Full Code of verl-project/verl for AI

gitextract_5e2u4bw9/

├── .gemini/
│   └── config.yaml
├── .git-blame-ignore-revs
├── .github/
│   ├── CODEOWNERS
│   ├── ISSUE_TEMPLATE/
│   │   ├── bug-report.yml
│   │   ├── config.yml
│   │   └── feature-request.yml
│   ├── PULL_REQUEST_TEMPLATE.md
│   ├── dependabot.yml
│   └── workflows/
│       ├── README.md
│       ├── check-pr-title.yml
│       ├── cpu_unit_tests.yml
│       ├── doc.yml
│       ├── docker-build-ascend-a2.yml
│       ├── docker-build-ascend-a3.yml
│       ├── e2e_ascend.yml
│       ├── e2e_fully_async_policy.yml
│       ├── e2e_fully_async_policy_ascend.yml
│       ├── e2e_one_step_off_policy.yml
│       ├── e2e_one_step_off_policy_ascend.yml
│       ├── e2e_ppo_grpo_trainer_trtllm.yml
│       ├── e2e_ppo_trainer.yml
│       ├── e2e_ppo_trainer_megatron_sglang.yml
│       ├── e2e_ppo_trainer_megatron_sglang_2.yml
│       ├── e2e_ppo_trainer_megatron_vllm.yml
│       ├── e2e_ppo_trainer_megatron_vllm_2.yml
│       ├── e2e_ppo_trainer_megatron_vllm_2_ascend.yml
│       ├── e2e_ppo_trainer_veomni_vllm.yml
│       ├── e2e_sft_llm.yml
│       ├── e2e_sft_llm_ascend.yml
│       ├── e2e_sft_vlm.yml
│       ├── gpu_unit_tests.yml
│       ├── model.yml
│       ├── model_ascend.yml
│       ├── nightly_ascend.yml
│       ├── npu_unit_tests.yml
│       ├── pre-commit.yml
│       ├── precommit-autofix.yml
│       ├── reward_model_sglang.yml
│       ├── reward_model_vllm.yml
│       ├── reward_model_vllm_ascend.yml
│       ├── sanity.yml
│       ├── scorecard.yml
│       ├── secrets_scan.yml
│       ├── sgl.yml
│       ├── type-coverage-check.yml
│       └── vllm.yml
├── .gitignore
├── .gitmodules
├── .pre-commit-config.yaml
├── .readthedocs.yaml
├── CONTRIBUTING.md
├── LICENSE
├── Notice.txt
├── README.md
├── docker/
│   ├── Dockerfile.isaaclab230
│   ├── Dockerfile.stable.sglang
│   ├── Dockerfile.stable.trtllm
│   ├── Dockerfile.stable.vllm
│   ├── README.md
│   ├── ascend/
│   │   ├── Dockerfile.ascend.sglang_8.3.rc1_a2
│   │   ├── Dockerfile.ascend.sglang_8.3.rc1_a3
│   │   ├── Dockerfile.ascend_8.2.rc1_a2
│   │   ├── Dockerfile.ascend_8.2.rc1_a3
│   │   ├── Dockerfile.ascend_8.3.rc1_a2
│   │   ├── Dockerfile.ascend_8.3.rc1_a3
│   │   ├── Dockerfile.ascend_8.5.0_a2
│   │   └── Dockerfile.ascend_8.5.0_a3
│   ├── aws/
│   │   ├── Dockerfile.extention.awsefa
│   │   └── Dockerfile.ngc.vllm0.8.sagemaker
│   ├── rocm/
│   │   ├── Apptainerfile.rocm
│   │   ├── Dockerfile.rocm
│   │   ├── Dockerfile.rocm7
│   │   ├── Dockerfile.rocm_verl-0.3.0.post1
│   │   └── Dockerfile.rocm_verl-0.4.1
│   ├── verl0.4-cu124-torch2.6-fa2.7.4/
│   │   ├── Dockerfile.app.sglang.vllm.mcore0.12
│   │   ├── Dockerfile.app.sglang.vllm.mcore0.12.deepep
│   │   ├── Dockerfile.app.sglang.vllm.mcore0.13.preview
│   │   ├── Dockerfile.app.vllm.mcore0.12
│   │   ├── Dockerfile.app.vllm.mcore0.12.deepep
│   │   ├── Dockerfile.app.vllm.mcore0.13.preview
│   │   ├── Dockerfile.base
│   │   └── README.md
│   ├── verl0.5-cu126-torch2.7-fa2.7.4/
│   │   ├── Dockerfile.app.sglang0.4.10.post2.mcore0.13
│   │   ├── Dockerfile.app.sglang0.4.9.post6.mcore0.13
│   │   ├── Dockerfile.app.vllm.mcore0.13
│   │   ├── Dockerfile.app.vllm.mcore0.15
│   │   ├── Dockerfile.base.torch2.7.1
│   │   └── README.md
│   ├── verl0.5-cu126-torch2.7.1-fa2.8.0/
│   │   ├── Dockerfile.app.sglang.mcore0.12
│   │   ├── Dockerfile.app.sglang.mcore0.13.preview
│   │   ├── Dockerfile.base
│   │   └── README.md
│   ├── verl0.5-preview-cu128-torch2.7.1-fa2.8.0/
│   │   ├── Dockerfile.app.sglang.megatron
│   │   ├── Dockerfile.base
│   │   └── README.md
│   ├── verl0.6-cu128-torch2.8.0-fa2.7.4/
│   │   ├── Dockerfile.app.sglang
│   │   ├── Dockerfile.base
│   │   └── Dockerfile.vllm011.mcore_gpt-oss
│   └── verl0.6.1-experimental/
│       ├── Dockerfile.sglang056exp
│       └── Dockerfile.vllm012exp
├── docs/
│   ├── Makefile
│   ├── README.md
│   ├── README_vllm0.7.md
│   ├── README_vllm0.8.md
│   ├── _static/
│   │   ├── custom.css
│   │   └── js/
│   │       ├── resizable-sidebar.js
│   │       └── runllm-widget.js
│   ├── advance/
│   │   ├── agent_loop.rst
│   │   ├── async-on-policy-distill.md
│   │   ├── attention_implementation.rst
│   │   ├── checkpoint.rst
│   │   ├── dpo_extension.rst
│   │   ├── fp8.md
│   │   ├── fsdp_extension.rst
│   │   ├── fully_async.md
│   │   ├── grafana_prometheus.md
│   │   ├── megatron_extension.rst
│   │   ├── mtp.md
│   │   ├── one_step_off.md
│   │   ├── placement.rst
│   │   ├── ppo_lora.rst
│   │   ├── reward_loop.rst
│   │   ├── rollout_skip.rst
│   │   ├── rollout_trace.rst
│   │   └── rope.rst
│   ├── algo/
│   │   ├── baseline.md
│   │   ├── collabllm.md
│   │   ├── dapo.md
│   │   ├── dppo.md
│   │   ├── entropy.md
│   │   ├── gpg.md
│   │   ├── grpo.md
│   │   ├── opo.md
│   │   ├── otb.md
│   │   ├── ppo.md
│   │   ├── rollout_corr.md
│   │   ├── rollout_corr_math.md
│   │   ├── spin.md
│   │   └── sppo.md
│   ├── amd_tutorial/
│   │   ├── amd_build_dockerfile_page.rst
│   │   └── amd_vllm_page.rst
│   ├── api/
│   │   ├── data.rst
│   │   ├── single_controller.rst
│   │   ├── trainer.rst
│   │   └── utils.rst
│   ├── ascend_tutorial/
│   │   ├── contribution_guide/
│   │   │   └── ascend_ci_guide_zh.rst
│   │   ├── examples/
│   │   │   ├── ascend_performance_analysis_guide.md
│   │   │   ├── ascend_retool_best_pratice.rst
│   │   │   ├── ascend_sglang_best_practices.rst
│   │   │   ├── dapo_multi_model_optimization_practice.md
│   │   │   ├── gspo_optimization_practice.md
│   │   │   └── run_qwen3_32B_megatron_1k_256k_npu.md
│   │   ├── faq/
│   │   │   └── faq.rst
│   │   ├── features/
│   │   │   ├── ascend_backend_features.md
│   │   │   └── ascend_consistency.rst
│   │   ├── profiling/
│   │   │   ├── ascend_profiling_en.rst
│   │   │   └── ascend_profiling_zh.rst
│   │   └── quick_start/
│   │       ├── ascend_quick_start.rst
│   │       ├── ascend_sglang_quick_start.rst
│   │       └── dockerfile_build_guidance.rst
│   ├── blog/
│   │   └── v0.7.md
│   ├── conf.py
│   ├── data/
│   │   └── transfer_queue.md
│   ├── examples/
│   │   ├── config.rst
│   │   ├── gsm8k_example.rst
│   │   ├── multi_modal_example.rst
│   │   ├── ppo_code_architecture.rst
│   │   ├── sandbox_fusion_example.rst
│   │   └── skypilot_examples.rst
│   ├── faq/
│   │   └── faq.rst
│   ├── hybrid_flow.rst
│   ├── index.rst
│   ├── perf/
│   │   ├── best_practices.rst
│   │   ├── device_tuning.rst
│   │   ├── dpsk.md
│   │   ├── nsight_profiling.md
│   │   ├── perf_tuning.rst
│   │   ├── perf_tuning_on_ascend.rst
│   │   ├── torch_profiling.md
│   │   └── verl_profiler_system.md
│   ├── preparation/
│   │   ├── prepare_data.rst
│   │   └── reward_function.rst
│   ├── requirements-docs.txt
│   ├── sglang_multiturn/
│   │   ├── interaction_system.rst
│   │   ├── multiturn.rst
│   │   ├── sandbox_fusion.rst
│   │   └── search_tool_example.rst
│   ├── single_controller.rst
│   ├── start/
│   │   ├── agentic_rl.rst
│   │   ├── install.rst
│   │   ├── more_resources.rst
│   │   ├── multinode.rst
│   │   ├── quickstart.rst
│   │   └── ray_debug_tutorial.rst
│   └── workers/
│       ├── automodel_workers.rst
│       ├── fsdp_workers.rst
│       ├── megatron_workers.rst
│       ├── model_engine.rst
│       ├── ray_trainer.rst
│       ├── sglang_worker.rst
│       └── trtllm_worker.rst
├── examples/
│   ├── cispo_trainer/
│   │   └── run_cispo_qwen2_5_0_5b_gsm8k.sh
│   ├── data_preprocess/
│   │   ├── aime2024_multiturn_w_tool.py
│   │   ├── dapo_multiturn_w_tool.py
│   │   ├── full_hh_rlhf.py
│   │   ├── geo3k.py
│   │   ├── geo3k_multiturn_w_tool.py
│   │   ├── gsm8k.py
│   │   ├── gsm8k_multiturn_sft.py
│   │   ├── gsm8k_multiturn_w_interaction.py
│   │   ├── gsm8k_multiturn_w_tool.py
│   │   ├── gsm8k_tool_agent_loop.py
│   │   ├── hellaswag.py
│   │   ├── math_dataset.py
│   │   ├── multiturn.py
│   │   ├── pokemon.py
│   │   └── preprocess_search_r1_dataset.py
│   ├── dppo_trainer/
│   │   ├── dppo.md
│   │   └── run_qwen30b_dppo.sh
│   ├── fapo_trainer/
│   │   ├── README.md
│   │   ├── prepare_data.py
│   │   ├── reward_fn.py
│   │   ├── run_qwen_7b_rm_colocate.sh
│   │   └── run_qwen_7b_rm_standalone.sh
│   ├── gdpo_trainer/
│   │   └── run_qwen1_5b_gdpo.sh
│   ├── generation/
│   │   ├── run_deepseek7b_mutli_node.sh
│   │   └── run_deepseek_v2_lite_math.sh
│   ├── gmpo_trainer/
│   │   ├── README.md
│   │   ├── run_qwen2_5-7b_math.sh
│   │   ├── test_dapo_7b_math.sh
│   │   └── test_dapo_qwen3_30b_math.sh
│   ├── gpg_trainer/
│   │   ├── gpg.md
│   │   ├── run_qwen2-7b_math.sh
│   │   └── run_qwen2-7b_math_megatron.sh
│   ├── grpo_trainer/
│   │   ├── README.md
│   │   ├── run_deepseek671b_math_megatron_80gb.sh
│   │   ├── run_deepseek671b_math_megatron_96gb.sh
│   │   ├── run_deepseek7b_llm.sh
│   │   ├── run_deepseek7b_llm_math.sh
│   │   ├── run_deepseek7b_llm_math_megatron.sh
│   │   ├── run_deepseek7b_llm_seq_balance.sh
│   │   ├── run_glm41v_9b.sh
│   │   ├── run_gptoss_20b.sh
│   │   ├── run_minicpmo2_6.sh
│   │   ├── run_mistral13b_skyworkrm_hhrlhf.sh
│   │   ├── run_moonlight16b_math_megatron.sh
│   │   ├── run_nemotron_nano_v3_megatron.sh
│   │   ├── run_qwen2-32b_sglang_fsdp_npu.sh
│   │   ├── run_qwen2-7b.sh
│   │   ├── run_qwen2-7b_math.sh
│   │   ├── run_qwen2-7b_math_megatron.sh
│   │   ├── run_qwen2-7b_math_megatron_lora.sh
│   │   ├── run_qwen2-7b_math_megatron_trtllm.sh
│   │   ├── run_qwen2-7b_math_trtllm.sh
│   │   ├── run_qwen2-7b_seq_balance.sh
│   │   ├── run_qwen2-7b_seq_balance_math_megatron.sh
│   │   ├── run_qwen2-7b_sgl_megatron.sh
│   │   ├── run_qwen2_5-32b_grpo_megatron_vllm_npu.sh
│   │   ├── run_qwen2_5-3b_gsm8k_grpo_lora.sh
│   │   ├── run_qwen2_5-3b_gsm8k_grpo_lora_from_adapter.sh
│   │   ├── run_qwen2_5-7b_math_megatron_diff_tp.sh
│   │   ├── run_qwen2_5_32b_grpo_npu.sh
│   │   ├── run_qwen2_5_7b_grpo_discrete_prof_npu.sh
│   │   ├── run_qwen2_5_7b_grpo_e2e_prof_npu.sh
│   │   ├── run_qwen2_5_7b_grpo_npu.sh
│   │   ├── run_qwen2_5_vl-7b-megatron.sh
│   │   ├── run_qwen2_5_vl-7b-sglang.sh
│   │   ├── run_qwen2_5_vl-7b-trtllm.sh
│   │   ├── run_qwen2_5_vl-7b.sh
│   │   ├── run_qwen2_5_vl-7b_freeze_vision.sh
│   │   ├── run_qwen2_5_vl-7b_lora.sh
│   │   ├── run_qwen2_5_vl-7b_seq_balance.sh
│   │   ├── run_qwen2_5_vl_32b_npu.sh
│   │   ├── run_qwen2_5_vl_3b_npu.sh
│   │   ├── run_qwen2_5_vl_3b_trtllm.sh
│   │   ├── run_qwen2_5_vl_7b_npu.sh
│   │   ├── run_qwen3-235b_megatron_96gb.sh
│   │   ├── run_qwen3-30b_dapo_megatron_fp8_trtllm.sh
│   │   ├── run_qwen3-32b_npu.sh
│   │   ├── run_qwen3-4b_gsm8k_grpo_lora_merge.sh
│   │   ├── run_qwen3-8b.sh
│   │   ├── run_qwen3-8b_npu.sh
│   │   ├── run_qwen3_235b_megatron_npu.sh
│   │   ├── run_qwen3_4b_grpo_vllm_1k_npu.sh
│   │   ├── run_qwen3_5-35b-megatron.sh
│   │   ├── run_qwen3_8b_grpo_sglang_1k_spmd_npu.sh
│   │   ├── run_qwen3_8b_grpo_sglang_32k_spmd_npu.sh
│   │   ├── run_qwen3_vl-235b-megatron.sh
│   │   ├── run_qwen3_vl-30b-megatron.sh
│   │   ├── run_qwen3_vl-8b-megatron.sh
│   │   ├── run_qwen3_vl-8b_npu.sh
│   │   ├── run_qwen3_vl_30b_vllm_fsdp_npu.sh
│   │   ├── run_qwen3moe-30b_grpo_megatron_vllm_npu.sh
│   │   ├── run_qwen3moe-30b_megatron_96gb.sh
│   │   ├── run_qwen3moe-30b_megatron_lora.sh
│   │   ├── run_qwen3moe-30b_megatron_lora_fp16.sh
│   │   ├── run_qwen3moe-30b_sglang_megatron_npu.sh
│   │   ├── run_qwen3next_80b_fsdp_npu.sh
│   │   └── run_seed_oss_36b.sh
│   ├── gspo_trainer/
│   │   ├── run_qwen30b_gspo.sh
│   │   ├── run_qwen3_32b_gspo_npu.sh
│   │   ├── test_gspo_3b_math.sh
│   │   ├── test_gspo_3b_math_slurm.sh
│   │   └── test_gspo_qwen30b_a3b_ep.sh
│   ├── mtp_trainer/
│   │   ├── runtime_env.yaml
│   │   ├── test_dapo_mimo_7b_with_mtp_math_megatron.sh
│   │   └── test_dapo_mimo_7b_with_mtp_math_megatron_4_4.sh
│   ├── otb_trainer/
│   │   └── run_qwen2_5-7b.sh
│   ├── ppo_trainer/
│   │   ├── README.md
│   │   ├── run_deepseek7b_llm.sh
│   │   ├── run_deepseek7b_llm_modelscope.sh
│   │   ├── run_deepseek7b_llm_pfppo.sh
│   │   ├── run_deepseek7b_llm_sandbox_fusion.sh
│   │   ├── run_deepseek7b_llm_sp2.sh
│   │   ├── run_deepseek_full_hh_rlhf.sh
│   │   ├── run_deepseek_math_gsm8k_megatron.sh
│   │   ├── run_deepseek_math_gsm8k_megatron_nsys.sh
│   │   ├── run_gemma.sh
│   │   ├── run_moonlight16b_a3b_gsm8k_megatron.sh
│   │   ├── run_qwen1.5_moe_a2.7b-gsm8k_megatron.sh
│   │   ├── run_qwen2-7b_math_gsm8k_megatron.sh
│   │   ├── run_qwen2-7b_rm.sh
│   │   ├── run_qwen2-7b_rm_reward_loop_colocate.sh
│   │   ├── run_qwen2-7b_rm_seq_balance.sh
│   │   ├── run_qwen2-7b_rm_seq_balance_fused_kernels.sh
│   │   ├── run_qwen2-7b_rm_seq_balance_nsys.sh
│   │   ├── run_qwen2-7b_seq_balance.sh
│   │   ├── run_qwen2-7b_sglang_seq_balance.sh
│   │   ├── run_qwen2.5-32b.sh
│   │   ├── run_qwen2.5-3b_rm_reward_loop_colocate.sh
│   │   └── run_qwen3-8b_npu.sh
│   ├── prefix_grouper/
│   │   ├── README.md
│   │   └── run_qwen3_prefix_grouper.sh
│   ├── ray/
│   │   └── tutorial.ipynb
│   ├── reinforce_plus_plus_trainer/
│   │   ├── run_qwen2-7b_math_rf.sh
│   │   └── run_qwen2-7b_math_rf_baseline.sh
│   ├── remax_trainer/
│   │   ├── run_qwen2.5-3b_seq_balance.sh
│   │   └── run_qwen2.5-7b_seq_balance.sh
│   ├── rloo_trainer/
│   │   └── run_qwen2-7b.sh
│   ├── rollout_correction/
│   │   ├── run_with_rollout_corr.sh
│   │   └── run_with_rollout_corr_multi_rs.sh
│   ├── router_replay/
│   │   ├── README.md
│   │   ├── run_qwen30_a3b_megatron_sglang.sh
│   │   └── run_qwen30_a3b_megatron_vllm.sh
│   ├── sapo_trainer/
│   │   ├── run_qwen30b_sapo.sh
│   │   └── run_qwen3_8b_sapo_npu.sh
│   ├── sft/
│   │   ├── gsm8k/
│   │   │   ├── run_deepseek_6b7.sh
│   │   │   ├── run_gemma_2b.sh
│   │   │   ├── run_gemma_7b.sh
│   │   │   ├── run_mimo_megatron_mtp.sh
│   │   │   ├── run_nemotron_nano_v3.sh
│   │   │   ├── run_qwen3_30b_automodel.sh
│   │   │   ├── run_qwen3_5_megatron.sh
│   │   │   ├── run_qwen3_8b_sft_peft_sp2_npu.sh
│   │   │   ├── run_qwen_05_automodel.sh
│   │   │   ├── run_qwen_05_peft.sh
│   │   │   ├── run_qwen_05_sp2.sh
│   │   │   ├── run_qwen_05_sp2_liger.sh
│   │   │   └── run_seed_oss_36b_sft.sh
│   │   ├── multiturn/
│   │   │   └── run_qwen_05_sp2.sh
│   │   └── vlm/
│   │       └── run_qwen3_vl_2b.sh
│   ├── sglang_multiturn/
│   │   ├── README.md
│   │   ├── config/
│   │   │   ├── geo3k_multiturn_grpo.yaml
│   │   │   ├── geo3k_multiturn_megatron_grpo.yaml
│   │   │   ├── gsm8k_multiturn_grpo.yaml
│   │   │   ├── gsm8k_multiturn_grpo_server.yaml
│   │   │   ├── gsm8k_multiturn_grpo_w_interaction.yaml
│   │   │   ├── gsm8k_multiturn_megatron_grpo.yaml
│   │   │   ├── interaction_config/
│   │   │   │   └── gsm8k_interaction_config.yaml
│   │   │   ├── retool_multiturn_grpo.yaml
│   │   │   ├── search_multiturn_grpo.yaml
│   │   │   ├── search_multiturn_grpo_one_step_off.yaml
│   │   │   └── tool_config/
│   │   │       ├── geo3k_tool_config.yaml
│   │   │       ├── gsm8k_tool_config.yaml
│   │   │       ├── mcp_server.json
│   │   │       ├── mcp_tool_config.yaml
│   │   │       ├── sandbox_fusion_tool_config.yaml
│   │   │       └── search_tool_config.yaml
│   │   ├── geo3k/
│   │   │   ├── run_qwen2.5-3b_geo3k_multiturn.sh
│   │   │   ├── run_qwen2.5-3b_geo3k_multiturn_4xgpu.sh
│   │   │   └── run_qwen2.5-3b_megatron_geo3k_multiturn.sh
│   │   ├── gsm8k_toolcall_shaping/
│   │   │   ├── gsm8k_toolcall_shaping.py
│   │   │   └── run_gsm8k_grpo_toolcall_shaping.sh
│   │   ├── run_qwen0.5b_gsm8k_multiturn_curriculum.sh
│   │   ├── run_qwen2.5-0.5b_gsm8k_multiturn_w_interaction.sh
│   │   ├── run_qwen2.5-3b_gsm8k_multiturn.sh
│   │   ├── run_qwen2.5-3b_gsm8k_multiturn_4xgpu.sh
│   │   ├── run_qwen2.5-3b_gsm8k_multiturn_4xgpu_server.sh
│   │   ├── run_qwen2.5-3b_gsm8k_multiturn_server.sh
│   │   ├── run_qwen2.5-3b_gsm8k_multiturn_vllm_fsdp.sh
│   │   ├── run_qwen2.5-3b_gsm8k_tool_agent_mlflow.sh
│   │   ├── run_qwen2.5-3b_megatron_gsm8k_multiturn.sh
│   │   ├── run_qwen3-4b_gsm8k_multiturn.sh
│   │   ├── run_qwen3_4b_dapo_multiturn.sh
│   │   └── search_r1_like/
│   │       ├── local_dense_retriever/
│   │       │   ├── download.py
│   │       │   └── retrieval_server.py
│   │       └── run_qwen2.5-3b_instruct_search_multiturn.sh
│   ├── skypilot/
│   │   ├── README.md
│   │   ├── verl-grpo.yaml
│   │   ├── verl-multiturn-tools.yaml
│   │   └── verl-ppo.yaml
│   ├── slurm/
│   │   └── ray_on_slurm.slurm
│   ├── split_placement/
│   │   ├── README.md
│   │   ├── config/
│   │   │   └── ppo_trainer_split.yaml
│   │   ├── main_ppo_split.py
│   │   ├── run_deepseek7b_llm.sh
│   │   └── split_monkey_patch.py
│   ├── tuning/
│   │   ├── 0.5b/
│   │   │   └── qwen2-0.5b_grpo-lora_1_h100_fsdp_vllm.sh
│   │   ├── 1.5b/
│   │   │   └── qwen2-1.5b_grpo-lora_1_h100_fsdp_vllm.sh
│   │   ├── 14b/
│   │   │   ├── qwen2-14b_grpo-lora_2_h100_fsdp_vllm.sh
│   │   │   └── qwen2_14b_grpo_4_h800_fsdp_vllm.sh
│   │   ├── 32b/
│   │   │   ├── qwen2-32b_grpo-lora_4_h100_fsdp_vllm.sh
│   │   │   └── qwen2_32B_grpo_8_h20_megatron_vllm.sh
│   │   ├── 3b/
│   │   │   └── qwen2-3b_grpo-lora_1_h100_fsdp_vllm.sh
│   │   ├── 70b/
│   │   │   ├── qwen2-70b_grpo_32_h20_fsdp_vllm.sh
│   │   │   ├── qwen2-70b_grpo_32_h800_fsdp_vllm.sh
│   │   │   └── qwen2-72b_grpo-lora_8_h100_fsdp_vllm.sh
│   │   └── 7b/
│   │       ├── qwen2-7b_grpo-lora_1_h100_fsdp_vllm.sh
│   │       └── qwen2-7b_grpo_2_h800_fsdp_vllm.sh
│   └── tutorial/
│       └── agent_loop_get_started/
│           ├── agent_loop_tutorial.ipynb
│           └── sandbox.py
├── pyproject.toml
├── requirements-cuda.txt
├── requirements-npu.txt
├── requirements-test.txt
├── requirements.txt
├── requirements_sglang.txt
├── scripts/
│   ├── __init__.py
│   ├── converter_hf_to_mcore.py
│   ├── diagnose.py
│   ├── generate_trainer_config.sh
│   ├── init_random_model.py
│   ├── install_sglang_mcore_npu.sh
│   ├── install_vllm_sglang_mcore.sh
│   ├── legacy_model_merger.py
│   ├── megatron_merge_lora.py
│   ├── print_cfg.py
│   ├── rollout_viewer.py
│   └── veomni/
│       ├── moe_merge.py
│       └── moe_split.py
├── setup.py
├── tests/
│   ├── README.md
│   ├── __init__.py
│   ├── checkpoint_engine/
│   │   ├── __init__.py
│   │   ├── test_correctness_on_gpu.py
│   │   ├── test_correctness_on_npu.py
│   │   ├── test_special_server_adapter.py
│   │   └── test_utils.py
│   ├── experimental/
│   │   ├── agent_loop/
│   │   │   ├── agent_utils.py
│   │   │   ├── qwen_vl_tool_chat_template.jinja2
│   │   │   ├── test_agent_loop_extra_fields_schema_on_cpu.py
│   │   │   ├── test_basic_agent_loop.py
│   │   │   ├── test_gpt_oss_tool_parser.py
│   │   │   ├── test_multi_modal.py
│   │   │   └── test_standalone_rollout.py
│   │   ├── reward_loop/
│   │   │   ├── reward_fn.py
│   │   │   ├── test_agent_reward_loop_colocate.py
│   │   │   ├── test_agent_reward_loop_standalone.py
│   │   │   ├── test_async_token_bucket_on_cpu.py
│   │   │   ├── test_math_verify.py
│   │   │   ├── test_rate_limited_reward_manager_on_cpu.py
│   │   │   ├── test_reward_model_disrm.py
│   │   │   └── test_reward_model_genrm.py
│   │   └── vla/
│   │       └── test_sim_envs.py
│   ├── interactions/
│   │   ├── __init__.py
│   │   ├── test_gsm8k_interaction.py
│   │   └── test_interaction_registry.py
│   ├── kill_github_tests.sh
│   ├── models/
│   │   ├── test_engine.py
│   │   ├── test_tiled_mlp_accuracy.py
│   │   ├── test_transformer.py
│   │   └── test_transformers_ulysses.py
│   ├── single_controller/
│   │   ├── __init__.py
│   │   ├── base/
│   │   │   └── test_decorator.py
│   │   ├── check_worker_alive/
│   │   │   └── main.py
│   │   ├── detached_worker/
│   │   │   ├── README.md
│   │   │   ├── client.py
│   │   │   ├── run.sh
│   │   │   └── server.py
│   │   ├── test_auto_padding_on_cpu.py
│   │   ├── test_colocated_workers.py
│   │   ├── test_colocated_workers_fused.py
│   │   ├── test_data_transfer.py
│   │   ├── test_decorator_on_cpu.py
│   │   ├── test_device_mesh_register.py
│   │   ├── test_driverfunc_to_worker.py
│   │   ├── test_fused_workers_on_cpu.py
│   │   ├── test_get_set_dispatch_collect_cpu.py
│   │   ├── test_high_level_scheduling_api.py
│   │   ├── test_nested_worker.py
│   │   ├── test_ray_collectives.py
│   │   ├── test_ray_local_envs_on_cpu.py
│   │   ├── test_ray_utils_on_cpu.py
│   │   ├── test_rvdz.py
│   │   ├── test_split_resource_pool.py
│   │   ├── test_worker_group_basics.py
│   │   └── test_worker_group_torch.py
│   ├── special_distributed/
│   │   ├── README.md
│   │   ├── run_all.sh
│   │   ├── test_fsdp_ckpt.py
│   │   ├── test_mcore_config_converter.py
│   │   ├── test_tensor_dict.py
│   │   └── test_torch_functional.py
│   ├── special_e2e/
│   │   ├── README.md
│   │   ├── __init__.py
│   │   ├── check_custom_rwd_fn.py
│   │   ├── check_results.py
│   │   ├── envs/
│   │   │   ├── __init__.py
│   │   │   └── digit_completion/
│   │   │       ├── __init__.py
│   │   │       ├── task.py
│   │   │       └── tokenizer.py
│   │   ├── generation/
│   │   │   ├── run_gen_qwen05.sh
│   │   │   └── run_gen_qwen05_server.sh
│   │   ├── ppo_trainer/
│   │   │   ├── expert_parallel/
│   │   │   │   ├── qwen2moe_minimal.json
│   │   │   │   └── qwen3moe_minimal.json
│   │   │   ├── run_function_reward.sh
│   │   │   ├── run_model_reward.sh
│   │   │   ├── run_single_gpu.sh
│   │   │   └── run_single_gpu_with_engine.sh
│   │   ├── run_dapo.sh
│   │   ├── run_fully_async_policy.sh
│   │   ├── run_geo3k_fsdp_sgl_multiturn_w_tool.sh
│   │   ├── run_grpo_lora_with_merge.sh
│   │   ├── run_gsm8k_fsdp_sgl_multiturn_sf_tool.sh
│   │   ├── run_gsm8k_fsdp_sgl_multiturn_w_tool.sh
│   │   ├── run_one_step_off_policy.sh
│   │   ├── run_ppo_trainer_megatron.sh
│   │   ├── run_ppo_trainer_torchtitan.sh
│   │   ├── run_ppo_trainer_veomni.sh
│   │   ├── run_test.sh
│   │   └── sft/
│   │       ├── compare_sft_engine_results.py
│   │       ├── run_sft.sh
│   │       ├── run_sft_engine.sh
│   │       └── test_sft_engine_all.sh
│   ├── special_npu/
│   │   ├── nightly_ci_ascend/
│   │   │   ├── run_grpo_qwen25-7b-instruct_fsdp_npu.sh
│   │   │   ├── run_grpo_qwen25-vl-3b-instruct_fsdp_npu.sh
│   │   │   └── run_ppo_qwen3-8b_fsdp_npu.sh
│   │   ├── run_qwen2_5_05b_grpo.sh
│   │   ├── run_qwen2_5_05b_grpo_mindspeed.sh
│   │   ├── run_qwen2_5_05b_sft_peft_sp2.sh
│   │   ├── run_qwen2_5_vl_3b_npu.sh
│   │   ├── run_qwen3_06b_ppo.sh
│   │   └── run_qwen3_30b_grpo_mindspeed.sh
│   ├── special_sanity/
│   │   ├── check_api_docs.py
│   │   ├── check_dataproto_usage.py
│   │   ├── check_device_api_usage.py
│   │   ├── check_docs_time_info.py
│   │   ├── check_docstrings.py
│   │   ├── check_license.py
│   │   ├── check_pr_description.py
│   │   ├── check_pr_title.py
│   │   ├── test_config_docs.py
│   │   ├── test_import.py
│   │   ├── type_coverage_check.py
│   │   ├── validate_imported_docs.py
│   │   └── validate_structure.py
│   ├── special_standalone/
│   │   ├── README.md
│   │   └── test_memory_buffers.py
│   ├── test_base_config_on_cpu.py
│   ├── test_protocol_on_cpu.py
│   ├── test_protocol_v2_on_cpu.py
│   ├── trainer/
│   │   ├── __init__.py
│   │   ├── config/
│   │   │   ├── __init__.py
│   │   │   ├── legacy_ppo_megatron_trainer.yaml
│   │   │   ├── legacy_ppo_trainer.yaml
│   │   │   ├── test_algo_config_on_cpu.py
│   │   │   └── test_legacy_config_on_cpu.py
│   │   └── ppo/
│   │       ├── __init__.py
│   │       ├── test_core_algos_on_cpu.py
│   │       ├── test_metric_utils_on_cpu.py
│   │       ├── test_rollout_corr.py
│   │       └── test_rollout_corr_integration.py
│   ├── utils/
│   │   ├── _test_module.py
│   │   ├── ckpt/
│   │   │   ├── test_checkpoint_cleanup_on_cpu.py
│   │   │   └── test_esi_save_ckpt_on_cpu.py
│   │   ├── dataset/
│   │   │   ├── test_create_rl_sampler_on_cpu.py
│   │   │   ├── test_multiturn_sft_dataset_on_cpu.py
│   │   │   ├── test_rl_collate_fn_on_cpu.py
│   │   │   └── test_rl_dataset_on_cpu.py
│   │   ├── debug/
│   │   │   └── test_metrics.py
│   │   ├── megatron/
│   │   │   └── test_pipeline_parallel.py
│   │   ├── reward_score/
│   │   │   ├── reward_score/
│   │   │   │   └── test_sandbox_fusion_on_cpu.py
│   │   │   └── test_sandbox_on_cpu.py
│   │   ├── test_activation_offload.py
│   │   ├── test_bucketed_weight_transfer.py
│   │   ├── test_check_ipc_version_support_on_npu.py
│   │   ├── test_check_profiler_output.py
│   │   ├── test_config_on_cpu.py
│   │   ├── test_flops_counter.py
│   │   ├── test_fs_on_cpu.py
│   │   ├── test_fsdp2_peft_wrapping.py
│   │   ├── test_fsdp_lora_merge.py
│   │   ├── test_groupwise.py
│   │   ├── test_import_utils_on_cpu.py
│   │   ├── test_linear_cross_entropy.py
│   │   ├── test_mlflow_key_sanitization.py
│   │   ├── test_model_on_cpu.py
│   │   ├── test_normalize_peft_param_name.py
│   │   ├── test_normalize_peft_param_name_on_cpu.py
│   │   ├── test_nvtx_profile.py
│   │   ├── test_padding_on_cpu.py
│   │   ├── test_prepare_micro_batches_with_group_size.py
│   │   ├── test_rollout_skip_on_cpu.py
│   │   ├── test_rollout_trace_on_cpu.py
│   │   ├── test_seqlen_balancing.py
│   │   ├── test_server_profiler.py
│   │   ├── test_shared_memory.py
│   │   ├── test_special_linear_cross_entropy_tp.py
│   │   ├── test_special_mstx_profile.py
│   │   ├── test_temp_env_on_cpu.py
│   │   ├── test_timeout_decorator_cpu.py
│   │   ├── test_tokenizer_normalize_on_cpu.py
│   │   ├── test_torch_functional.py
│   │   └── test_torch_profile.py
│   └── workers/
│       ├── actor/
│       │   └── test_special_dp_actor.py
│       ├── config/
│       │   ├── test_actor_config_on_cpu.py
│       │   ├── test_critic_config_on_cpu.py
│       │   ├── test_engine_config_on_cpu.py
│       │   ├── test_model_config_on_cpu.py
│       │   └── test_optim_config_on_cpu.py
│       ├── critic/
│       │   └── test_special_dp_critic.py
│       ├── reward_manager/
│       │   └── test_registry_on_cpu.py
│       ├── rollout/
│       │   ├── perf/
│       │   │   └── vllm_async_rollout.py
│       │   ├── resource/
│       │   │   └── tool_configs/
│       │   │       ├── mcp_server.json
│       │   │       ├── mcp_tool_config
│       │   │       ├── sandbox_fusion_tool_config
│       │   │       └── search_tool_config
│       │   ├── rollout_sglang/
│       │   │   └── test_http_server_engine.py
│       │   ├── rollout_trtllm/
│       │   │   ├── __init__.py
│       │   │   ├── test_adapter.py
│       │   │   ├── test_async_server.py
│       │   │   └── test_trtllm_rollout_utils.py
│       │   ├── rollout_vllm/
│       │   │   ├── run_fsdp_vllm.py
│       │   │   └── test_vllm_abort.py
│       │   ├── test_hf_rollout.py
│       │   ├── test_sglang_async_rollout_multimodal_delta.py
│       │   ├── test_sglang_rollout_sharding_manager.py
│       │   └── test_vllm_cli_args_on_cpu.py
│       ├── test_fsdp_attn_implementation.py
│       └── test_fsdp_workers.py
└── verl/
    ├── __init__.py
    ├── base_config.py
    ├── checkpoint_engine/
    │   ├── README.md
    │   ├── __init__.py
    │   ├── base.py
    │   ├── hccl_checkpoint_engine.py
    │   ├── kimi_checkpoint_engine.py
    │   ├── mooncake_checkpoint_engine.py
    │   ├── nccl_checkpoint_engine.py
    │   └── nixl_checkpoint_engine.py
    ├── experimental/
    │   ├── __init__.py
    │   ├── agent_loop/
    │   │   ├── __init__.py
    │   │   ├── agent_loop.py
    │   │   ├── prometheus_utils.py
    │   │   ├── single_turn_agent_loop.py
    │   │   ├── tool_agent_loop.py
    │   │   ├── tool_parser.py
    │   │   └── utils.py
    │   ├── dataset/
    │   │   ├── __init__.py
    │   │   └── sampler.py
    │   ├── dynamic_dataset/
    │   │   ├── __init__.py
    │   │   └── dynamicgen_dataset.py
    │   ├── fully_async_policy/
    │   │   ├── README.md
    │   │   ├── README_zh.md
    │   │   ├── agent_loop/
    │   │   │   ├── __init__.py
    │   │   │   └── agent_loop.py
    │   │   ├── config/
    │   │   │   ├── fully_async_ppo_megatron_trainer.yaml
    │   │   │   └── fully_async_ppo_trainer.yaml
    │   │   ├── detach_utils.py
    │   │   ├── fully_async_main.py
    │   │   ├── fully_async_rollouter.py
    │   │   ├── fully_async_trainer.py
    │   │   ├── message_queue.py
    │   │   ├── shell/
    │   │   │   ├── dapo_30b_a3b_base_math_fsdp.sh
    │   │   │   ├── dapo_7b_async_retool.sh
    │   │   │   ├── dapo_7b_math_fsdp2_16_16.sh
    │   │   │   ├── dapo_7b_math_fsdp2_32_32.sh
    │   │   │   ├── dapo_7b_math_fsdp2_4_12.sh
    │   │   │   ├── dapo_7b_math_fsdp2_4_4.sh
    │   │   │   ├── dapo_7b_math_fsdp2_64_64.sh
    │   │   │   ├── dapo_7b_math_fsdp2_64_64_mis.sh
    │   │   │   ├── dapo_7b_math_fsdp2_8_8.sh
    │   │   │   ├── geo3k_qwen25vl_7b_megatron_4_4.sh
    │   │   │   ├── grpo_30b_a3b_base_math_megatron_96_32.sh
    │   │   │   ├── grpo_30b_a3b_base_math_megatron_96_32_mis.sh
    │   │   │   └── runtime_env.yaml
    │   │   └── unittest/
    │   │       └── simple_streaming_demo.py
    │   ├── one_step_off_policy/
    │   │   ├── README.md
    │   │   ├── config/
    │   │   │   ├── one_step_off_ppo_megatron_trainer.yaml
    │   │   │   └── one_step_off_ppo_trainer.yaml
    │   │   ├── main_ppo.py
    │   │   ├── ray_trainer.py
    │   │   └── shell/
    │   │       ├── dapo_7b_math_fsdp2_4_12.sh
    │   │       ├── dapo_7b_math_fsdp2_64_64.sh
    │   │       ├── dapo_7b_math_fsdp2_64_64_ris.sh
    │   │       ├── dapo_7b_math_fsdp2_colocate.sh
    │   │       ├── dapo_7b_math_fsdp2_sglang_4_12.sh
    │   │       ├── dapo_7b_math_fsdp2_sglang_colocate.sh
    │   │       ├── dapo_7b_math_megatron_4_12.sh
    │   │       ├── dapo_7b_math_megatron_colocate.sh
    │   │       ├── grpo_0.6b_gsm8k_fsdp2_2_6.sh
    │   │       ├── grpo_0.6b_gsm8k_fsdp2_sglang_2_6.sh
    │   │       ├── grpo_3b_gsm8k_fsdp2_2_6.sh
    │   │       └── grpo_qwen3_8b_gsm8k_fsdp2_8_8_npu.sh
    │   ├── reward_loop/
    │   │   ├── __init__.py
    │   │   ├── reward_loop.py
    │   │   ├── reward_manager/
    │   │   │   ├── __init__.py
    │   │   │   ├── base.py
    │   │   │   ├── dapo.py
    │   │   │   ├── gdpo.py
    │   │   │   ├── limited.py
    │   │   │   ├── naive.py
    │   │   │   ├── registry.py
    │   │   │   └── remote.py
    │   │   ├── reward_model.py
    │   │   └── router/
    │   │       ├── inner_sglang_router.py
    │   │       └── naive_router.py
    │   ├── separation/
    │   │   ├── __init__.py
    │   │   ├── engine_workers.py
    │   │   ├── ray_trainer.py
    │   │   └── utils.py
    │   └── vla/
    │       ├── README.md
    │       ├── config/
    │       │   ├── rob_ppo_trainer.yaml
    │       │   └── rob_sac_trainer.yaml
    │       ├── dp_rob.py
    │       ├── env_loop.py
    │       ├── envs/
    │       │   ├── __init__.py
    │       │   ├── action_utils.py
    │       │   ├── isaac_env/
    │       │   │   ├── __init__.py
    │       │   │   └── isaac_env.py
    │       │   └── libero_env/
    │       │       ├── __init__.py
    │       │       ├── libero_env.py
    │       │       ├── utils.py
    │       │       └── venv.py
    │       ├── fsdp_workers.py
    │       ├── main_ppo.py
    │       ├── main_sac.py
    │       ├── models/
    │       │   ├── __init__.py
    │       │   ├── modules/
    │       │   │   └── mlp.py
    │       │   ├── openvla_oft/
    │       │   │   ├── __init__.py
    │       │   │   ├── configuration_prismatic.py
    │       │   │   ├── constants.py
    │       │   │   ├── modeling_prismatic.py
    │       │   │   ├── processing_prismatic.py
    │       │   │   └── train_utils.py
    │       │   ├── pi0_torch/
    │       │   │   ├── __init__.py
    │       │   │   ├── configuration_pi0_torch.py
    │       │   │   ├── model/
    │       │   │   │   ├── modeling_pi0.py
    │       │   │   │   └── paligemma_with_expert.py
    │       │   │   ├── modeling_pi0_torch.py
    │       │   │   ├── pi0_utils.py
    │       │   │   └── policy/
    │       │   │       ├── __init__.py
    │       │   │       ├── base.py
    │       │   │       └── libero_policy.py
    │       │   └── register_vla_models.py
    │       ├── naive_rollout_rob.py
    │       ├── prepare_libero_dataset.py
    │       ├── requirements_vla.txt
    │       ├── rob_ray_trainer.py
    │       ├── run_pi05_libero_sac.sh
    │       ├── run_pi05_libero_sac_disagg.sh
    │       ├── run_simpleVLA_isaac_disagg.sh
    │       ├── run_simpleVLA_libero_grpo.sh
    │       ├── sac/
    │       │   ├── base.py
    │       │   ├── naive_rollout_pi05.py
    │       │   ├── replay_pool.py
    │       │   ├── sac_actor.py
    │       │   └── sac_ray_trainer.py
    │       └── workers/
    │           └── env/
    │               ├── env_loop_wg_test.py
    │               ├── env_manager.py
    │               └── env_worker.py
    ├── interactions/
    │   ├── __init__.py
    │   ├── base.py
    │   ├── gsm8k_interaction.py
    │   ├── utils/
    │   │   ├── __init__.py
    │   │   └── interaction_registry.py
    │   └── weather_interaction.py
    ├── model_merger/
    │   ├── __init__.py
    │   ├── __main__.py
    │   ├── base_model_merger.py
    │   ├── fsdp_model_merger.py
    │   └── megatron_model_merger.py
    ├── models/
    │   ├── README.md
    │   ├── __init__.py
    │   ├── llama/
    │   │   ├── __init__.py
    │   │   └── megatron/
    │   │       ├── __init__.py
    │   │       ├── checkpoint_utils/
    │   │       │   ├── __init__.py
    │   │       │   ├── llama_loader.py
    │   │       │   ├── llama_loader_depracated.py
    │   │       │   └── llama_saver.py
    │   │       ├── layers/
    │   │       │   ├── __init__.py
    │   │       │   ├── parallel_attention.py
    │   │       │   ├── parallel_decoder.py
    │   │       │   ├── parallel_linear.py
    │   │       │   ├── parallel_mlp.py
    │   │       │   └── parallel_rmsnorm.py
    │   │       └── modeling_llama_megatron.py
    │   ├── mcore/
    │   │   ├── __init__.py
    │   │   ├── bridge.py
    │   │   ├── config_converter.py
    │   │   ├── loader.py
    │   │   ├── mbridge.py
    │   │   ├── model_forward.py
    │   │   ├── model_forward_1f1b_overlap.py
    │   │   ├── model_forward_fused.py
    │   │   ├── model_initializer.py
    │   │   ├── mtp_patch.py
    │   │   ├── patch.py
    │   │   ├── qwen2_5_vl/
    │   │   │   ├── __init__.py
    │   │   │   ├── attention.py
    │   │   │   ├── model.py
    │   │   │   ├── rope_utils.py
    │   │   │   ├── vision_config.py
    │   │   │   ├── vision_model.py
    │   │   │   └── vision_transformer_block.py
    │   │   ├── readme.md
    │   │   ├── registry.py
    │   │   ├── saver.py
    │   │   ├── util.py
    │   │   └── weight_converter.py
    │   ├── qwen2/
    │   │   ├── __init__.py
    │   │   └── megatron/
    │   │       ├── __init__.py
    │   │       ├── checkpoint_utils/
    │   │       │   ├── __init__.py
    │   │       │   ├── qwen2_loader.py
    │   │       │   ├── qwen2_loader_depracated.py
    │   │       │   └── qwen2_saver.py
    │   │       ├── layers/
    │   │       │   ├── __init__.py
    │   │       │   ├── parallel_attention.py
    │   │       │   ├── parallel_decoder.py
    │   │       │   ├── parallel_linear.py
    │   │       │   ├── parallel_mlp.py
    │   │       │   └── parallel_rmsnorm.py
    │   │       └── modeling_qwen2_megatron.py
    │   ├── registry.py
    │   ├── transformers/
    │   │   ├── __init__.py
    │   │   ├── apertus.py
    │   │   ├── dense_common.py
    │   │   ├── glm4v.py
    │   │   ├── kimi_vl.py
    │   │   ├── llama.py
    │   │   ├── monkey_patch.py
    │   │   ├── npu_patch.py
    │   │   ├── qwen2.py
    │   │   ├── qwen2_vl.py
    │   │   ├── qwen3_vl.py
    │   │   └── tiled_mlp.py
    │   └── weight_loader_registry.py
    ├── protocol.py
    ├── py.typed
    ├── single_controller/
    │   ├── __init__.py
    │   ├── base/
    │   │   ├── __init__.py
    │   │   ├── decorator.py
    │   │   ├── worker.py
    │   │   └── worker_group.py
    │   └── ray/
    │       ├── __init__.py
    │       └── base.py
    ├── third_party/
    │   ├── __init__.py
    │   ├── torch/
    │   │   ├── __init__.py
    │   │   └── distributed/
    │   │       ├── __init__.py
    │   │       ├── _state_dict_utils.py
    │   │       └── checkpoint/
    │   │           ├── __init__.py
    │   │           └── state_dict.py
    │   └── vllm/
    │       └── __init__.py
    ├── tools/
    │   ├── __init__.py
    │   ├── base_tool.py
    │   ├── geo3k_tool.py
    │   ├── gsm8k_tool.py
    │   ├── image_zoom_in_tool.py
    │   ├── mcp_base_tool.py
    │   ├── mcp_search_tool.py
    │   ├── sandbox_fusion_tools.py
    │   ├── schemas.py
    │   ├── search_tool.py
    │   └── utils/
    │       ├── __init__.py
    │       ├── mcp_clients/
    │       │   ├── McpClientManager.py
    │       │   └── utils.py
    │       ├── search_r1_like_utils.py
    │       └── tool_registry.py
    ├── trainer/
    │   ├── README.md
    │   ├── __init__.py
    │   ├── config/
    │   │   ├── __init__.py
    │   │   ├── _generated_ppo_megatron_trainer.yaml
    │   │   ├── _generated_ppo_torchtitan_trainer.yaml
    │   │   ├── _generated_ppo_trainer.yaml
    │   │   ├── _generated_ppo_veomni_trainer.yaml
    │   │   ├── actor/
    │   │   │   ├── actor.yaml
    │   │   │   ├── dp_actor.yaml
    │   │   │   ├── megatron_actor.yaml
    │   │   │   ├── torchtitan_actor.yaml
    │   │   │   └── veomni_actor.yaml
    │   │   ├── algorithm/
    │   │   │   └── rollout_correction.yaml
    │   │   ├── algorithm.py
    │   │   ├── config.py
    │   │   ├── critic/
    │   │   │   ├── critic.yaml
    │   │   │   ├── dp_critic.yaml
    │   │   │   ├── megatron_critic.yaml
    │   │   │   ├── torchtitan_critic.yaml
    │   │   │   └── veomni_critic.yaml
    │   │   ├── data/
    │   │   │   └── legacy_data.yaml
    │   │   ├── engine/
    │   │   │   ├── automodel.yaml
    │   │   │   ├── fsdp.yaml
    │   │   │   ├── megatron.yaml
    │   │   │   ├── torchtitan.yaml
    │   │   │   └── veomni.yaml
    │   │   ├── evaluation.yaml
    │   │   ├── legacy_reward_impl.yaml
    │   │   ├── model/
    │   │   │   └── hf_model.yaml
    │   │   ├── model_engine/
    │   │   │   ├── dp.yaml
    │   │   │   ├── torchtitan.yaml
    │   │   │   └── veomni.yaml
    │   │   ├── npu_profile/
    │   │   │   └── npu_profile.yaml
    │   │   ├── optim/
    │   │   │   ├── automodel.yaml
    │   │   │   ├── fsdp.yaml
    │   │   │   ├── megatron.yaml
    │   │   │   ├── torchtitan.yaml
    │   │   │   └── veomni.yaml
    │   │   ├── ppo_megatron_trainer.yaml
    │   │   ├── ppo_trainer.yaml
    │   │   ├── profiler/
    │   │   │   └── profiler.yaml
    │   │   ├── ref/
    │   │   │   ├── dp_ref.yaml
    │   │   │   ├── megatron_ref.yaml
    │   │   │   ├── ref.yaml
    │   │   │   ├── torchtitan_ref.yaml
    │   │   │   └── veomni_ref.yaml
    │   │   ├── reward/
    │   │   │   └── reward.yaml
    │   │   ├── rollout/
    │   │   │   └── rollout.yaml
    │   │   └── sft_trainer_engine.yaml
    │   ├── constants_ppo.py
    │   ├── main_eval.py
    │   ├── main_generation_server.py
    │   ├── main_ppo.py
    │   ├── ppo/
    │   │   ├── __init__.py
    │   │   ├── core_algos.py
    │   │   ├── metric_utils.py
    │   │   ├── prefix_grouper_utils.py
    │   │   ├── ray_trainer.py
    │   │   ├── reward.py
    │   │   ├── rollout_corr_helper.py
    │   │   └── utils.py
    │   ├── runtime_env.yaml
    │   ├── sft_trainer.py
    │   └── sft_trainer_ray.py
    ├── utils/
    │   ├── __init__.py
    │   ├── activation_offload.py
    │   ├── attention_utils.py
    │   ├── chat_template.py
    │   ├── checkpoint/
    │   │   ├── __init__.py
    │   │   ├── checkpoint_handler.py
    │   │   ├── checkpoint_manager.py
    │   │   ├── fsdp_checkpoint_manager.py
    │   │   └── megatron_checkpoint_manager.py
    │   ├── config.py
    │   ├── dataset/
    │   │   ├── README.md
    │   │   ├── __init__.py
    │   │   ├── dataset_utils.py
    │   │   ├── multiturn_sft_dataset.py
    │   │   ├── rl_dataset.py
    │   │   ├── rm_dataset.py
    │   │   └── vision_utils.py
    │   ├── debug/
    │   │   ├── __init__.py
    │   │   ├── metrics.py
    │   │   ├── performance.py
    │   │   └── trajectory_tracker.py
    │   ├── device.py
    │   ├── distributed.py
    │   ├── experimental/
    │   │   ├── __init__.py
    │   │   └── torch_functional.py
    │   ├── flops_counter.py
    │   ├── fp8_utils.py
    │   ├── fs.py
    │   ├── fsdp_utils.py
    │   ├── groupwise.py
    │   ├── hdfs_io.py
    │   ├── import_utils.py
    │   ├── kernel/
    │   │   ├── __init__.py
    │   │   ├── fp8_kernel.py
    │   │   ├── kernels.py
    │   │   └── linear_cross_entropy.py
    │   ├── logger/
    │   │   ├── __init__.py
    │   │   └── aggregate_logger.py
    │   ├── logging_utils.py
    │   ├── megatron/
    │   │   ├── __init__.py
    │   │   ├── dist_checkpointing.py
    │   │   ├── memory.py
    │   │   ├── optimizer.py
    │   │   ├── pipeline_parallel.py
    │   │   ├── router_replay_patch.py
    │   │   ├── router_replay_utils.py
    │   │   ├── sequence_parallel.py
    │   │   └── tensor_parallel.py
    │   ├── megatron_peft_utils.py
    │   ├── megatron_utils.py
    │   ├── memory_utils.py
    │   ├── metric/
    │   │   ├── __init__.py
    │   │   └── utils.py
    │   ├── model.py
    │   ├── net_utils.py
    │   ├── npu_flash_attn_utils.py
    │   ├── profiler/
    │   │   ├── __init__.py
    │   │   ├── config.py
    │   │   ├── empty_annotations.py
    │   │   ├── mstx_profile.py
    │   │   ├── nvtx_profile.py
    │   │   ├── performance.py
    │   │   ├── profile.py
    │   │   └── torch_profile.py
    │   ├── py_functional.py
    │   ├── qat/
    │   │   ├── __init__.py
    │   │   ├── core.py
    │   │   ├── linear.py
    │   │   ├── quantizer.py
    │   │   └── vllm_patch.py
    │   ├── ray_utils.py
    │   ├── rendezvous/
    │   │   ├── __init__.py
    │   │   └── ray_backend.py
    │   ├── reward_score/
    │   │   ├── __init__.py
    │   │   ├── geo3k.py
    │   │   ├── gsm8k.py
    │   │   ├── math_batch.py
    │   │   ├── math_dapo.py
    │   │   ├── math_reward.py
    │   │   ├── math_verify.py
    │   │   ├── prime_code/
    │   │   │   ├── README.md
    │   │   │   ├── __init__.py
    │   │   │   ├── testing_util.py
    │   │   │   └── utils.py
    │   │   ├── prime_math/
    │   │   │   ├── __init__.py
    │   │   │   ├── grader.py
    │   │   │   └── math_normalize.py
    │   │   ├── rlla.py
    │   │   ├── sandbox_fusion/
    │   │   │   ├── __init__.py
    │   │   │   └── utils.py
    │   │   └── search_r1_like_qa_em.py
    │   ├── rollout_skip.py
    │   ├── rollout_trace.py
    │   ├── seqlen_balancing.py
    │   ├── sglang/
    │   │   └── sglang_fp8_utils.py
    │   ├── tensordict_utils.py
    │   ├── tokenizer.py
    │   ├── torch_dtypes.py
    │   ├── torch_functional.py
    │   ├── tracking.py
    │   ├── transformers_compat.py
    │   ├── trtllm/
    │   │   └── trtllm_fp8_utils.py
    │   ├── ulysses.py
    │   └── vllm/
    │       ├── __init__.py
    │       ├── npu_vllm_patch.py
    │       ├── patch.py
    │       ├── utils.py
    │       └── vllm_fp8_utils.py
    ├── version/
    │   └── version
    └── workers/
        ├── __init__.py
        ├── actor/
        │   ├── __init__.py
        │   ├── base.py
        │   ├── dp_actor.py
        │   └── megatron_actor.py
        ├── config/
        │   ├── __init__.py
        │   ├── actor.py
        │   ├── critic.py
        │   ├── engine.py
        │   ├── megatron_peft.py
        │   ├── model.py
        │   ├── optimizer.py
        │   ├── reward.py
        │   └── rollout.py
        ├── critic/
        │   ├── __init__.py
        │   ├── base.py
        │   ├── dp_critic.py
        │   └── megatron_critic.py
        ├── engine/
        │   ├── __init__.py
        │   ├── automodel/
        │   │   ├── __init__.py
        │   │   ├── transformer_impl.py
        │   │   └── utils.py
        │   ├── base.py
        │   ├── fsdp/
        │   │   ├── __init__.py
        │   │   ├── transformer_impl.py
        │   │   └── utils.py
        │   ├── megatron/
        │   │   ├── __init__.py
        │   │   ├── transformer_impl.py
        │   │   └── utils.py
        │   ├── mindspeed/
        │   │   ├── __init__.py
        │   │   └── transformer_impl.py
        │   ├── torchtitan/
        │   │   ├── __init__.py
        │   │   ├── transformer_impl.py
        │   │   └── utils.py
        │   ├── utils.py
        │   └── veomni/
        │       ├── __init__.py
        │       ├── transformer_impl.py
        │       └── utils.py
        ├── engine_workers.py
        ├── fsdp_workers.py
        ├── megatron_workers.py
        ├── reward_manager/
        │   ├── __init__.py
        │   ├── abstract.py
        │   ├── batch.py
        │   ├── dapo.py
        │   ├── naive.py
        │   ├── prime.py
        │   └── registry.py
        ├── rollout/
        │   ├── __init__.py
        │   ├── base.py
        │   ├── hf_rollout.py
        │   ├── naive/
        │   │   ├── __init__.py
        │   │   └── naive_rollout.py
        │   ├── replica.py
        │   ├── schemas.py
        │   ├── sglang_rollout/
        │   │   ├── __init__.py
        │   │   ├── async_sglang_server.py
        │   │   ├── http_server_engine.py
        │   │   ├── sglang_rollout.py
        │   │   └── utils.py
        │   ├── tokenizer.py
        │   ├── trtllm_rollout/
        │   │   ├── trtllm_async_rollout.md
        │   │   ├── trtllm_async_server.py
        │   │   ├── trtllm_rollout.py
        │   │   └── trtllm_worker_extension.py
        │   ├── utils.py
        │   └── vllm_rollout/
        │       ├── __init__.py
        │       ├── bucketed_weight_transfer.py
        │       ├── utils.py
        │       ├── vllm_async_server.py
        │       └── vllm_rollout.py
        ├── sharding_manager/
        │   ├── __init__.py
        │   ├── base.py
        │   └── fsdp_ulysses.py
        └── utils/
            ├── __init__.py
            ├── losses.py
            └── padding.py

SYMBOL INDEX (4887 symbols across 489 files)

FILE: docs/_static/js/resizable-sidebar.js
  function setupNavigationFix (line 136) | function setupNavigationFix() {

FILE: examples/data_preprocess/aime2024_multiturn_w_tool.py
  function make_map_fn (line 49) | def make_map_fn(split):

FILE: examples/data_preprocess/dapo_multiturn_w_tool.py
  function make_map_fn (line 49) | def make_map_fn(split):

FILE: examples/data_preprocess/full_hh_rlhf.py
  function generate_sft_dataset (line 30) | def generate_sft_dataset(target_hdfs_path_dir, local_dir="~/data/full_hh...
  function generate_rm_dataset (line 61) | def generate_rm_dataset(target_hdfs_path_dir, local_dir="~/data/full_hh_...
  function generate_rl_dataset (line 93) | def generate_rl_dataset(target_hdfs_path_dir, local_dir="~/data/full_hh_...

FILE: examples/data_preprocess/geo3k.py
  function make_map_fn (line 58) | def make_map_fn(split):

FILE: examples/data_preprocess/geo3k_multiturn_w_tool.py
  function make_map_fn (line 60) | def make_map_fn(split):

FILE: examples/data_preprocess/gsm8k.py
  function extract_solution (line 27) | def extract_solution(solution_str):
  function make_map_fn (line 60) | def make_map_fn(split):

FILE: examples/data_preprocess/gsm8k_multiturn_sft.py
  function extract_solution (line 27) | def extract_solution(solution_str):
  function make_map_fn (line 60) | def make_map_fn(split):

FILE: examples/data_preprocess/gsm8k_multiturn_w_interaction.py
  function extract_solution (line 29) | def extract_solution(solution_str):
  function make_map_fn (line 62) | def make_map_fn(split):

FILE: examples/data_preprocess/gsm8k_multiturn_w_tool.py
  function extract_solution (line 29) | def extract_solution(solution_str):
  function make_map_fn (line 62) | def make_map_fn(split):

FILE: examples/data_preprocess/gsm8k_tool_agent_loop.py
  function extract_solution (line 29) | def extract_solution(solution_str):
  function make_map_fn (line 62) | def make_map_fn(split):

FILE: examples/data_preprocess/hellaswag.py
  function preprocess (line 28) | def preprocess(text):
  function make_map_fn (line 62) | def make_map_fn(split):

FILE: examples/data_preprocess/math_dataset.py
  function extract_solution (line 28) | def extract_solution(solution_str):
  function make_map_fn (line 63) | def make_map_fn(split):

FILE: examples/data_preprocess/multiturn.py
  function main (line 24) | def main():

FILE: examples/data_preprocess/pokemon.py
  function map_fn (line 38) | def map_fn(row: dict):

FILE: examples/data_preprocess/preprocess_search_r1_dataset.py
  function process_single_row (line 45) | def process_single_row(row, current_split_name, row_index):
  function main (line 101) | def main():

FILE: examples/fapo_trainer/prepare_data.py
  function example_map_fn (line 27) | def example_map_fn(example, idx, process_fn, data_source, ability, split):
  function build_aime2024_dataset (line 39) | def build_aime2024_dataset():
  function build_aime2025_dataset (line 53) | def build_aime2025_dataset():
  function build_gpqa_diamond_dataset (line 67) | def build_gpqa_diamond_dataset():
  function build_dapo_train_dataset (line 107) | def build_dapo_train_dataset():

FILE: examples/fapo_trainer/reward_fn.py
  function verify (line 29) | def verify(
  function compute_score_baseline (line 45) | async def compute_score_baseline(
  function post_request (line 95) | async def post_request(router_address: str, payload: dict, endpoint: str...
  function compute_score_fapo (line 134) | async def compute_score_fapo(

FILE: examples/sglang_multiturn/gsm8k_toolcall_shaping/gsm8k_toolcall_shaping.py
  function toolcall_shaping_reward (line 23) | def toolcall_shaping_reward(
  function compute_score (line 46) | def compute_score(

FILE: examples/sglang_multiturn/search_r1_like/local_dense_retriever/retrieval_server.py
  function load_corpus (line 34) | def load_corpus(corpus_path: str):
  function load_docs (line 39) | def load_docs(corpus, doc_idxs):
  function load_model (line 44) | def load_model(model_path: str, use_fp16: bool = False):
  function pooling (line 54) | def pooling(pooler_output, last_hidden_state, attention_mask=None, pooli...
  class Encoder (line 66) | class Encoder:
    method __init__ (line 67) | def __init__(self, model_name, model_path, pooling_method, max_length,...
    method encode (line 78) | def encode(self, query_list: list[str], is_query=True) -> np.ndarray:
  class BaseRetriever (line 124) | class BaseRetriever:
    method __init__ (line 125) | def __init__(self, config):
    method _search (line 133) | def _search(self, query: str, num: int, return_score: bool):
    method _batch_search (line 136) | def _batch_search(self, query_list: list[str], num: int, return_score:...
    method search (line 139) | def search(self, query: str, num: int = None, return_score: bool = Fal...
    method batch_search (line 142) | def batch_search(self, query_list: list[str], num: int = None, return_...
  class BM25Retriever (line 146) | class BM25Retriever(BaseRetriever):
    method __init__ (line 147) | def __init__(self, config):
    method _check_contain_doc (line 157) | def _check_contain_doc(self):
    method _search (line 160) | def _search(self, query: str, num: int = None, return_score: bool = Fa...
    method _batch_search (line 193) | def _batch_search(self, query_list: list[str], num: int = None, return...
  class DenseRetriever (line 206) | class DenseRetriever(BaseRetriever):
    method __init__ (line 207) | def __init__(self, config):
    method _search (line 227) | def _search(self, query: str, num: int = None, return_score: bool = Fa...
    method _batch_search (line 240) | def _batch_search(self, query_list: list[str], num: int = None, return...
  function get_retriever (line 273) | def get_retriever(config):
  class Config (line 285) | class Config:
    method __init__ (line 291) | def __init__(
  class QueryRequest (line 320) | class QueryRequest(BaseModel):
  function retrieve_endpoint (line 330) | def retrieve_endpoint(request: QueryRequest):

FILE: examples/split_placement/main_ppo_split.py
  function _select_rm_score_fn (line 30) | def _select_rm_score_fn(data_source):
  class RewardManager (line 39) | class RewardManager:
    method __init__ (line 40) | def __init__(self, tokenizer, num_examine) -> None:
    method __call__ (line 44) | def __call__(self, data: DataProto, return_dict: bool = False):
  function main (line 96) | def main(config):
  function main_task (line 111) | def main_task(config):

FILE: examples/split_placement/split_monkey_patch.py
  function fit (line 38) | def fit(self):

FILE: examples/tutorial/agent_loop_get_started/sandbox.py
  class SandboxTool (line 22) | class SandboxTool(BaseTool):
    method __init__ (line 23) | def __init__(self, config: dict, tool_schema: OpenAIFunctionToolSchema):
    method code_interpreter (line 28) | async def code_interpreter(self, code: str) -> str:
    method get_openai_tool_schema (line 47) | def get_openai_tool_schema(self) -> OpenAIFunctionToolSchema:
    method execute (line 51) | async def execute(self, instance_id: str, parameters: dict, **kwargs) ...

FILE: scripts/converter_hf_to_mcore.py
  function _init_args (line 51) | def _init_args():
  function test_conversion (line 73) | def test_conversion(megatron_model_provider, tfconfig, output_path, model):
  function convert_checkpoint_from_transformers_to_megatron (line 122) | def convert_checkpoint_from_transformers_to_megatron(
  function safe_copy (line 209) | def safe_copy(
  function convert_checkpoint_from_transformers_to_megatron_qwen2_5_vl (line 223) | def convert_checkpoint_from_transformers_to_megatron_qwen2_5_vl(hfmodel,...
  function convert_checkpoint_from_transformers_to_megatron_dpskv3 (line 330) | def convert_checkpoint_from_transformers_to_megatron_dpskv3(
  function noop_context (line 434) | def noop_context() -> Any:
  function support_distributed_convert (line 438) | def support_distributed_convert(hf_config: AutoConfig) -> bool:
  function convert_hf_to_mcore (line 445) | def convert_hf_to_mcore(

FILE: scripts/diagnose.py
  function test_connection (line 50) | def test_connection(name, url, timeout=10):
  function check_python (line 70) | def check_python():
  function check_pip (line 78) | def check_pip():
  function _get_current_git_commit (line 89) | def _get_current_git_commit():
  function check_verl (line 101) | def check_verl():
  function check_os (line 126) | def check_os():
  function check_hardware (line 135) | def check_hardware():
  function check_network (line 151) | def check_network(args):
  function check_environment (line 170) | def check_environment():
  function check_pip_package_versions (line 177) | def check_pip_package_versions():
  function check_cuda_versions (line 187) | def check_cuda_versions():
  function _get_cpu_memory (line 208) | def _get_cpu_memory():
  function _get_gpu_info (line 216) | def _get_gpu_info():
  function _get_system_info (line 244) | def _get_system_info():
  function check_system_info (line 253) | def check_system_info():
  function parse_args (line 263) | def parse_args():

FILE: scripts/init_random_model.py
  function _init_args (line 37) | def _init_args():
  function check_output_path (line 51) | def check_output_path(output_path: str):
  function check_configs (line 60) | def check_configs(original_config: dict[str, Any], new_config: dict[str,...
  function init_random_model (line 77) | def init_random_model(hf_model_path, new_config_path, output_path, trust...

FILE: scripts/legacy_model_merger.py
  class ModelMergerConfig (line 77) | class ModelMergerConfig:
    method __post_init__ (line 91) | def __post_init__(self):
  class BaseModelMerger (line 99) | class BaseModelMerger(ABC):
    method __init__ (line 100) | def __init__(self, config: ModelMergerConfig):
    method get_transformers_auto_model_class (line 117) | def get_transformers_auto_model_class(self):
    method patch_model_generation_config (line 141) | def patch_model_generation_config(self, model):
    method save_lora_adapter (line 157) | def save_lora_adapter(self, state_dict: dict[str, torch.Tensor]):
    method save_hf_model_and_tokenizer (line 214) | def save_hf_model_and_tokenizer(self, state_dict: dict[str, torch.Tens...
    method upload_to_huggingface (line 243) | def upload_to_huggingface(self):
    method merge_and_save (line 251) | def merge_and_save(self):
  class FSDPModelMerger (line 255) | class FSDPModelMerger(BaseModelMerger):
    method _get_world_size (line 256) | def _get_world_size(self) -> int:
    method _load_rank_zero_state_dict (line 266) | def _load_rank_zero_state_dict(self, world_size: int) -> dict:
    method _extract_device_mesh_info (line 273) | def _extract_device_mesh_info(self, state_dict: dict, world_size: int)...
    method _calculate_shard_configuration (line 293) | def _calculate_shard_configuration(
    method _merge_by_placement (line 309) | def _merge_by_placement(self, tensors: list[torch.Tensor], placement: ...
    method _load_and_merge_state_dicts (line 320) | def _load_and_merge_state_dicts(
    method merge_and_save (line 383) | def merge_and_save(self):
    method _test_state_dict (line 406) | def _test_state_dict(self, state_dict: dict[str, torch.Tensor]):
  class MegatronModelMerger (line 440) | class MegatronModelMerger(BaseModelMerger):
    method __init__ (line 441) | def __init__(self, config: ModelMergerConfig):
    method _get_tp_pp_rank_from_sharded_dir (line 484) | def _get_tp_pp_rank_from_sharded_dir(self, sharded_dir: str) -> tuple[...
    method _check_megatron_checkpoint_path (line 498) | def _check_megatron_checkpoint_path(self, model_path: str) -> tuple[li...
    method _merge_across_tp (line 513) | def _merge_across_tp(
    method _load_state_dicts (line 569) | def _load_state_dicts(
    method _check_megatron_state_key (line 587) | def _check_megatron_state_key(self, key: str) -> bool:
    method _merge_state_dicts (line 611) | def _merge_state_dicts(
    method merge_and_save (line 663) | def merge_and_save(self):
    method _test_state_dict (line 685) | def _test_state_dict(self, state_dict: dict[str, torch.Tensor]):
    method _replace_name (line 706) | def _replace_name(self, megatron_name: str, name_mapping: dict[str, st...
  function main (line 718) | def main():

FILE: scripts/megatron_merge_lora.py
  class CustomSaveWorker (line 33) | class CustomSaveWorker(ActorRolloutRefWorker):
    method save_merged_weights (line 35) | def save_merged_weights(self, hf_ckpt_path):
  function main (line 56) | def main(config):
  function run_merge (line 69) | def run_merge(config) -> None:
  function main_task (line 84) | def main_task(config):

FILE: scripts/print_cfg.py
  function main (line 21) | def main(config):

FILE: scripts/rollout_viewer.py
  function check_textual_version (line 42) | def check_textual_version():
  function load_path (line 54) | async def load_path(p: Path, data: dict, mask_strs: str, idx: int, pbar):
  function load_dir (line 74) | async def load_dir(path: Path, data: dict[int, dict], pbar, mask_strs: s...
  class Highlighter (line 83) | class Highlighter(ReprHighlighter):
  function center_word_with_equals_exactly (line 90) | def center_word_with_equals_exactly(word: str, total_length: int, char: ...
  function highlight_keyword (line 100) | def highlight_keyword(content: str, keyword: Optional[str]):
  class JsonLineViewer (line 129) | class JsonLineViewer(App):
    method __init__ (line 175) | def __init__(self, step_num: int, data: dict[int, dict], pbar):
    method compose (line 200) | def compose(self) -> ComposeResult:
    method on_mount (line 250) | async def on_mount(self) -> None:
    method update_result_options (line 268) | def update_result_options(self, offset: int = 0, sort_desc: Optional[b...
    method update_content (line 292) | async def update_content(self, search_keyword: Optional[str] = None):
    method on_reqid_submitted (line 332) | async def on_reqid_submitted(self, event: Input.Submitted) -> None:
    method _update_fields_select (line 373) | def _update_fields_select(self, keys):
    method step_changed (line 395) | async def step_changed(self, event):
    method sample_changed (line 401) | async def sample_changed(self, event):
    method sort_changed (line 407) | async def sort_changed(self, event):
    method fields_changed (line 413) | async def fields_changed(self, event):
    method fields_all_changed (line 417) | async def fields_all_changed(self, event):
    method action_focus_previous (line 424) | def action_focus_previous(self):
    method action_focus_next (line 427) | def action_focus_next(self):
    method action_next_step (line 430) | async def action_next_step(self) -> None:
    method action_next_sample (line 438) | async def action_next_sample(self) -> None:
    method action_previous_step (line 446) | async def action_previous_step(self) -> None:
    method action_previous_sample (line 454) | async def action_previous_sample(self) -> None:
    method action_swith_render (line 462) | async def action_swith_render(self):
    method action_toggle_search (line 466) | def action_toggle_search(self) -> None:
    method action_cancel_search (line 469) | async def action_cancel_search(self) -> None:
    method _clear_search (line 474) | async def _clear_search(self):
    method on_search_submitted (line 480) | async def on_search_submitted(self, event: Input.Submitted) -> None:
    method action_next_search (line 507) | async def action_next_search(self) -> None:
    method action_page_up (line 521) | def action_page_up(self):
    method action_page_down (line 524) | def action_page_down(self):
    method action_page_home (line 527) | def action_page_home(self):
    method action_page_end (line 530) | def action_page_end(self):
  function _run (line 534) | async def _run(path: Path, mask_str: str):
  function run (line 556) | def run(

FILE: scripts/veomni/moe_merge.py
  class StateDictIterator (line 46) | class StateDictIterator:
    method __iter__ (line 49) | def __iter__(self) -> Generator[tuple[str, "torch.Tensor"], None, None]:
  function main (line 61) | def main(raw_hf_path, merge_hf_path):

FILE: scripts/veomni/moe_split.py
  class StateDictIterator (line 43) | class StateDictIterator:
    method __iter__ (line 46) | def __iter__(self) -> Generator[tuple[str, "torch.Tensor"], None, None]:
  function main (line 58) | def main(merge_hf_path, split_hf_path):

FILE: tests/checkpoint_engine/test_correctness_on_gpu.py
  function test_nccl_checkpoint_engine (line 34) | async def test_nccl_checkpoint_engine(
  function test_nixl_checkpoint_engine (line 83) | async def test_nixl_checkpoint_engine(
  function test_kimi_checkpoint_engine (line 139) | async def test_kimi_checkpoint_engine(

FILE: tests/checkpoint_engine/test_correctness_on_npu.py
  function test_hccl_checkpoint_engine (line 34) | async def test_hccl_checkpoint_engine(
  function test_kimi_checkpoint_engine (line 83) | async def test_kimi_checkpoint_engine(
  function test_mooncake_checkpoint_engine (line 130) | async def test_mooncake_checkpoint_engine(

FILE: tests/checkpoint_engine/test_special_server_adapter.py
  function init_config (line 34) | def init_config() -> DictConfig:
  function _run_update_weights_with_global_steps_none (line 57) | async def _run_update_weights_with_global_steps_none(
  function _run_server_manager_without_resume (line 83) | async def _run_server_manager_without_resume(
  function _run_server_manager_with_resume (line 124) | async def _run_server_manager_with_resume(
  function test_server_adapter (line 175) | async def test_server_adapter(init_config):

FILE: tests/checkpoint_engine/test_utils.py
  class TrainingWorkerTest (line 31) | class TrainingWorkerTest(TrainingWorker):
    method __init__ (line 32) | def __init__(self, config: TrainingWorkerConfig, checkpoint_engine_con...
    method update_weights (line 43) | async def update_weights(self, global_steps: int = None):
    method execute_checkpoint_engine (line 48) | def execute_checkpoint_engine(self, method: str, *args, **kwargs):
  class MockServerAdapter (line 52) | class MockServerAdapter(BaseRollout):
    method __init__ (line 53) | def __init__(self, config: RolloutConfig, model_config: HFModelConfig,...
    method resume (line 59) | async def resume(self, tags: list[str]):
    method release (line 62) | async def release(self):
    method update_weights (line 65) | async def update_weights(
    method check_weights (line 75) | def check_weights(self):
  class MockReplica (line 90) | class MockReplica(RolloutReplica):
    method init_hybrid (line 91) | async def init_hybrid(self, worker_group: RayWorkerGroup):
    method get_ray_class_with_init_args (line 101) | def get_ray_class_with_init_args(self) -> RayClassWithInitArgs:
    method launch_servers (line 105) | async def launch_servers(self):
  class CheckpointEngineWorkerTest (line 110) | class CheckpointEngineWorkerTest(CheckpointEngineWorker):
    method __init__ (line 111) | def __init__(
    method check_weights (line 118) | def check_weights(self):
  function create_trainer_worker_group (line 122) | def create_trainer_worker_group(
  function create_rollout_worker_group (line 150) | async def create_rollout_worker_group(

FILE: tests/experimental/agent_loop/agent_utils.py
  function init_agent_loop_manager (line 28) | def init_agent_loop_manager(config: DictConfig) -> AgentLoopManager | Ra...

FILE: tests/experimental/agent_loop/test_agent_loop_extra_fields_schema_on_cpu.py
  class _FakeServerManager (line 37) | class _FakeServerManager:
    method generate (line 38) | async def generate(
    method generate_for_partial (line 51) | async def generate_for_partial(
  class _FakeTokenizer (line 67) | class _FakeTokenizer:
    method apply_chat_template (line 70) | def apply_chat_template(
    method pad (line 83) | def pad(
    method decode (line 113) | def decode(self, ids: list[int] | torch.Tensor, skip_special_tokens: b...
  function _pad_1d (line 118) | def _pad_1d(ids: list[int], *, length: int, pad_id: int = 0) -> list[int]:
  function _to_internal (line 124) | def _to_internal(
  function test_agent_loop_extra_fields_schema_stable_for_training_concat_on_cpu (line 170) | async def test_agent_loop_extra_fields_schema_stable_for_training_concat...
  function test_agent_loop_postprocess_accepts_read_only_routed_experts_on_cpu (line 252) | async def test_agent_loop_postprocess_accepts_read_only_routed_experts_o...

FILE: tests/experimental/agent_loop/test_basic_agent_loop.py
  function init_config (line 36) | def init_config() -> DictConfig:
  function test_single_turn (line 68) | def test_single_turn(init_config):
  class WeatherTool (line 130) | class WeatherTool(BaseTool):
    method get_current_temperature (line 131) | def get_current_temperature(self, location: str, unit: str = "celsius"):
    method get_openai_tool_schema (line 148) | def get_openai_tool_schema(self) -> OpenAIFunctionToolSchema:
    method execute (line 152) | async def execute(self, instance_id: str, parameters: dict[str, Any], ...
  class WeatherToolWithData (line 160) | class WeatherToolWithData(BaseTool):
    method get_openai_tool_schema (line 161) | def get_openai_tool_schema(self) -> OpenAIFunctionToolSchema:
    method get_temperature_date (line 165) | def get_temperature_date(self, location: str, date: str, unit: str = "...
    method execute (line 184) | async def execute(self, instance_id: str, parameters: dict[str, Any], ...
  function test_tool_agent (line 192) | def test_tool_agent(init_config):
  function test_tool_agent_with_interaction (line 306) | def test_tool_agent_with_interaction(init_config):
  function test_get_trajectory_info (line 441) | async def test_get_trajectory_info():
  function ray_for_lb (line 464) | def ray_for_lb():
  class TestLoadBalancerRouting (line 470) | class TestLoadBalancerRouting:
    method test_distributes_across_servers (line 473) | def test_distributes_across_servers(self, ray_for_lb):
    method test_new_requests_route_to_least_loaded (line 478) | def test_new_requests_route_to_least_loaded(self, ray_for_lb):
    method test_release_rebalances (line 490) | def test_release_rebalances(self, ray_for_lb):
    method test_release_invalid_server_raises (line 501) | def test_release_invalid_server_raises(self, ray_for_lb):
    method test_release_without_inflight_raises (line 507) | def test_release_without_inflight_raises(self, ray_for_lb):
  class TestLoadBalancerStickySession (line 514) | class TestLoadBalancerStickySession:
    method test_same_request_id_same_server (line 517) | def test_same_request_id_same_server(self, ray_for_lb):

FILE: tests/experimental/agent_loop/test_gpt_oss_tool_parser.py
  function test_gpt_oss_tool_parser (line 22) | async def test_gpt_oss_tool_parser():

FILE: tests/experimental/agent_loop/test_multi_modal.py
  function parse_multi_modal_type (line 32) | def parse_multi_modal_type(messages: list[dict]) -> str:
  function init_config (line 47) | def init_config() -> DictConfig:
  class ImageGeneratorTool (line 75) | class ImageGeneratorTool(BaseTool):
    method generate_image (line 76) | def generate_image(self, description: str, size: str = "256x256"):
    method get_openai_tool_schema (line 113) | def get_openai_tool_schema(self) -> OpenAIFunctionToolSchema:
    method execute (line 117) | async def execute(self, instance_id: str, parameters: dict[str, Any], ...
  function test_multimodal_tool_agent (line 127) | def test_multimodal_tool_agent(init_config):
  function test_multimodal_single_turn_agent (line 297) | def test_multimodal_single_turn_agent(init_config):

FILE: tests/experimental/agent_loop/test_standalone_rollout.py
  function init_config (line 29) | def init_config() -> DictConfig:
  function test_standalone_rollout (line 48) | async def test_standalone_rollout(init_config, tp_size):
  function test_hybrid_rollout_with_ep (line 104) | def test_hybrid_rollout_with_ep(init_config):

FILE: tests/experimental/reward_loop/reward_fn.py
  function chat_complete (line 41) | async def chat_complete(router_address: str, chat_complete_request: dict):
  function compute_score_gsm8k (line 56) | async def compute_score_gsm8k(
  function compute_score_math_verify (line 87) | def compute_score_math_verify(

FILE: tests/experimental/reward_loop/test_agent_reward_loop_colocate.py
  function test_agent_reward_loop_standalone (line 34) | def test_agent_reward_loop_standalone():

FILE: tests/experimental/reward_loop/test_agent_reward_loop_standalone.py
  function test_agent_reward_loop_standalone (line 28) | def test_agent_reward_loop_standalone():

FILE: tests/experimental/reward_loop/test_async_token_bucket_on_cpu.py
  class TestAsyncTokenBucket (line 23) | class TestAsyncTokenBucket:
    method test_basic_acquire (line 27) | async def test_basic_acquire(self):
    method test_refill_mechanism (line 40) | async def test_refill_mechanism(self):
    method test_waiting_for_tokens (line 59) | async def test_waiting_for_tokens(self):
    method test_max_tokens_cap (line 75) | async def test_max_tokens_cap(self):
    method test_fractional_tokens (line 90) | async def test_fractional_tokens(self):
    method test_concurrent_acquires (line 102) | async def test_concurrent_acquires(self):
    method test_high_rate_limit (line 123) | async def test_high_rate_limit(self):
    method test_zero_initial_state (line 137) | async def test_zero_initial_state(self):
    method test_rate_limit_accuracy (line 149) | async def test_rate_limit_accuracy(self):
    method test_sequential_acquires (line 166) | async def test_sequential_acquires(self):
    method test_default_max_tokens (line 187) | async def test_default_max_tokens(self):
    method test_single_token_acquire (line 195) | async def test_single_token_acquire(self):
    method test_large_token_acquire (line 204) | async def test_large_token_acquire(self):
    method test_thread_safety_with_lock (line 217) | async def test_thread_safety_with_lock(self):
    method test_multiple_wait_cycles (line 237) | async def test_multiple_wait_cycles(self):
    method test_rapid_small_acquires (line 253) | async def test_rapid_small_acquires(self):

FILE: tests/experimental/reward_loop/test_math_verify.py
  function test_agent_reward_loop_standalone (line 27) | def test_agent_reward_loop_standalone():

FILE: tests/experimental/reward_loop/test_rate_limited_reward_manager_on_cpu.py
  class MockAPICounter (line 29) | class MockAPICounter:
    method __init__ (line 32) | def __init__(self):
    method record_call (line 37) | async def record_call(self):
    method reset (line 42) | def reset(self):
    method get_rate_per_second (line 46) | def get_rate_per_second(self, window_start: float = None):
  function mock_sync_reward_function (line 70) | def mock_sync_reward_function(
  function mock_async_reward_function (line 82) | async def mock_async_reward_function(
  function mock_slow_api_function (line 97) | async def mock_slow_api_function(
  function mock_failing_api_function (line 105) | async def mock_failing_api_function(
  function mock_dict_result_function (line 113) | async def mock_dict_result_function(
  function create_test_data_proto (line 124) | def create_test_data_proto(tokenizer, response_text: str, ground_truth: ...
  class TestRateLimitedRewardManager (line 143) | class TestRateLimitedRewardManager:
    method setup_and_teardown (line 147) | def setup_and_teardown(self):
    method tokenizer (line 160) | def tokenizer(self):
    method test_basic_reward_computation (line 165) | async def test_basic_reward_computation(self, tokenizer):
    method test_rpm_rate_limiting (line 183) | async def test_rpm_rate_limiting(self, tokenizer):
    method test_tpm_rate_limiting (line 218) | async def test_tpm_rate_limiting(self, tokenizer):
    method test_concurrency_limiting (line 254) | async def test_concurrency_limiting(self, tokenizer):
    method test_timeout_handling (line 287) | async def test_timeout_handling(self, tokenizer):
    method test_error_handling (line 311) | async def test_error_handling(self, tokenizer):
    method test_dict_result_format (line 330) | async def test_dict_result_format(self, tokenizer):
    method test_sync_reward_function (line 347) | async def test_sync_reward_function(self, tokenizer):
    method test_combined_rate_limits (line 362) | async def test_combined_rate_limits(self, tokenizer):
    method test_correct_vs_incorrect_answers (line 398) | async def test_correct_vs_incorrect_answers(self, tokenizer):
    method test_high_throughput (line 417) | async def test_high_throughput(self, tokenizer):
    method test_class_initialization_once (line 459) | async def test_class_initialization_once(self, tokenizer):
    method test_extra_info_handling (line 474) | async def test_extra_info_handling(self, tokenizer):

FILE: tests/experimental/reward_loop/test_reward_model_disrm.py
  function create_data_samples (line 27) | def create_data_samples(tokenizer) -> DataProto:
  function test_reward_model_manager (line 107) | def test_reward_model_manager():

FILE: tests/experimental/reward_loop/test_reward_model_genrm.py
  function create_data_samples (line 28) | def create_data_samples(tokenizer) -> DataProto:
  function test_reward_model_manager (line 108) | def test_reward_model_manager():

FILE: tests/experimental/vla/test_sim_envs.py
  function test_sim_env_creation_and_step (line 25) | def test_sim_env_creation_and_step(simulator_type):

FILE: tests/interactions/test_gsm8k_interaction.py
  class TestGsm8kInteraction (line 24) | class TestGsm8kInteraction:
    method setup_method (line 27) | def setup_method(self):
    method test_init (line 32) | def test_init(self):
    method test_start_interaction_with_instance_id (line 39) | async def test_start_interaction_with_instance_id(self):
    method test_start_interaction_without_instance_id (line 53) | async def test_start_interaction_without_instance_id(self):
    method test_start_interaction_without_ground_truth (line 65) | async def test_start_interaction_without_ground_truth(self):
    method test_generate_response_correct_answer_with_prefix (line 75) | async def test_generate_response_correct_answer_with_prefix(self):
    method test_generate_response_correct_answer_without_prefix (line 97) | async def test_generate_response_correct_answer_without_prefix(self):
    method test_generate_response_incorrect_answer (line 118) | async def test_generate_response_incorrect_answer(self):
    method test_generate_response_multiple_messages (line 139) | async def test_generate_response_multiple_messages(self):
    method test_generate_response_no_assistant_message (line 164) | async def test_generate_response_no_assistant_message(self):
    method test_calculate_score_direct_call (line 183) | async def test_calculate_score_direct_call(self):
    method test_calculate_score_with_kwargs (line 201) | async def test_calculate_score_with_kwargs(self):
    method test_finalize_interaction (line 219) | async def test_finalize_interaction(self):
    method test_finalize_interaction_with_kwargs (line 234) | async def test_finalize_interaction_with_kwargs(self):
    method test_finalize_nonexistent_interaction (line 249) | async def test_finalize_nonexistent_interaction(self):
    method test_full_interaction_workflow_correct (line 258) | async def test_full_interaction_workflow_correct(self):
    method test_full_interaction_workflow_incorrect (line 281) | async def test_full_interaction_workflow_incorrect(self):
    method test_multiple_concurrent_interactions (line 316) | async def test_multiple_concurrent_interactions(self):
    method test_edge_case_empty_messages (line 349) | async def test_edge_case_empty_messages(self):
    method test_edge_case_message_without_content (line 369) | async def test_edge_case_message_without_content(self):
    method test_inheritance_from_base_interaction (line 390) | def test_inheritance_from_base_interaction(self):
    method test_name_attribute_initialization (line 408) | def test_name_attribute_initialization(self):

FILE: tests/interactions/test_interaction_registry.py
  class TestInteractionRegistry (line 30) | class TestInteractionRegistry:
    method test_get_interaction_class (line 31) | def test_get_interaction_class(self):
    method test_initialize_single_interaction_from_config (line 41) | def test_initialize_single_interaction_from_config(self):
    method test_initialize_multiple_interactions_from_config (line 69) | def test_initialize_multiple_interactions_from_config(self):
    method test_initialize_interaction_without_explicit_name (line 111) | def test_initialize_interaction_without_explicit_name(self):
    method test_initialize_empty_config (line 132) | def test_initialize_empty_config(self):
    method test_invalid_class_name (line 146) | def test_invalid_class_name(self):
    method test_duplicate_interaction_names (line 162) | def test_duplicate_interaction_names(self):
    method test_auto_name_generation_edge_cases (line 185) | def test_auto_name_generation_edge_cases(self):

FILE: tests/models/test_engine.py
  function get_test_language_model (line 56) | def get_test_language_model(device_count):
  function create_training_config (line 65) | def create_training_config(model_type, strategy, device_count, model):
  function test_actor_engine (line 114) | def test_actor_engine(strategy):
  function create_value_model (line 234) | def create_value_model(language_model_path, output_path):
  function test_critic_engine (line 250) | def test_critic_engine(strategy):
  function create_actor_model (line 353) | def create_actor_model(tmp_path, config):
  function _worker (line 361) | def _worker(rank: int, world_size: int, rendezvous_file: str, strategy: ...
  function test_per_tensor_generator (line 431) | def test_per_tensor_generator(world_size, tmp_path, config, strategy):

FILE: tests/models/test_tiled_mlp_accuracy.py
  function setup_distributed (line 26) | def setup_distributed():
  function create_model (line 34) | def create_model(model_name="Qwen/Qwen3-1.7B", num_layers=2):
  function apply_fsdp2 (line 51) | def apply_fsdp2(model, device_mesh):
  function run_forward_backward (line 59) | def run_forward_backward(model, input_ids, labels):
  function compare_results (line 78) | def compare_results(logits1, grads1, logits2, grads2, rank):
  function main (line 119) | def main():

FILE: tests/models/test_transformer.py
  function test_hf_casual_models (line 47) | def test_hf_casual_models():
  function test_hf_value_models (line 117) | def test_hf_value_models():
  function test_attn_implementation_override (line 172) | def test_attn_implementation_override():
  function test_fsdp_worker_attn_implementation_integration (line 207) | def test_fsdp_worker_attn_implementation_integration():

FILE: tests/models/test_transformers_ulysses.py
  class SequenceParallelConfig (line 49) | class SequenceParallelConfig:
  function test_configs (line 55) | def test_configs():
  function sync_model_parameters_global (line 92) | def sync_model_parameters_global(layer):
  function test_hf_casual_fwd_bwd (line 99) | def test_hf_casual_fwd_bwd(test_config):
  function _hf_casual_fwd (line 112) | def _hf_casual_fwd(config, sp_size, dp_size):
  function _hf_casual_fwd_bwd (line 191) | def _hf_casual_fwd_bwd(config, sp_size, dp_size):

FILE: tests/single_controller/base/test_decorator.py
  function reset_dispatch_registry (line 29) | def reset_dispatch_registry():
  function test_register_new_dispatch_mode (line 38) | def test_register_new_dispatch_mode(reset_dispatch_registry):
  function test_update_existing_dispatch_mode (line 60) | def test_update_existing_dispatch_mode(reset_dispatch_registry):

FILE: tests/single_controller/check_worker_alive/main.py
  class TestActor (line 27) | class TestActor(Worker):
    method __init__ (line 28) | def __init__(self) -> None:
    method foo (line 32) | def foo(self, wait_time):

FILE: tests/single_controller/detached_worker/client.py
  function compute_position_id_with_mask (line 27) | def compute_position_id_with_mask(mask):

FILE: tests/single_controller/detached_worker/server.py
  class Trainer (line 44) | class Trainer(Worker):
    method __init__ (line 45) | def __init__(self):
    method init_model (line 74) | def init_model(self):
    method train_model (line 117) | def train_model(self, data: DataProto) -> DataProto:

FILE: tests/single_controller/test_auto_padding_on_cpu.py
  class Actor (line 30) | class Actor(Worker):
    method __init__ (line 31) | def __init__(self) -> None:
    method add (line 35) | def add(self, data: DataProto):
  function test_auto_padding (line 40) | def test_auto_padding():

FILE: tests/single_controller/test_colocated_workers.py
  class Actor (line 30) | class Actor(Worker):
    method __init__ (line 31) | def __init__(self) -> None:
    method add (line 35) | def add(self, data: DataProto):
  class Critic (line 41) | class Critic(Worker):
    method __init__ (line 42) | def __init__(self, config) -> None:
    method sub (line 47) | async def sub(self, data: DataProto):
  function test_colocated_workers (line 52) | def test_colocated_workers():

FILE: tests/single_controller/test_colocated_workers_fused.py
  class Actor (line 30) | class Actor(Worker):
    method __init__ (line 31) | def __init__(self) -> None:
    method add (line 35) | def add(self, data: DataProto):
  class Critic (line 41) | class Critic(Worker):
    method __init__ (line 42) | def __init__(self, config) -> None:
    method sub (line 47) | def sub(self, data: DataProto):
  function test_colocated_workers_fused (line 52) | def test_colocated_workers_fused():

FILE: tests/single_controller/test_data_transfer.py
  class DummyWorker (line 34) | class DummyWorker(Worker):
    method __init__ (line 35) | def __init__(self):
    method do_nothing (line 40) | def do_nothing(self, data):
  function test_data_transfer (line 48) | def test_data_transfer():

FILE: tests/single_controller/test_decorator_on_cpu.py
  function ray_init_shutdown (line 32) | def ray_init_shutdown():
  class DecoratorTestWorker (line 40) | class DecoratorTestWorker(Worker):
    method __init__ (line 41) | def __init__(self, initial_value=0):
    method dp_compute (line 51) | def dp_compute(self, data: DataProto) -> DataProto:
    method async_dp_compute (line 59) | async def async_dp_compute(self, data: DataProto) -> DataProto:
    method dp_compute_td (line 67) | def dp_compute_td(self, data: TensorDict) -> TensorDict:
  function test_decorator_dp_compute (line 82) | def test_decorator_dp_compute(ray_init_shutdown):
  function test_decorator_async_function (line 118) | def test_decorator_async_function(ray_init_shutdown):
  function test_decorator_dp_compute_td (line 161) | def test_decorator_dp_compute_td(ray_init_shutdown):

FILE: tests/single_controller/test_device_mesh_register.py
  class TestActor (line 29) | class TestActor(Worker):
    method __init__ (line 30) | def __init__(self):
    method generate_data_proto (line 56) | def generate_data_proto(self, data: DataProto):
    method generate_tensordict (line 63) | def generate_tensordict(self, data: TensorDict):
    method train_data_proto (line 70) | def train_data_proto(self, data: DataProto):
    method train_tensordict (line 80) | def train_tensordict(self, data: TensorDict):
    method generate_nested_tensor (line 90) | def generate_nested_tensor(self, data: TensorDict):
  function test_dist_global_info_wg (line 100) | def test_dist_global_info_wg():

FILE: tests/single_controller/test_driverfunc_to_worker.py
  class ModelActor (line 32) | class ModelActor(Worker):
    method __init__ (line 33) | def __init__(self):
  class HackSelf (line 37) | class HackSelf:
    method __init__ (line 38) | def __init__(self):
  function get_aux_metrics (line 42) | def get_aux_metrics(self, test_proto):
  function test (line 55) | def test():

FILE: tests/single_controller/test_fused_workers_on_cpu.py
  class Actor (line 28) | class Actor(Worker):
    method __init__ (line 29) | def __init__(self) -> None:
    method add (line 33) | def add(self, x):
  class Critic (line 39) | class Critic(Worker):
    method __init__ (line 40) | def __init__(self, val) -> None:
    method sub (line 45) | def sub(self, x):
  class HybridWorker (line 57) | class HybridWorker(FusedBaseClass):
    method foo (line 59) | def foo(self, x):
  function test_fused_workers (line 63) | def test_fused_workers():

FILE: tests/single_controller/test_get_set_dispatch_collect_cpu.py
  function test_get_set_dispatch_collect_cpu (line 21) | def test_get_set_dispatch_collect_cpu():

FILE: tests/single_controller/test_high_level_scheduling_api.py
  class TestActor (line 25) | class TestActor(Worker):
    method __init__ (line 27) | def __init__(self, cuda_visible_devices=None) -> None:
    method get_node_id (line 30) | def get_node_id(self):
  function test (line 34) | def test():

FILE: tests/single_controller/test_nested_worker.py
  class TestActor (line 24) | class TestActor(Worker):
    method __init__ (line 26) | def __init__(self, x) -> None:
    method get (line 31) | def get(self):
  class TestHighLevelActor (line 35) | class TestHighLevelActor(Worker):
    method __init__ (line 36) | def __init__(self, x=None) -> None:
    method get (line 41) | def get(self):
  function test_nested_worker (line 45) | def test_nested_worker():

FILE: tests/single_controller/test_ray_collectives.py
  class Actor (line 33) | class Actor(Worker):
    method init (line 35) | def init(self):
    method send_tensors (line 41) | def send_tensors(self):
  class Rollout (line 47) | class Rollout(Worker):
    method init (line 49) | def init(self):
    method receive_tensors (line 59) | def receive_tensors(self):
    method get_tensors (line 67) | def get_tensors(self):
  function test_ray_collective_group (line 71) | def test_ray_collective_group():

FILE: tests/single_controller/test_ray_local_envs_on_cpu.py
  class TestActor (line 27) | class TestActor(Worker):
    method __init__ (line 28) | def __init__(self) -> None:
    method getenv (line 31) | def getenv(self, key):
  function test_basics (line 36) | def test_basics():
  function test_customized_worker_env (line 53) | def test_customized_worker_env():

FILE: tests/single_controller/test_ray_utils_on_cpu.py
  function init_ray (line 23) | def init_ray():
  function test_parallel_put_basic (line 29) | def test_parallel_put_basic(init_ray):
  function test_parallel_put_empty (line 37) | def test_parallel_put_empty(init_ray):
  function test_parallel_put_workers (line 43) | def test_parallel_put_workers(init_ray):

FILE: tests/single_controller/test_rvdz.py
  class TestWorker (line 19) | class TestWorker:
    method __init__ (line 20) | def __init__(self, rank, world_size, group_name):
    method init (line 26) | def init(self):
    method test (line 31) | def test(self):
  function test_rvdz (line 37) | def test_rvdz():

FILE: tests/single_controller/test_split_resource_pool.py
  class Actor (line 33) | class Actor(Worker):
    method __init__ (line 34) | def __init__(self, worker_id) -> None:
    method add (line 45) | def add(self, data: DataProto):
  function test_split_resource_pool_with_split_size (line 50) | def test_split_resource_pool_with_split_size():
  function test_split_resource_pool_with_split_size_list (line 78) | def test_split_resource_pool_with_split_size_list():
  function test_split_resource_pool_with_split_size_list_cross_nodes (line 112) | def test_split_resource_pool_with_split_size_list_cross_nodes():
  function test_split_resource_pool_with_split_twice (line 147) | def test_split_resource_pool_with_split_twice():

FILE: tests/single_controller/test_worker_group_basics.py
  function two_to_all_dispatch_fn (line 27) | def two_to_all_dispatch_fn(worker_group, *args, **kwargs):
  function get_ray_remote_options (line 42) | def get_ray_remote_options() -> str:
  class TestActor (line 56) | class TestActor(Worker):
    method __init__ (line 58) | def __init__(self, x) -> None:
    method foo (line 62) | def foo(self, y):
    method foo_rank_zero (line 66) | def foo_rank_zero(self, x, y):
    method foo_one_to_all (line 70) | def foo_one_to_all(self, x, y):
    method foo_all_to_all (line 74) | def foo_all_to_all(self, x, y):
    method foo_custom (line 78) | def foo_custom(self, x, y):
  function remote_call_wg (line 83) | def remote_call_wg(worker_names):
  function add_one (line 99) | def add_one(data):
  function test_basics (line 106) | def test_basics():

FILE: tests/single_controller/test_worker_group_torch.py
  class TestAllGatherActor (line 30) | class TestAllGatherActor(Worker):
    method __init__ (line 31) | def __init__(self, size) -> None:
    method init (line 35) | def init(self):
    method all_gather (line 40) | def all_gather(self):
  class TestAllGatherActorV2 (line 50) | class TestAllGatherActorV2(Worker):
    method __init__ (line 51) | def __init__(self, size) -> None:
    method all_gather (line 59) | def all_gather(self):
  function test_all_gather_torch (line 68) | def test_all_gather_torch():
  function test_all_gather_torch_v2 (line 94) | def test_all_gather_torch_v2():

FILE: tests/special_distributed/test_fsdp_ckpt.py
  function create_random_input_ids (line 31) | def create_random_input_ids(batch_size, seq_len, vocab_size):
  function test_fsdp_ckpt (line 50) | def test_fsdp_ckpt(strategy="fsdp"):

FILE: tests/special_distributed/test_mcore_config_converter.py
  function check_config_converter_results (line 36) | def check_config_converter_results(tf_config: TransformerConfig | MLATra...
  function modify_hf_config (line 67) | def modify_hf_config(name: str, hf_config: PretrainedConfig):
  function test_mcore_config_converter (line 74) | def test_mcore_config_converter():

FILE: tests/special_distributed/test_tensor_dict.py
  function test_all_gather_data_proto (line 28) | def test_all_gather_data_proto():
  function test_vocab_parallel_entropy (line 61) | def test_vocab_parallel_entropy():

FILE: tests/special_e2e/check_custom_rwd_fn.py
  function check_congratulations_in_file (line 18) | def check_congratulations_in_file(output_file):

FILE: tests/special_e2e/check_results.py
  function extract_reward_from_line (line 20) | def extract_reward_from_line(line):

FILE: tests/special_e2e/envs/digit_completion/task.py
  class DigitCompletion (line 19) | class DigitCompletion:
    method __init__ (line 35) | def __init__(self, max_number: int, max_diff: int, max_num_in_response...
    method __str__ (line 56) | def __str__(self):
    method get_state (line 63) | def get_state(self):
    method set_state (line 66) | def set_state(self, state):
    method prompt_length (line 71) | def prompt_length(self):
    method response_length (line 75) | def response_length(self):
    method add (line 80) | def add(self, a, b):
    method get_all_prompts (line 83) | def get_all_prompts(self):
    method sample_str_prompts (line 93) | def sample_str_prompts(self):
    method sample_batch_str_prompts (line 102) | def sample_batch_str_prompts(self, batch_size):
  function compute_attention_mask (line 109) | def compute_attention_mask(prompts, pad_token_id):
  function compute_position_id_with_mask (line 115) | def compute_position_id_with_mask(mask):
  function generate_ground_truth_response (line 119) | def generate_ground_truth_response(prompt: str):
  function compute_reward (line 139) | def compute_reward(prompt: str, response: str, sequence_reward=1.0):

FILE: tests/special_e2e/envs/digit_completion/tokenizer.py
  class CharTokenizer (line 29) | class CharTokenizer(PreTrainedTokenizer):
    method __init__ (line 30) | def __init__(self, characters: Sequence[str], model_max_length: int, c...
    method vocab_size (line 83) | def vocab_size(self) -> int:
    method get_vocab (line 86) | def get_vocab(self):
    method _tokenize (line 89) | def _tokenize(self, text: str) -> list[str]:
    method _convert_token_to_id (line 92) | def _convert_token_to_id(self, token: str) -> int:
    method _convert_id_to_token (line 95) | def _convert_id_to_token(self, index: int) -> str:
    method convert_tokens_to_string (line 98) | def convert_tokens_to_string(self, tokens):
    method build_inputs_with_special_tokens (line 101) | def build_inputs_with_special_tokens(
    method get_special_tokens_mask (line 111) | def get_special_tokens_mask(
    method get_config (line 129) | def get_config(self) -> dict:
    method from_config (line 137) | def from_config(cls, config: dict):
    method save_pretrained (line 144) | def save_pretrained(self, save_directory: str | os.PathLike, **kwargs):
    method from_pretrained (line 151) | def from_pretrained(cls, save_directory: str | os.PathLike, **kwargs):

FILE: tests/special_e2e/sft/compare_sft_engine_results.py
  function get_result (line 21) | def get_result(file):
  function compare_results (line 31) | def compare_results(golden_results, other_result):

FILE: tests/special_sanity/check_api_docs.py
  function iter_submodules (line 57) | def iter_submodules(root: ModuleType) -> Iterable[ModuleType]:
  function names_missing_doc (line 72) | def names_missing_doc(mod: ModuleType) -> list[str]:
  function check_module (line 92) | def check_module(qualname: str) -> list[str]:
  function autodiscover_packages (line 106) | def autodiscover_packages() -> list[str]:
  function main (line 115) | def main() -> None:

FILE: tests/special_sanity/check_docs_time_info.py
  function is_allowed (line 41) | def is_allowed(path: Path) -> bool:
  function main (line 52) | def main():

FILE: tests/special_sanity/check_docstrings.py
  class DocstringChecker (line 25) | class DocstringChecker(ast.NodeVisitor):
    method __init__ (line 28) | def __init__(self, filename: str):
    method visit_FunctionDef (line 34) | def visit_FunctionDef(self, node: ast.FunctionDef):
    method visit_AsyncFunctionDef (line 45) | def visit_AsyncFunctionDef(self, node: ast.AsyncFunctionDef):
    method visit_ClassDef (line 56) | def visit_ClassDef(self, node: ast.ClassDef):
    method _has_docstring (line 67) | def _has_docstring(self, node) -> bool:
  function check_file_docstrings (line 72) | def check_file_docstrings(filepath: str) -> list[tuple[str, str, int]]:
  function main (line 88) | def main():

FILE: tests/special_sanity/check_license.py
  function get_py_files (line 49) | def get_py_files(path_arg: Path) -> Iterable[Path]:

FILE: tests/special_sanity/check_pr_description.py
  class TemplateFileError (line 24) | class TemplateFileError(Exception):
  class PRBodyLoadError (line 28) | class PRBodyLoadError(Exception):
  class PRDescriptionError (line 32) | class PRDescriptionError(Exception):
  function load_template (line 40) | def load_template(path):
  function load_pr_body (line 58) | def load_pr_body(event_path):
  function check_pr_description (line 67) | def check_pr_description(body, template_lines):
  function main (line 84) | def main():

FILE: tests/special_sanity/test_config_docs.py
  function validate_yaml_format (line 19) | def validate_yaml_format(yaml_lines):
  function test_trainer_config_doc (line 60) | def test_trainer_config_doc():

FILE: tests/special_sanity/test_import.py
  function test_import (line 16) | def test_import():
  function test_single_controller_import (line 22) | def test_single_controller_import():

FILE: tests/special_sanity/type_coverage_check.py
  function get_changed_files (line 27) | def get_changed_files() -> list[Path]:
  function get_changed_lines (line 34) | def get_changed_lines(file_path: Path) -> set[int]:
  function should_check_type (line 61) | def should_check_type(arg_name: str) -> bool:
  function has_type_annotations (line 69) | def has_type_annotations(node: ast.AST, debug: bool = False) -> int:
  function check_file (line 87) | def check_file(
  function main (line 116) | def main() -> None:

FILE: tests/special_sanity/validate_imported_docs.py
  function _parse_args (line 32) | def _parse_args() -> argparse.Namespace:
  function _import_attr (line 57) | def _import_attr(module_name: str, attr_name: str):
  function _check_file (line 63) | def _check_file(py_file: pathlib.Path, project_root: pathlib.Path, allow...
  function main (line 110) | def main() -> None:

FILE: tests/special_sanity/validate_structure.py
  function discover_allowed_modules (line 39) | def discover_allowed_modules(impl_root: Path, extra: list[str]) -> set[s...
  function find_violations (line 46) | def find_violations(tests_root: Path, allowed: set[str], allowed_files: ...
  function main (line 66) | def main() -> None:

FILE: tests/special_standalone/test_memory_buffers.py
  function test_memory_buffers (line 26) | def test_memory_buffers():

FILE: tests/test_base_config_on_cpu.py
  function base_config_mock (line 21) | def base_config_mock():
  function test_getitem_success (line 28) | def test_getitem_success(base_config_mock):
  function test_getitem_nonexistent_attribute (line 33) | def test_getitem_nonexistent_attribute(base_config_mock):
  function test_getitem_invalid_key_type (line 39) | def test_getitem_invalid_key_type(base_config_mock):

FILE: tests/test_protocol_on_cpu.py
  function test_union_tensor_dict (line 36) | def test_union_tensor_dict():
  function test_union_numpy_dict (line 51) | def test_union_numpy_dict():
  function test_tensor_dict_constructor (line 141) | def test_tensor_dict_constructor():
  function test_tensor_dict_make_iterator (line 155) | def test_tensor_dict_make_iterator():
  function test_reorder (line 184) | def test_reorder():
  function test_chunk_concat (line 195) | def test_chunk_concat():
  function test_concat_metrics_from_multiple_workers (line 219) | def test_concat_metrics_from_multiple_workers():
  function test_concat_with_empty_and_non_list_meta_info (line 249) | def test_concat_with_empty_and_non_list_meta_info():
  function test_concat_first_worker_missing_metrics (line 272) | def test_concat_first_worker_missing_metrics():
  function test_concat_non_list_metrics (line 295) | def test_concat_non_list_metrics():
  function test_concat_merge_different_non_metric_keys (line 315) | def test_concat_merge_different_non_metric_keys():
  function test_concat_conflicting_non_metric_keys (line 339) | def test_concat_conflicting_non_metric_keys():
  function test_pop (line 357) | def test_pop():
  function test_repeat (line 370) | def test_repeat():
  function test_dataproto_pad_unpad (line 395) | def test_dataproto_pad_unpad():
  function test_dataproto_fold_unfold (line 447) | def test_dataproto_fold_unfold():
  function test_torch_save_data_proto (line 470) | def test_torch_save_data_proto():
  function test_len (line 486) | def test_len():
  function test_dataproto_index (line 506) | def test_dataproto_index():
  function test_old_vs_new_from_single_dict (line 570) | def test_old_vs_new_from_single_dict():
  function test_dataproto_no_batch (line 607) | def test_dataproto_no_batch():
  function test_sample_level_repeat (line 617) | def test_sample_level_repeat():
  function test_dataproto_unfold_column_chunks (line 642) | def test_dataproto_unfold_column_chunks():
  function test_dataproto_chunk_after_index (line 708) | def test_dataproto_chunk_after_index():
  function test_to_tensordict (line 754) | def test_to_tensordict():
  function test_from_tensordict (line 768) | def test_from_tensordict():
  function test_to_tensordict_with_nested_lists (line 785) | def test_to_tensordict_with_nested_lists():
  function test_to_tensordict_with_nested_dicts (line 810) | def test_to_tensordict_with_nested_dicts():
  function test_to_tensordict_with_complex_nested_structures (line 834) | def test_to_tensordict_with_complex_nested_structures():
  function test_to_tensordict_and_back_with_nested_data (line 862) | def test_to_tensordict_and_back_with_nested_data():
  function test_to_tensordict_agent_loop_scenario (line 926) | def test_to_tensordict_agent_loop_scenario():
  function test_serialize_deserialize_single_tensor (line 993) | def test_serialize_deserialize_single_tensor():
  function test_serialize_deserialize_tensordict_regular_tensors (line 1010) | def test_serialize_deserialize_tensordict_regular_tensors():
  function test_serialize_deserialize_tensordict_nested_tensors (line 1039) | def test_serialize_deserialize_tensordict_nested_tensors():
  function test_serialize_deserialize_tensordict_mixed_types (line 1092) | def test_serialize_deserialize_tensordict_mixed_types():
  function test_serialize_deserialize_tensordict_with_device (line 1177) | def test_serialize_deserialize_tensordict_with_device():
  function test_serialize_dataproto_with_empty_tensordict (line 1208) | def test_serialize_dataproto_with_empty_tensordict():

FILE: tests/test_protocol_v2_on_cpu.py
  function test_union_tensor_dict (line 30) | def test_union_tensor_dict():
  function test_tensor_dict_constructor (line 67) | def test_tensor_dict_constructor():
  function test_index_select_tensor_dict (line 92) | def test_index_select_tensor_dict():
  function test_tensordict_with_images (line 131) | def test_tensordict_with_images():
  function test_tensordict_with_packing (line 159) | def test_tensordict_with_packing():
  function test_tensordict_eq (line 185) | def test_tensordict_eq():
  function test_tensor_dict_make_iterator (line 248) | def test_tensor_dict_make_iterator():
  function test_reorder (line 300) | def test_reorder():
  function test_chunk_concat (line 313) | def test_chunk_concat():
  function test_pop (line 350) | def test_pop():
  function test_get (line 382) | def test_get():
  function test_repeat (line 412) | def test_repeat():
  function test_dataproto_pad_unpad (line 437) | def test_dataproto_pad_unpad():
  function test_torch_save_data_proto (line 488) | def test_torch_save_data_proto():
  function test_len (line 506) | def test_len():
  function test_dataproto_index (line 523) | def test_dataproto_index():
  function test_select (line 583) | def test_select():
  function test_dataproto_no_batch (line 596) | def test_dataproto_no_batch():
  function test_sample_level_repeat (line 607) | def test_sample_level_repeat():
  function test_dataproto_chunk_after_index (line 633) | def test_dataproto_chunk_after_index():
  function test_concat_nested_tensor (line 676) | def test_concat_nested_tensor():
  function test_concat_tensordict (line 731) | def test_concat_tensordict():
  function test_chunk_tensordict (line 781) | def test_chunk_tensordict():
  function test_assign_non_tensor_stack_with_nested_lists (line 840) | def test_assign_non_tensor_stack_with_nested_lists():
  function test_assign_non_tensor_stack_with_nested_dicts (line 855) | def test_assign_non_tensor_stack_with_nested_dicts():
  function test_assign_non_tensor_stack_with_complex_nested (line 870) | def test_assign_non_tensor_stack_with_complex_nested():
  function test_assign_non_tensor_handles_wrappers (line 889) | def test_assign_non_tensor_handles_wrappers():
  function test_assign_non_tensor_stack_batch_size_check (line 904) | def test_assign_non_tensor_stack_batch_size_check():
  function test_assign_non_tensor_with_auto_detection (line 912) | def test_assign_non_tensor_with_auto_detection():
  function test_get_tensordict_with_nested_lists (line 935) | def test_get_tensordict_with_nested_lists():
  function test_get_tensordict_with_nested_dicts (line 950) | def test_get_tensordict_with_nested_dicts():
  function test_get_tensordict_with_complex_nested_structures (line 962) | def test_get_tensordict_with_complex_nested_structures():
  function test_get_tensordict_agent_loop_scenario (line 977) | def test_get_tensordict_agent_loop_scenario():
  function test_contiguous (line 1040) | def test_contiguous():

FILE: tests/trainer/config/test_algo_config_on_cpu.py
  class TestAlgoConfig (line 30) | class TestAlgoConfig(unittest.TestCase):
    method setUp (line 33) | def setUp(self):
    method test_dataclass_creation_from_dict (line 56) | def test_dataclass_creation_from_dict(self):
    method test_dataclass_creation_from_omega_config (line 69) | def test_dataclass_creation_from_omega_config(self):
    method test_nested_configs (line 77) | def test_nested_configs(self):
    method test_default_values (line 92) | def test_default_values(self):
    method test_get_method_backward_compatibility (line 105) | def test_get_method_backward_compatibility(self):
    method test_post_init_nested_configs (line 117) | def test_post_init_nested_configs(self):
    method test_config_init_from_yaml (line 127) | def test_config_init_from_yaml(self):
  class TestAlgoCompute (line 140) | class TestAlgoCompute(unittest.TestCase):
    method setUp (line 143) | def setUp(self):
    method test_advantage_estimator_with_cfg (line 157) | def test_advantage_estimator_with_cfg(self):
    method test_grpo_advantage_estimator_with_cfg (line 182) | def test_grpo_advantage_estimator_with_cfg(self):

FILE: tests/trainer/config/test_legacy_config_on_cpu.py
  class TestConfigComparison (line 35) | class TestConfigComparison(unittest.TestCase):
    method _compare_configs_recursively (line 55) | def _compare_configs_recursively(
    method test_ppo_trainer_config_matches_legacy (line 114) | def test_ppo_trainer_config_matches_legacy(self):
    method test_ppo_megatron_trainer_config_matches_legacy (line 138) | def test_ppo_megatron_trainer_config_matches_legacy(self):
    method test_load_component (line 160) | def test_load_component(self):

FILE: tests/trainer/ppo/test_core_algos_on_cpu.py
  function mock_test_fn (line 34) | def mock_test_fn():
  class TestRegisterAdvEst (line 38) | class TestRegisterAdvEst(unittest.TestCase):
    method setUp (line 39) | def setUp(self):
    method tearDown (line 48) | def tearDown(self) -> None:
    method test_register_new_function (line 52) | def test_register_new_function(self):
    method test_register_with_enum (line 62) | def test_register_with_enum(self):
    method test_duplicate_registration_same_function (line 76) | def test_duplicate_registration_same_function(self):
    method test_duplicate_registration_different_function (line 83) | def test_duplicate_registration_different_function(self):
    method test_decorator_preserves_function (line 96) | def test_decorator_preserves_function(self):
    method test_multiple_registrations (line 105) | def test_multiple_registrations(self):
    method test_get_adv_estimator_fn_valid_names (line 121) | def test_get_adv_estimator_fn_valid_names(self):
    method test_get_adv_estimator_fn_invalid_name (line 131) | def test_get_adv_estimator_fn_invalid_name(self):
    method test_get_adv_estimator_fn_case_sensitive (line 137) | def test_get_adv_estimator_fn_case_sensitive(self):
  function test_multi_turn_compute_gae_advantage_return (line 143) | def test_multi_turn_compute_gae_advantage_return():
  function _make_group_index (line 200) | def _make_group_index(batch_size: int, num_groups: int) -> np.ndarray:
  function _rand_mask (line 214) | def _rand_mask(batch_size: int, seq_len: int) -> torch.Tensor:
  function test_rloo_and_vectorized_equivalence (line 230) | def test_rloo_and_vectorized_equivalence(batch_size: int, seq_len: int, ...
  function test_grpo_and_vectorized_equivalence (line 270) | def test_grpo_and_vectorized_equivalence(batch_size: int, seq_len: int, ...

FILE: tests/trainer/ppo/test_metric_utils_on_cpu.py
  class TestReduceMetrics (line 41) | class TestReduceMetrics(unittest.TestCase):
    method test_reduce_metrics_basic (line 44) | def test_reduce_metrics_basic(self):
    method test_reduce_metrics_empty (line 55) | def test_reduce_metrics_empty(self):
    method test_reduce_metrics_single_value (line 64) | def test_reduce_metrics_single_value(self):
  class TestMetric (line 74) | class TestMetric(unittest.TestCase):
    method test_init_with_string_aggregation (line 77) | def test_init_with_string_aggregation(self):
    method test_init_with_enum_aggregation (line 83) | def test_init_with_enum_aggregation(self):
    method test_init_with_value (line 89) | def test_init_with_value(self):
    method test_init_with_invalid_aggregation (line 94) | def test_init_with_invalid_aggregation(self):
    method test_append_float (line 99) | def test_append_float(self):
    method test_append_int (line 106) | def test_append_int(self):
    method test_append_tensor (line 113) | def test_append_tensor(self):
    method test_append_non_scalar_tensor_raises (line 120) | def test_append_non_scalar_tensor_raises(self):
    method test_append_metric (line 126) | def test_append_metric(self):
    method test_extend_with_list (line 136) | def test_extend_with_list(self):
    method test_extend_with_metric (line 142) | def test_extend_with_metric(self):
    method test_extend_aggregation_mismatch_raises (line 153) | def test_extend_aggregation_mismatch_raises(self):
    method test_aggregate_mean (line 161) | def test_aggregate_mean(self):
    method test_aggregate_sum (line 167) | def test_aggregate_sum(self):
    method test_aggregate_min (line 173) | def test_aggregate_min(self):
    method test_aggregate_max (line 179) | def test_aggregate_max(self):
    method test_aggregate_dp_sum_mean (line 185) | def test_aggregate_dp_sum_mean(self):
    method test_aggregate_dp_min_max (line 215) | def test_aggregate_dp_min_max(self):
    method test_aggregate_dp_mismatched_lengths (line 245) | def test_aggregate_dp_mismatched_lengths(self):
    method test_from_dict (line 256) | def test_from_dict(self):
    method test_init_list (line 267) | def test_init_list(self):
    method test_reduce_metrics_with_metric (line 277) | def test_reduce_metrics_with_metric(self):
  class TestComputeDataMetrics (line 292) | class TestComputeDataMetrics(unittest.TestCase):
    method setUp (line 295) | def setUp(self):
    method test_compute_data_metrics_with_critic (line 320) | def test_compute_data_metrics_with_critic(self):
    method test_compute_data_metrics_without_critic (line 338) | def test_compute_data_metrics_without_critic(self):
  class TestComputeTimingMetrics (line 352) | class TestComputeTimingMetrics(unittest.TestCase):
    method setUp (line 355) | def setUp(self):
    method test_compute_timing_metrics (line 377) | def test_compute_timing_metrics(self, mock_compute_response_info):
  class TestComputeThroughputMetrics (line 403) | class TestComputeThroughputMetrics(unittest.TestCase):
    method setUp (line 406) | def setUp(self):
    method test_compute_throughout_metrics (line 414) | def test_compute_throughout_metrics(self):
  class TestBootstrapMetric (line 435) | class TestBootstrapMetric(unittest.TestCase):
    method test_bootstrap_metric_basic (line 438) | def test_bootstrap_metric_basic(self):
    method test_bootstrap_metric_empty (line 462) | def test_bootstrap_metric_empty(self):
  class TestCalcMajVal (line 468) | class TestCalcMajVal(unittest.TestCase):
    method test_calc_maj_val_basic (line 471) | def test_calc_maj_val_basic(self):
    method test_calc_maj_val_tie (line 484) | def test_calc_maj_val_tie(self):
  class TestProcessValidationMetrics (line 501) | class TestProcessValidationMetrics(unittest.TestCase):
    method test_process_validation_metrics_basic (line 504) | def test_process_validation_metrics_basic(self):
    method test_process_validation_metrics_with_pred (line 527) | def test_process_validation_metrics_with_pred(self):

FILE: tests/trainer/ppo/test_rollout_corr.py
  function test_basic_rollout_correction (line 41) | def test_basic_rollout_correction():
  function test_each_supported_rollout_rs_option (line 143) | def test_each_supported_rollout_rs_option(option: str, threshold):
  function test_rollout_rs_multiple_options (line 168) | def test_rollout_rs_multiple_options():
  function test_metrics_completeness (line 194) | def test_metrics_completeness():
  function test_offpolicy_metrics (line 254) | def test_offpolicy_metrics():
  function test_mask_mode (line 311) | def test_mask_mode():

FILE: tests/trainer/ppo/test_rollout_corr_integration.py
  class TestRolloutISIntegration (line 28) | class TestRolloutISIntegration:
    method sample_data (line 32) | def sample_data(self):
    method config_with_rollout_is (line 46) | def config_with_rollout_is(self):
    method test_policy_loss_with_rollout_is (line 60) | def test_policy_loss_with_rollout_is(self, sample_data, config_with_ro...
    method test_rollout_is_weights_computation (line 96) | def test_rollout_is_weights_computation(self, sample_data):
    method test_all_aggregation_levels (line 120) | def test_all_aggregation_levels(self, sample_data):
    method test_both_bounding_modes (line 146) | def test_both_bounding_modes(self, sample_data):
    method test_offpolicy_metrics (line 172) | def test_offpolicy_metrics(self, sample_data):
    method test_metrics_only_mode (line 186) | def test_metrics_only_mode(self, sample_data, config_with_rollout_is):
  class TestRolloutCorrectionConfigNormalization (line 234) | class TestRolloutCorrectionConfigNormalization:
    method test_alias_normalization_and_threshold_parsing (line 237) | def test_alias_normalization_and_threshold_parsing(self):
    method test_missing_threshold_raises (line 250) | def test_missing_threshold_raises(self):
    method test_float_threshold_conversion_in_factory (line 255) | def test_float_threshold_conversion_in_factory(self):

FILE: tests/utils/_test_module.py
  class TestClass (line 17) | class TestClass:
    method __init__ (line 20) | def __init__(self, value=None):
    method get_value (line 23) | def get_value(self):
  function test_function (line 30) | def test_function():

FILE: tests/utils/ckpt/test_checkpoint_cleanup_on_cpu.py
  class TestCheckpointCleanupLogic (line 22) | class TestCheckpointCleanupLogic:
    method setup (line 26) | def setup(self):
    method manager (line 33) | def manager(self, monkeypatch):
    method _create_checkpoint_dir (line 56) | def _create_checkpoint_dir(self, step: int) -> str:
    method test_max_ckpt_1_preserves_existing_before_save (line 64) | def test_max_ckpt_1_preserves_existing_before_save(self, manager):
    method test_max_ckpt_1_deletes_old_after_save (line 76) | def test_max_ckpt_1_deletes_old_after_save(self, manager):
    method test_max_ckpt_2_keeps_one_before_save (line 88) | def test_max_ckpt_2_keeps_one_before_save(self, manager):
    method test_max_ckpt_0_keeps_all (line 100) | def test_max_ckpt_0_keeps_all(self, manager):
    method test_full_save_cycle_max_ckpt_1 (line 115) | def test_full_save_cycle_max_ckpt_1(self, manager):

FILE: tests/utils/ckpt/test_esi_save_ckpt_on_cpu.py
  class TestShouldSaveCkptEsi (line 22) | class TestShouldSaveCkptEsi(TestCase):
    method test_no_expiration_timestamp (line 23) | def test_no_expiration_timestamp(self):
    method test_mlp_expiration_valid (line 29) | def test_mlp_expiration_valid(self):
    method test_mlp_expiration_passed (line 35) | def test_mlp_expiration_passed(self):
    method test_mlp_invalid_timestamp (line 41) | def test_mlp_invalid_timestamp(self):
    method test_mlp_expiration_not_reached (line 46) | def test_mlp_expiration_not_reached(self):
    method test_aws_expiration_not_reached (line 52) | def test_aws_expiration_not_reached(self):
    method test_redundant_time (line 59) | def test_redundant_time(self):
    method test_zero_max_steps_duration (line 66) | def test_zero_max_steps_duration(self):

FILE: tests/utils/dataset/test_create_rl_sampler_on_cpu.py
  class RandomCurriculumSampler (line 29) | class RandomCurriculumSampler(AbstractCurriculumSampler):
    method __init__ (line 30) | def __init__(
    method __iter__ (line 40) | def __iter__(self):
    method __len__ (line 43) | def __len__(self) -> int:
    method update (line 46) | def update(self, batch) -> None:
  class MockIncorrectSampler (line 50) | class MockIncorrectSampler:
    method __init__ (line 53) | def __init__(self, data_source, data_config):
  class MockChatDataset (line 57) | class MockChatDataset(Dataset):
    method __init__ (line 58) | def __init__(self):
    method __getitem__ (line 70) | def __getitem__(self, index):
    method __len__ (line 73) | def __len__(self):
  function test_create_custom_curriculum_samper (line 77) | def test_create_custom_curriculum_samper():
  function test_create_custom_curriculum_samper_wrong_class (line 94) | def test_create_custom_curriculum_samper_wrong_class():

FILE: tests/utils/dataset/test_multiturn_sft_dataset_on_cpu.py
  function test_multiturn_sft_dataset (line 47) | def test_multiturn_sft_dataset(model_path: str, ignore_input_ids_mismatc...
  function generate_image (line 239) | def generate_image(description: str, size: str = "256x256"):
  function vlm_data_file (line 253) | def vlm_data_file():
  function test_multiturn_sft_vlm_dataset_on_cpu (line 349) | def test_multiturn_sft_vlm_dataset_on_cpu(model_path, vlm_data_file):
  function test_multiturn_sft_vlm_dataloader_on_cpu (line 411) | def test_multiturn_sft_vlm_dataloader_on_cpu(model_path, vlm_data_file):

FILE: tests/utils/dataset/test_rl_collate_fn_on_cpu.py
  function test_rl_collate_fn (line 17) | def test_rl_collate_fn():

FILE: tests/utils/dataset/test_rl_dataset_on_cpu.py
  function get_gsm8k_data (line 28) | def get_gsm8k_data():
  function test_rl_dataset (line 36) | def test_rl_dataset():
  function test_rl_dataset_with_max_samples (line 67) | def test_rl_dataset_with_max_samples():
  function test_image_rl_data (line 83) | def test_image_rl_data():
  function video_data_file (line 131) | def video_data_file():
  function test_video_rl_data (line 166) | def test_video_rl_data(video_data_file):

FILE: tests/utils/debug/test_metrics.py
  class TestMetrics (line 22) | class TestMetrics(unittest.TestCase):
    method test_calculate_debug_metrics (line 23) | def test_calculate_debug_metrics(self):

FILE: tests/utils/megatron/test_pipeline_parallel.py
  function test_make_batch_generator_no_vpp (line 21) | def test_make_batch_generator_no_vpp():
  function test_make_batch_generator_with_vpp (line 28) | def test_make_batch_generator_with_vpp():
  function test_make_batch_generator_empty (line 40) | def test_make_batch_generator_empty():
  function test_get_dynamic_pipeline_shards (line 63) | def test_get_dynamic_pipeline_shards(layer_num, pp_size, gt):

FILE: tests/utils/reward_score/reward_score/test_sandbox_fusion_on_cpu.py
  function test_integration_success_correct (line 78) | def test_integration_success_correct():
  function test_integration_success_wrong_output (line 89) | def test_integration_success_wrong_output():
  function test_integration_compile_error (line 99) | def test_integration_compile_error():
  function test_integration_runtime_error (line 108) | def test_integration_runtime_error():
  function test_integration_runtime_timeout (line 117) | def test_integration_runtime_timeout():
  function test_integration_concurrency_high_load (line 127) | def test_integration_concurrency_high_load():
  function test_unit_concurrency_order (line 254) | def test_unit_concurrency_order(mock_call_sandbox_api):
  function test_unit_api_timeout_error_concurrent (line 298) | def test_unit_api_timeout_error_concurrent(mock_call_sandbox_api):
  function _mock_api_call_for_concurrency_tracking (line 351) | def _mock_api_call_for_concurrency_tracking(
  function _process_pool_worker_for_concurrency_test (line 391) | def _process_pool_worker_for_concurrency_test(
  function test_multiprocess_global_concurrency_limit_with_semaphore (line 458) | def test_multiprocess_global_concurrency_limit_with_semaphore():
  function test_unit_invalid_input_format (line 556) | def test_unit_invalid_input_format():
  function test_unit_input_output_mismatch (line 572) | def test_unit_input_output_mismatch():
  function test_integration_concurrency_all_timeout (line 581) | def test_integration_concurrency_all_timeout():
  function test_fn_name_success_single_case (line 633) | def test_fn_name_success_single_case():
  function test_none_and_empty_stdin_passed_correctly (line 672) | def test_none_and_empty_stdin_passed_correctly():
  function test_assert_case_success (line 696) | def test_assert_case_success():

FILE: tests/utils/reward_score/test_sandbox_on_cpu.py
  function test_parallelism (line 95) | def test_parallelism():
  function test_prime_code (line 118) | def test_prime_code():
  function test_prime_code_sandbox_fusion (line 130) | def test_prime_code_sandbox_fusion():
  function test_continuous_score_consistency (line 147) | def test_continuous_score_consistency():
  function test_check_correctness (line 176) | def test_check_correctness():
  function test_prime_math (line 186) | def test_prime_math():

FILE: tests/utils/test_activation_offload.py
  function create_random_input_ids (line 33) | def create_random_input_ids(batch_size, seq_len, vocab_size):
  function _fsdp_activation_offloading_test (line 52) | def _fsdp_activation_offloading_test(rank, world_size, rendezvous_file, ...
  function test_activation_offloading (line 166) | def test_activation_offloading(world_size, strategy, tmp_path):

FILE: tests/utils/test_bucketed_weight_transfer.py
  function _unique_zmq_handle (line 37) | def _unique_zmq_handle():
  function _generate_weights (line 41) | def _generate_weights(weight_specs, seed):
  function _sender_fn (line 64) | def _sender_fn(zmq_handle, weight_specs, seed, bucket_size_mb, use_shm):
  function _receiver_fn (line 77) | def _receiver_fn(zmq_handle, use_shm, result_queue):
  function _transfer_and_validate (line 98) | def _transfer_and_validate(weight_specs, bucket_size_mb, use_shm):
  class TestBucketedWeightTransferSHM (line 149) | class TestBucketedWeightTransferSHM:
    method test_single_small_weight (line 152) | def test_single_small_weight(self):
    method test_multiple_weights_single_bucket (line 156) | def test_multiple_weights_single_bucket(self):
    method test_multiple_buckets (line 164) | def test_multiple_buckets(self):
    method test_mixed_dtypes (line 169) | def test_mixed_dtypes(self):
    method test_empty_weights (line 177) | def test_empty_weights(self):
  class TestBucketedWeightTransferIPC (line 185) | class TestBucketedWeightTransferIPC:
    method test_single_small_weight (line 188) | def test_single_small_weight(self):
    method test_multiple_weights_single_bucket (line 192) | def test_multiple_weights_single_bucket(self):
    method test_multiple_buckets (line 200) | def test_multiple_buckets(self):
    method test_mixed_dtypes (line 204) | def test_mixed_dtypes(self):
    method test_empty_weights (line 212) | def test_empty_weights(self):
    method test_exact_bucket_boundary (line 215) | def test_exact_bucket_boundary(self):

FILE: tests/utils/test_check_ipc_version_support_on_npu.py
  class TestCheckIPCVersionSupport (line 22) | class TestCheckIPCVersionSupport(unittest.TestCase):
    method setUp (line 25) | def setUp(self):
    method tearDown (line 30) | def tearDown(self):
    method test_standard_version_with_support (line 34) | def test_standard_version_with_support(self):
    method test_standard_version_newer (line 40) | def test_standard_version_newer(self):
    method test_rc_version_format (line 46) | def test_rc_version_format(self):
    method test_exact_rc_version (line 53) | def test_exact_rc_version(self):
    method test_t_suffix_version (line 60) | def test_t_suffix_version(self):
    method test_t_suffix_version_older (line 67) | def test_t_suffix_version_older(self):
    method test_software_version_below_minimum (line 74) | def test_software_version_below_minimum(self):
    method test_cann_version_below_minimum (line 80) | def test_cann_version_below_minimum(self):
    method test_both_versions_below_minimum (line 87) | def test_both_versions_below_minimum(self):
    method test_invalid_software_version (line 94) | def test_invalid_software_version(self):
    method test_invalid_cann_version (line 100) | def test_invalid_cann_version(self):
    method test_rc_with_more_parts (line 106) | def test_rc_with_more_parts(self):
    method test_standard_with_more_parts (line 112) | def test_standard_with_more_parts(self):
    method test_rc_edge_case_versions (line 118) | def test_rc_edge_case_versions(self):
    method test_major_version_differences (line 128) | def test_major_version_differences(self):
  class TestGetNPUVersions (line 139) | class TestGetNPUVersions(unittest.TestCase):
    method test_get_npu_versions_success (line 146) | def test_get_npu_versions_success(self, mock_file, mock_exists, mock_m...
    method test_get_npu_versions_missing_software_version (line 163) | def test_get_npu_versions_missing_software_version(self, mock_run):
    method test_get_npu_versions_unsupported_architecture (line 176) | def test_get_npu_versions_unsupported_architecture(self, mock_file, mo...
    method test_get_npu_versions_cann_path_not_exists (line 192) | def test_get_npu_versions_cann_path_not_exists(self, mock_file, mock_e...
    method test_get_npu_versions_info_file_not_exists (line 208) | def test_get_npu_versions_info_file_not_exists(self, mock_file, mock_e...
    method test_get_npu_versions_missing_cann_version (line 226) | def test_get_npu_versions_missing_cann_version(self, mock_file, mock_e...

FILE: tests/utils/test_check_profiler_output.py
  class DeviceCheckConfig (line 29) | class DeviceCheckConfig:
  class ProfilerChecker (line 40) | class ProfilerChecker:
    method __init__ (line 45) | def __init__(self, device_type: str, profiler_dir: str):
    method _init_device_config (line 56) | def _init_device_config(self):
    method _validate_stage_dirs (line 80) | def _validate_stage_dirs(self, stage: str) -> bool:
    method check (line 104) | def check(self) -> bool:
  function parse_args (line 122) | def parse_args():
  function main (line 141) | def main():

FILE: tests/utils/test_config_on_cpu.py
  class TestDataclass (line 25) | class TestDataclass(BaseConfig):
  class TestTrainConfig (line 31) | class TestTrainConfig(BaseConfig):
  class TestConfigOnCPU (line 46) | class TestConfigOnCPU(unittest.TestCase):
    method setUp (line 55) | def setUp(self):
    method test_omega_conf_to_dataclass (line 58) | def test_omega_conf_to_dataclass(self):
    method test_nested_omega_conf_to_dataclass (line 65) | def test_nested_omega_conf_to_dataclass(self):
  class TestPrintCfgCommand (line 74) | class TestPrintCfgCommand(unittest.TestCase):
    method test_command_with_override (line 77) | def test_command_with_override(self):

FILE: tests/utils/test_flops_counter.py
  class Config (line 24) | class Config:
    method __init__ (line 25) | def __init__(self, config_dict):
  function test_flops_counter (line 454) | def test_flops_counter(config_type: str):

FILE: tests/utils/test_fs_on_cpu.py
  function test_record_and_check_directory_structure (line 21) | def test_record_and_check_directory_structure(tmp_path):
  function test_copy_from_hdfs_with_mocks (line 43) | def test_copy_from_hdfs_with_mocks(tmp_path, monkeypatch):
  function test_always_recopy_flag (line 66) | def test_always_recopy_flag(tmp_path, monkeypatch):

FILE: tests/utils/test_fsdp2_peft_wrapping.py
  class MockDecoderLayer (line 30) | class MockDecoderLayer(nn.Module):
    method __init__ (line 33) | def __init__(self, hidden_size=64):
  class MockModulesToSaveWrapper (line 39) | class MockModulesToSaveWrapper(nn.Module):
    method __init__ (line 46) | def __init__(self, original_module):
  class MockCausalLM (line 52) | class MockCausalLM(nn.Module):
    method __init__ (line 57) | def __init__(self, vocab_size=1000, hidden_size=64, num_layers=2, tie_...
  class TestFSDP2PeftWrapping (line 69) | class TestFSDP2PeftWrapping(unittest.TestCase):
    method _get_wrapped_names (line 72) | def _get_wrapped_names(self, model, cls_names):
    method test_vanilla_model_wraps_layers_and_embedding (line 79) | def test_vanilla_model_wraps_layers_and_embedding(self):
    method test_peft_wrapped_model_wraps_embed_tokens_by_name (line 89) | def test_peft_wrapped_model_wraps_embed_tokens_by_name(self):
    method test_tied_embeddings_skips_name_based_wrapping (line 101) | def test_tied_embeddings_skips_name_based_wrapping(self):
    method test_peft_wrapped_tied_embeddings_skips_wrapping (line 110) | def test_peft_wrapped_tied_embeddings_skips_wrapping(self):
    method test_no_duplicate_wrapping_for_vanilla_embedding (line 121) | def test_no_duplicate_wrapping_for_vanilla_embedding(self):

FILE: tests/utils/test_fsdp_lora_merge.py
  function _test_merged_lora_context_worker (line 36) | def _test_merged_lora_context_worker(
  function test_merged_lora_context_qwen2 (line 161) | def test_merged_lora_context_qwen2(world_size, strategy, backup_adapters...
  function test_merged_lora_context_gptoss (line 190) | def test_merged_lora_context_gptoss(world_size, strategy, backup_adapter...

FILE: tests/utils/test_groupwise.py
  function test_as_torch_index_basic_integers (line 27) | def test_as_torch_index_basic_integers():
  function test_as_torch_index_near_integer_floats (line 36) | def test_as_torch_index_near_integer_floats():
  function test_as_torch_index_factorization_mixed (line 43) | def test_as_torch_index_factorization_mixed():
  function test_group_mean_std_simple (line 51) | def test_group_mean_std_simple():
  function test_group_mean_std_empty (line 68) | def test_group_mean_std_empty():
  function test_group_mean_std_default_device_no_force_env (line 75) | def test_group_mean_std_default_device_no_force_env(monkeypatch):

FILE: tests/utils/test_import_utils_on_cpu.py
  function test_load_extern_object_class (line 25) | def test_load_extern_object_class():
  function test_load_extern_object_function (line 42) | def test_load_extern_object_function():
  function test_load_extern_object_constant (line 55) | def test_load_extern_object_constant():
  function test_load_extern_object_nonexistent_file (line 64) | def test_load_extern_object_nonexistent_file():
  function test_load_extern_object_nonexistent_type (line 70) | def test_load_extern_object_nonexistent_type():
  function test_load_extern_object_none_path (line 76) | def test_load_extern_object_none_path():
  function test_load_extern_object_invalid_module (line 82) | def test_load_extern_object_invalid_module():

FILE: tests/utils/test_linear_cross_entropy.py
  function run_torch_entropy (line 49) | def run_torch_entropy(
  function run_verl_original_entropy (line 65) | def run_verl_original_entropy(
  function run_verl_torch_fused_entropy (line 83) | def run_verl_torch_fused_entropy(
  class TestLinearCrossEntropy (line 100) | class TestLinearCrossEntropy:
    method __init__ (line 101) | def __init__(self, test_case_idx: int, temperature: float = 1.5) -> None:
    method cleanup (line 105) | def cleanup(self):
    method generate_hyper (line 113) | def generate_hyper(self):
    method generate_forward_inputs (line 146) | def generate_forward_inputs(self):
    method generate_backward_inputs (line 160) | def generate_backward_inputs(self):
    method verify_correctness (line 165) | def verify_correctness(self, iterations=5):
    method check_storage (line 323) | def check_storage(self, method_name, run_forward):
    method check_storage_all (line 345) | def check_storage_all(self):
  function test_lce_non_divisible_vocab_padding (line 352) | def test_lce_non_divisible_vocab_padding():

FILE: tests/utils/test_mlflow_key_sanitization.py
  class TestMlflowLoggingAdapter (line 21) | class TestMlflowLoggingAdapter(unittest.TestCase):
    method test_sanitize_key_and_warning (line 22) | def test_sanitize_key_and_warning(self):

FILE: tests/utils/test_model_on_cpu.py
  function test_update_model_config (line 30) | def test_update_model_config(override_kwargs):

FILE: tests/utils/test_normalize_peft_param_name.py
  function _test_normalize_peft_with_fsdp_worker (line 37) | def _test_normalize_peft_with_fsdp_worker(rank, world_size, rendezvous_f...
  function test_normalize_peft_param_name_with_fsdp (line 200) | def test_normalize_peft_param_name_with_fsdp(world_size, strategy, tmp_p...

FILE: tests/utils/test_normalize_peft_param_name_on_cpu.py
  function create_base_model (line 23) | def create_base_model():
  function create_peft_model (line 36) | def create_peft_model():
  function base_model (line 46) | def base_model():
  function peft_model (line 52) | def peft_model():
  function test_normalize_peft_param_name_keys_match_base_model (line 57) | def test_normalize_peft_param_name_keys_match_base_model():
  function test_normalize_peft_param_name_removes_lora_keys (line 86) | def test_normalize_peft_param_name_removes_lora_keys(peft_model):
  function test_normalize_peft_param_name_removes_base_model_prefix (line 102) | def test_normalize_peft_param_name_removes_base_model_prefix(peft_model):
  function test_normalize_peft_param_name_removes_base_layer_suffix (line 118) | def test_normalize_peft_param_name_removes_base_layer_suffix(peft_model):
  function test_normalize_peft_param_name_tensor_shapes_match (line 134) | def test_normalize_peft_param_name_tensor_shapes_match(base_model, peft_...
  function test_normalize_peft_param_name_empty_dict (line 150) | def test_normalize_peft_param_name_empty_dict():
  function test_normalize_peft_param_name_filters_lora_patterns (line 165) | def test_normalize_peft_param_name_filters_lora_patterns(lora_key_pattern):

FILE: tests/utils/test_nvtx_profile.py
  class TestProfilerConfig (line 24) | class TestProfilerConfig(unittest.TestCase):
    method test_config_init (line 25) | def test_config_init(self):
    method test_frozen_config (line 51) | def test_frozen_config(self):
  class TestNsightSystemsProfiler (line 73) | class TestNsightSystemsProfiler(unittest.TestCase):
    method setUp (line 84) | def setUp(self):
    method test_initialization (line 89) | def test_initialization(self):
    method test_start_stop_profiling (line 93) | def test_start_stop_profiling(self):
    method test_annotate_decorator (line 118) | def test_annotate_decorator(self):

FILE: tests/utils/test_padding_on_cpu.py
  function test_padding_conversion_with_log_probs (line 21) | def test_padding_conversion_with_log_probs():
  function test_padding_conversion_without_log_probs (line 99) | def test_padding_conversion_without_log_probs():
  function test_padding_roundtrip (line 130) | def test_padding_roundtrip():
  function test_no_padding_2_padding_varying_lengths (line 178) | def test_no_padding_2_padding_varying_lengths():

FILE: tests/utils/test_prepare_micro_batches_with_group_size.py
  function _make_batch (line 34) | def _make_batch(seq_lens: list[int], force_group_size: int, max_token_le...
  function _verify_group_integrity (line 70) | def _verify_group_integrity(batch_idx_list: list[list[int]], force_group...
  function test_force_group_size_2_basic (line 99) | def test_force_group_size_2_basic():
  function test_force_group_size_4_basic (line 116) | def test_force_group_size_4_basic():
  function test_force_group_size_reconstruction (line 150) | def test_force_group_size_reconstruction():
  function test_force_group_size_single_micro_batch (line 174) | def test_force_group_size_single_micro_batch():
  function test_force_group_size_large_group (line 191) | def test_force_group_size_large_group():
  function test_force_group_size_1_unchanged (line 227) | def test_force_group_size_1_unchanged():

FILE: tests/utils/test_rollout_skip_on_cpu.py
  function temp_dir (line 28) | def temp_dir():
  function build_generate_fn (line 36) | def build_generate_fn(gen_bs, n):
  function mock_rollout_wg (line 56) | def mock_rollout_wg(request):
  class TestRolloutSkip (line 74) | class TestRolloutSkip:
    method test_initialization (line 75) | def test_initialization(self, capsys):
    method test_generate_without_wrap (line 95) | def test_generate_without_wrap(self, mock_rollout_wg):
    method test_dump (line 110) | def test_dump(self, mock_rollout_wg, capsys):
    method test_generate_with_wrap (line 125) | def test_generate_with_wrap(self, mock_rollout_wg, capsys):

FILE: tests/utils/test_rollout_trace_on_cpu.py
  function reset_rollout_trace_config_singleton (line 25) | def reset_rollout_trace_config_singleton():
  function mock_weave_client (line 31) | def mock_weave_client():
  class TracedClass (line 46) | class TracedClass:
    method my_method (line 50) | async def my_method(self, a, b="default"):
    method middle_method (line 56) | async def middle_method(self, a, b="default"):
    method my_method_with_exception (line 62) | async def my_method_with_exception(self):
    method upper_method (line 65) | async def upper_method(self):
  class UntracedClass (line 71) | class UntracedClass:
    method my_method (line 73) | async def my_method(self, x):
  function test_rollout_trace_on_untraced_class (line 77) | async def test_rollout_trace_on_untraced_class():
  function test_rollout_trace_with_tracer (line 83) | async def test_rollout_trace_with_tracer(mock_weave_client):
  function test_rollout_trace_with_exception (line 102) | async def test_rollout_trace_with_exception(mock_weave_client):
  function test_rollout_trace_with_dummy_backend (line 121) | async def test_rollout_trace_with_dummy_backend(mock_weave_client):
  function test_trace_disabled_with_trace_false (line 131) | async def test_trace_disabled_with_trace_false(mock_weave_client):
  function test_trace_false_disables_nested_trace_ops (line 157) | async def test_trace_false_disables_nested_trace_ops(mock_weave_client):
  function test_trace_enabled_restored_after_exception (line 182) | async def test_trace_enabled_restored_after_exception(mock_weave_client):
  function test_rollout_trace_with_real_weave_backend (line 211) | async def test_rollout_trace_with_real_weave_backend():
  function test_rollout_trace_with_real_mlflow_backend (line 232) | async def test_rollout_trace_with_real_mlflow_backend():

FILE: tests/utils/test_seqlen_balancing.py
  function test_seqlen_balancing (line 31) | def test_seqlen_balancing():
  function test_dynamic_batch (line 50) | def test_dynamic_batch():
  function _worker (line 64) | def _worker(rank, world_size, init_method, max_token_len, use_same_dp, m...
  function test_dataproto_split_uneven (line 128) | def test_dataproto_split_uneven():
  function test_seqlen_balancing_distributed_params (line 182) | def test_seqlen_balancing_distributed_params(tmp_path):
  function test_group_balanced_partitions (line 205) | def test_group_balanced_partitions():
  function test_group_balanced_partitions_single_sample_groups (line 237) | def test_group_balanced_partitions_single_sample_groups():
  function test_group_balanced_partitions_equal_size (line 254) | def test_group_balanced_partitions_equal_size():

FILE: tests/utils/test_server_profiler.py
  class TestServerProfilerArgs (line 28) | class TestServerProfilerArgs(unittest.TestCase):
    method test_build_vllm_profiler_args (line 29) | def test_build_vllm_profiler_args(self):
    method test_build_sglang_profiler_args (line 52) | def test_build_sglang_profiler_args(self):
  class TestServerProfilerFunctionality (line 63) | class TestServerProfilerFunctionality(unittest.IsolatedAsyncioTestCase):
    method test_vllm_start_stop_profile (line 64) | async def test_vllm_start_stop_profile(self):
    method test_sglang_start_stop_profile (line 93) | async def test_sglang_start_stop_profile(self):

FILE: tests/utils/test_shared_memory.py
  class TestSharedMemory (line 24) | class TestSharedMemory(unittest.TestCase):
    method setUp (line 27) | def setUp(self):
    method tearDown (line 35) | def tearDown(self):
    method test_create_shared_memory_new (line 41) | def test_create_shared_memory_new(self):
    method test_create_shared_memory_attach_existing (line 56) | def test_create_shared_memory_attach_existing(self):
    method test_rebuild_shared_memory_default_dtype (line 78) | def test_rebuild_shared_memory_default_dtype(self):
    method test_rebuild_shared_memory_custom_dtype (line 101) | def test_rebuild_shared_memory_custom_dtype(self):
    method test_shared_memory_data_integrity (line 124) | def test_shared_memory_data_integrity(self):
    method test_shared_memory_different_dtypes (line 145) | def test_shared_memory_different_dtypes(self):
    method test_shared_memory_multiple_operations (line 176) | def test_shared_memory_multiple_operations(self):
  function child_process_function (line 200) | def child_process_function(name, size, test_data_bytes):
  class TestSharedMemoryIntegration (line 230) | class TestSharedMemoryIntegration(unittest.TestCase):
    method test_cross_process_shared_memory (line 233) | def test_cross_process_shared_memory(self):

FILE: tests/utils/test_special_linear_cross_entropy_tp.py
  function run_torch_entropy (line 57) | def run_torch_entropy(
  class TorchEntropyTP (line 79) | class TorchEntropyTP(torch.autograd.Function):
    method forward (line 86) | def forward(
    method backward (line 128) | def backward(ctx, g_logprobs: torch.Tensor, g_entropy: torch.Tensor):
  class TestLinearCrossEntropy_TensorParallel (line 181) | class TestLinearCrossEntropy_TensorParallel:
    method __init__ (line 182) | def __init__(self):
    method initialize (line 192) | def initialize(self, test_case_idx: int, temperature: float = 1.5):
    method shutdown (line 196) | def shutdown(self):
    method cleanup (line 199) | def cleanup(self):
    method generate_hyper (line 207) | def generate_hyper(self):
    method generate_forward_inputs (line 242) | def generate_forward_inputs(self):
    method generate_backward_inputs (line 256) | def generate_backward_inputs(self):
    method verify_torch_itself (line 261) | def verify_torch_itself(self, iterations: int = 5):
    method check_torch_storage (line 331) | def check_torch_storage(self):
    method verify_kernel_correctness (line 364) | def verify_kernel_correctness(self, iterations: int = 5):
    method check_kernel_storage (line 455) | def check_kernel_storage(self):

FILE: tests/utils/test_special_mstx_profile.py
  class TestNPUProfilerInitialization (line 23) | class TestNPUProfilerInitialization(unittest.TestCase):
    method setUp (line 24) | def setUp(self):
    method test_init_with_default_config (line 27) | def test_init_with_default_config(self):
    method test_init_with_disabled_config (line 33) | def test_init_with_disabled_config(self):
    method test_init_with_all_ranks_true (line 39) | def test_init_with_all_ranks_true(self):
    method test_init_with_ranks_list (line 45) | def test_init_with_ranks_list(self):
    method test_init_with_rank_not_in_ranks (line 51) | def test_init_with_rank_not_in_ranks(self):
  class TestNPUProfilerStart (line 58) | class TestNPUProfilerStart(unittest.TestCase):
    method setUp (line 59) | def setUp(self):
    method test_start_when_enabled_and_this_rank (line 65) | def test_start_when_enabled_and_this_rank(self, mock_get_profiler):
    method test_start_when_not_this_rank (line 73) | def test_start_when_not_this_rank(self, mock_get_profiler):
    method test_start_discrete_mode_does_not_increase_count (line 81) | def test_start_discrete_mode_does_not_increase_count(self, mock_get_pr...
    method test_multiple_start_calls_do_not_increase_count (line 89) | def test_multiple_start_calls_do_not_increase_count(self, mock_get_pro...
  class TestNPUProfilerStartStopInteraction (line 97) | class TestNPUProfilerStartStopInteraction(unittest.TestCase):
    method setUp (line 98) | def setUp(self):
    method test_start_stop_cycle (line 104) | def test_start_stop_cycle(self, mock_get_profiler):
    method test_multiple_instances_share_define_count (line 118) | def test_multiple_instances_share_define_count(self, mock_get_profiler):
  class TestNPUProfilerAnnotate (line 132) | class TestNPUProfilerAnnotate(unittest.TestCase):
    method setUp (line 133) | def setUp(self):
    method test_annotate_decorator_applied_correctly (line 138) | def test_annotate_decorator_applied_correctly(self):
    method test_annotate_when_profiler_disabled (line 166) | def test_annotate_when_profiler_disabled(self):
    method test_annotate_when_this_step_disabled (line 189) | def test_annotate_when_this_step_disabled(self):
    method test_annotate_discrete_mode_enabled (line 212) | def test_annotate_discrete_mode_enabled(self):
    method test_annotate_with_default_message (line 250) | def test_annotate_with_default_message(self):

FILE: tests/utils/test_temp_env_on_cpu.py
  function clean_env (line 23) | def clean_env():
  function test_set_new_env_var (line 42) | def test_set_new_env_var():
  function test_restore_existing_env_var (line 56) | def test_restore_existing_env_var():
  function test_env_var_restored_on_exception (line 69) | def test_env_var_restored_on_exception():
  function test_nested_context_managers (line 85) | def test_nested_context_managers():
  function test_multiple_different_vars (line 103) | def test_multiple_different_vars():
  function test_empty_string_value (line 118) | def test_empty_string_value():
  function test_overwrite_with_empty_string (line 128) | def test_overwrite_with_empty_string():
  function test_context_manager_returns_none (line 139) | def test_context_manager_returns_none():

FILE: tests/utils/test_timeout_decorator_cpu.py
  function quick_task (line 30) | def quick_task(x):
  function slow_task (line 37) | def slow_task(x):
  function task_raises_value_error (line 44) | def task_raises_value_error():  # Now truly not globally decorated
  function top_level_decorated_quick_task_signal (line 52) | def top_level_decorated_quick_task_signal():
  function top_level_decorated_slow_task_signal (line 62) | def top_level_decorated_slow_task_signal():
  function run_target_and_put_in_queue (line 69) | def run_target_and_put_in_queue(target_func, q):
  function set_macos_start_method (line 83) | def set_macos_start_method():
  function test_quick_task (line 97) | def test_quick_task():  # Renamed from test_multiprocessing_quick_task
  function test_slow_task_timeout (line 104) | def test_slow_task_timeout():  # Renamed from test_multiprocessing_slow_...
  function test_internal_exception (line 113) | def test_internal_exception():  # Renamed from test_multiprocessing_inte...
  function test_signal_quick_task_main_process (line 127) | def test_signal_quick_task_main_process():  # Removed self
  function test_signal_slow_task_main_process_timeout (line 139) | def test_signal_slow_task_main_process_timeout():  # Removed self
  function test_signal_in_thread_does_not_timeout (line 155) | def test_signal_in_thread_does_not_timeout():
  function test_in_thread_timeout (line 200) | def test_in_thread_timeout():

FILE: tests/utils/test_tokenizer_normalize_on_cpu.py
  class DummyBatchEncoding (line 21) | class DummyBatchEncoding:
    method __init__ (line 22) | def __init__(self, input_ids):
  class DummyToList (line 26) | class DummyToList:
    method __init__ (line 27) | def __init__(self, data):
    method tolist (line 30) | def tolist(self):
  function test_normalize_token_ids_valid_outputs (line 53) | def test_normalize_token_ids_valid_outputs(tokenized_output, expected):
  function test_normalize_token_ids_invalid_outputs (line 66) | def test_normalize_token_ids_invalid_outputs(tokenized_output):

FILE: tests/utils/test_torch_functional.py
  function _worker_mean (line 31) | def _worker_mean(rank: int, world_size: int, rendezvous_file: str):
  function test_masked_mean (line 68) | def test_masked_mean(value, mask, gt):
  function test_distributed_mean_max_min_std (line 75) | def test_distributed_mean_max_min_std(world_size, tmp_path):
  function _worker_mask (line 87) | def _worker_mask(rank: int, world_size: int, rendezvous_file: str):
  function test_distributed_masked_mean (line 113) | def test_distributed_masked_mean(world_size, tmp_path):
  function test_expand_as_nested (line 125) | def test_expand_as_nested():

FILE: tests/utils/test_torch_profile.py
  class TestTorchProfile (line 24) | class TestTorchProfile(unittest.TestCase):
    method setUp (line 25) | def setUp(self):
    method test_get_torch_profiler (line 30) | def test_get_torch_profiler(self, mock_profile):
    method test_profiler_lifecycle (line 47) | def test_profiler_lifecycle(self, mock_get_profiler):
    method test_discrete_mode (line 71) | def test_discrete_mode(self, mock_get_profiler):

FILE: tests/workers/actor/test_special_dp_actor.py
  class MockTransformerModel (line 28) | class MockTransformerModel(nn.Module):
    method __init__ (line 31) | def __init__(self, vocab_size=1000, hidden_size=64):
    method forward (line 41) | def forward(self, input_ids, attention_mask=None, position_ids=None, u...
  class TestDataParallelPPOActor (line 55) | class TestDataParallelPPOActor(unittest.TestCase):
    method setUpClass (line 59) | def setUpClass(cls):
    method setUp (line 83) | def setUp(self):
    method tearDownClass (line 108) | def tearDownClass(cls):
    method _create_test_data_for_compute_log_prob (line 113) | def _create_test_data_for_compute_log_prob(self):
    method _create_test_data_for_update_policy (line 140) | def _create_test_data_for_update_policy(self):
    method test_compute_log_prob (line 173) | def test_compute_log_prob(self):
    method test_compute_log_prob_without_entropy (line 193) | def test_compute_log_prob_without_entropy(self):
    method test_update_policy (line 209) | def test_update_policy(self):
    method test_dataparallelppoactor_initialization (line 233) | def test_dataparallelppoactor_initialization(self):
    method test_dataparallelppoactor_with_qwen3_model (line 243) | def test_dataparallelppoactor_with_qwen3_model(self):

FILE: tests/workers/config/test_actor_config_on_cpu.py
  class TestActorConfig (line 27) | class TestActorConfig(unittest.TestCase):
    method test_config_inheritance (line 30) | def test_config_inheritance(self):
    method test_actor_config_from_yaml (line 66) | def test_actor_config_from_yaml(self):
    method test_fsdp_actor_config_from_yaml (line 78) | def test_fsdp_actor_config_from_yaml(self):
    method test_megatron_actor_config_from_yaml (line 90) | def test_megatron_actor_config_from_yaml(self):
    method test_config_get_method (line 102) | def test_config_get_method(self):
    method test_config_dict_like_access (line 123) | def test_config_dict_like_access(self):
    method test_frozen_fields_modification_raises_exception (line 147) | def test_frozen_fields_modification_raises_exception(self):
    method test_actor_config_validation_exceptions (line 171) | def test_actor_config_validation_exceptions(self):
    method test_fsdp_actor_config_validation_exceptions (line 217) | def test_fsdp_actor_config_validation_exceptions(self):
    method test_actor_config_validate_method_exceptions (line 233) | def test_actor_config_validate_method_exceptions(self):

FILE: tests/workers/config/test_critic_config_on_cpu.py
  class TestCriticConfig (line 34) | class TestCriticConfig:
    method config_dir (line 38) | def config_dir(self):
    method test_megatron_critic_config_instantiation_from_yaml (line 42) | def test_megatron_critic_config_instantiation_from_yaml(self, config_d...
    method test_fsdp_critic_config_instantiation_from_yaml (line 74) | def test_fsdp_critic_config_instantiation_from_yaml(self, config_dir):
    method test_config_inheritance_hierarchy (line 107) | def test_config_inheritance_hierarchy(self):
    method test_config_dict_interface (line 122) | def test_config_dict_interface(self):
    method test_frozen_fields_immutability (line 139) | def test_frozen_fields_immutability(self):
    method test_batch_size_fields_modifiable (line 162) | def test_batch_size_fields_modifiable(self):
    method test_profiler_config_type_validation (line 183) | def test_profiler_config_type_validation(self):
    method test_critic_config_validation_logic (line 211) | def test_critic_config_validation_logic(self):
    method test_micro_batch_size_divisibility_validation (line 254) | def test_micro_batch_size_divisibility_validation(self):
    method test_fsdp_sequence_parallelism_validation (line 279) | def test_fsdp_sequence_parallelism_validation(self):

FILE: tests/workers/config/test_engine_config_on_cpu.py
  class TestMcoreEngineConfig (line 20) | class TestMcoreEngineConfig:
    method test_default_values (line 21) | def test_default_values(self):
    method test_post_init_validation (line 27) | def test_post_init_validation(self):
    method test_mutable_fields (line 36) | def test_mutable_fields(self):
    method test_offload_flags (line 43) | def test_offload_flags(self, offload_field):
  class TestFSDPEngineConfigCPU (line 48) | class TestFSDPEngineConfigCPU:
    method test_default_values (line 49) | def test_default_values(self):
    method test_offload_combinations (line 59) | def test_offload_combinations(self, offload_params):
    method test_wrap_policy_configuration (line 64) | def test_wrap_policy_configuration(self):

FILE: tests/workers/config/test_model_config_on_cpu.py
  class TestHFModelConfigCPU (line 23) | class TestHFModelConfigCPU:
    method test_target_modules_accepts_list_via_omegaconf (line 26) | def test_target_modules_accepts_list_via_omegaconf(self):
    method test_target_modules_accepts_none_via_omegaconf (line 55) | def test_target_modules_accepts_none_via_omegaconf(self):
    method test_target_modules_accepts_string_via_omegaconf (line 70) | def test_target_modules_accepts_string_via_omegaconf(self):
    method test_target_modules_raises_on_invalid_type (line 85) | def test_target_modules_raises_on_invalid_type(self):

FILE: tests/workers/config/test_optim_config_on_cpu.py
  class TestFSDPOptimizerConfigCPU (line 20) | class TestFSDPOptimizerConfigCPU:
    method test_default_configuration (line 21) | def test_default_configuration(self):
    method test_valid_lr_scheduler_types (line 28) | def test_valid_lr_scheduler_types(self, lr_scheduler_type):
    method test_valid_warmup_style_types (line 33) | def test_valid_warmup_style_types(self, warmup_style):
    method test_invalid_lr_scheduler_type (line 37) | def test_invalid_lr_scheduler_type(self):
    method test_invalid_warmup_style_type (line 41) | def test_invalid_warmup_style_type(self):
    method test_num_cycles_configuration (line 46) | def test_num_cycles_configuration(self, num_cycles):

FILE: tests/workers/critic/test_special_dp_critic.py
  class TestCriticWorker (line 33) | class TestCriticWorker(unittest.TestCase):
    method setUpClass (line 35) | def setUpClass(cls):
    method tearDownClass (line 52) | def tearDownClass(cls):
    method setUp (line 57) | def setUp(self):
    method tearDown (line 88) | def tearDown(self):
    method _create_test_data_for_compute_values (line 94) | def _create_test_data_for_compute_values(self, batch_size=2, seq_len=1...
    method _create_test_data_for_update_critic (line 119) | def _create_test_data_for_update_critic(self, batch_size=2, seq_len=10...
    method test_init_model (line 149) | def test_init_model(self):
    method test_compute_values (line 159) | def test_compute_values(self):
    method test_update_critic (line 177) | def test_update_critic(self):
    method test_critic_attn_implementation_override_functionality (line 202) | def test_critic_attn_implementation_override_functionality(self, mock_...
    method test_critic_model_config_structure (line 260) | def test_critic_model_config_structure(self):
    method test_critic_hydra_config_compatibility (line 290) | def test_critic_hydra_config_compatibility(self):
    method test_critic_backward_compatibility (line 310) | def test_critic_backward_compatibility(self):
    method test_critic_and_actor_independent_configuration (line 333) | def test_critic_and_actor_independent_configuration(self):

FILE: tests/workers/reward_manager/test_registry_on_cpu.py
  function setup (line 22) | def setup():
  function test_get_existing_manager (line 29) | def test_get_existing_manager(setup):
  function test_get_nonexistent_manager (line 35) | def test_get_nonexistent_manager(setup):
  function test_case_sensitivity (line 42) | def test_case_sensitivity(setup):
  function test_empty_registry (line 50) | def test_empty_registry(setup):
  function test_register_new_class (line 58) | def test_register_new_class(setup):
  function test_register_different_classes_same_name (line 69) | def test_register_different_classes_same_name(setup):
  function test_decorator_returns_original_class (line 85) | def test_decorator_returns_original_class(setup):

FILE: tests/workers/rollout/perf/vllm_async_rollout.py
  function init_config (line 48) | def init_config(n_gpus_per_node) -> DictConfig:
  function initialize (line 77) | def initialize(config, backend) -> tuple[AgentLoopManager | RayWorkerGro...
  function perf_rollout (line 107) | def perf_rollout(mode, backend, n_gpus_per_node, num_steps):

FILE: tests/workers/rollout/rollout_sglang/test_http_server_engine.py
  function event_loop (line 63) | def event_loop():
  function basic_adapter_kwargs (line 71) | def basic_adapter_kwargs():
  function router_adapter_kwargs (line 82) | def router_adapter_kwargs():
  function non_master_adapter_kwargs (line 95) | def non_master_adapter_kwargs():
  function mock_launch_server_process (line 106) | def mock_launch_server_process():
  function mock_multiprocessing_process (line 119) | def mock_multiprocessing_process():
  function mock_requests_session (line 132) | def mock_requests_session():
  function mock_requests_post (line 148) | def mock_requests_post():
  function mock_requests_get (line 161) | def mock_requests_get():
  function mock_aiohttp_session (line 174) | def mock_aiohttp_session():
  function mock_kill_process_tree (line 193) | def mock_kill_process_tree():
  function sglang_test_model_path (line 203) | def sglang_test_model_path():
  function real_adapter_kwargs (line 215) | def real_adapter_kwargs(sglang_test_model_path):
  function mock_server_args_post_init (line 226) | def mock_server_args_post_init():
  class TestLaunchServerProcess (line 236) | class TestLaunchServerProcess:
    method test_launch_server_process_success (line 239) | def test_launch_server_process_success(
    method test_launch_server_process_non_master (line 264) | def test_launch_server_process_non_master(self, mock_multiprocessing_p...
    method test_launch_server_process_timeout (line 279) | def test_launch_server_process_timeout(self, mock_multiprocessing_proc...
    method test_launch_server_process_died (line 305) | def test_launch_server_process_died(self, real_adapter_kwargs):
  class TestHttpServerEngineAdapter (line 322) | class TestHttpServerEngineAdapter:
    method test_init_with_router_registration (line 325) | def test_init_with_router_registration(self, mock_launch_server_proces...
    method test_init_without_router (line 334) | def test_init_without_router(self, mock_launch_server_process, basic_a...
    method test_register_with_router_failure (line 342) | def test_register_with_router_failure(self, mock_launch_server_process...
    method test_make_request_success (line 353) | def test_make_request_success(self, mock_launch_server_process, basic_...
    method test_make_request_get_method (line 372) | def test_make_request_get_method(self, mock_launch_server_process, bas...
    method test_make_request_non_master (line 387) | def test_make_request_non_master(self, mock_launch_server_process):
    method test_make_request_retry_logic (line 395) | def test_make_request_retry_logic(self, mock_launch_server_process, ba...
    method test_make_request_http_error (line 414) | def test_make_request_http_error(self, mock_launch_server_process, bas...
    method test_make_request_max_attempts_exceeded (line 426) | def test_make_request_max_attempts_exceeded(self, mock_launch_server_p...
    method test_update_weights_from_tensor_strict (line 439) | def test_update_weights_from_tensor_strict(self, mock_launch_server_pr...
    method test_update_weights_from_tensor_empty (line 473) | def test_update_weights_from_tensor_empty(self, mock_launch_server_pro...
    method test_update_weights_from_tensor_none (line 502) | def test_update_weights_from_tensor_none(self, mock_launch_server_proc...
    method test_generate (line 531) | def test_generate(self, mock_launch_server_process, basic_adapter_kwar...
    method test_flush_cache (line 555) | def test_flush_cache(self, mock_launch_server_process, basic_adapter_k...
    method test_flush_cache_non_master (line 574) | def test_flush_cache_non_master(self, mock_launch_server_process):
    method test_memory_management_methods (line 582) | def test_memory_management_methods(self, mock_launch_server_process, b...
    method test_generation_control_methods (line 599) | def test_generation_control_methods(self, mock_launch_server_process, ...
    method test_shutdown (line 606) | def test_shutdown(self, mock_launch_server_process, mock_kill_process_...
    method test_shutdown_with_errors (line 622) | def test_shutdown_with_errors(self, mock_launch_server_process, mock_k...
    method test_empty_and_none_parameters (line 643) | def test_empty_and_none_parameters(self, mock_launch_server_process, b...
    method test_large_payload_handling (line 667) | def test_large_payload_handling(self, mock_launch_server_process, basi...
    method test_timeout_edge_cases (line 690) | def test_timeout_edge_cases(self, mock_launch_server_process):
    method test_extreme_configuration_values (line 702) | def test_extreme_configuration_values(self, mock_launch_server_process):
  class TestAsyncHttpServerEngineAdapter (line 721) | class TestAsyncHttpServerEngineAdapter:
    method test_init (line 724) | def test_init(self, mock_launch_server_process, basic_adapter_kwargs):
    method test_make_async_request_success (line 731) | async def test_make_async_request_success(self, mock_launch_server_pro...
    method test_make_async_request_get_method (line 764) | async def test_make_async_request_get_method(self, mock_launch_server_...
    method test_make_async_request_non_master (line 793) | async def test_make_async_request_non_master(self, mock_launch_server_...
    method test_async_generate (line 802) | async def test_async_generate(self, mock_launch_server_process, basic_...
    method test_async_memory_management (line 819) | async def test_async_memory_management(self, mock_launch_server_proces...
  class TestErrorRecovery (line 840) | class TestErrorRecovery:
    method test_flush_cache_recovery (line 843) | def test_flush_cache_recovery(self, mock_launch_server_process, basic_...
    method test_flush_cache_max_attempts (line 860) | def test_flush_cache_max_attempts(self, mock_launch_server_process, ba...
    method test_network_partition_recovery (line 872) | def test_network_partition_recovery(self, mock_launch_server_process, ...
  class TestResourceManagement (line 889) | class TestResourceManagement:
    method test_resource_cleanup_on_exception (line 892) | def test_resource_cleanup_on_exception(
    method test_multiple_shutdown_calls (line 909) | def test_multiple_shutdown_calls(self, mock_launch_server_process, bas...
  class TestDataTypeHandling (line 919) | class TestDataTypeHandling:
    method test_complex_data_structures (line 922) | def test_complex_data_structures(self, mock_launch_server_process, bas...
  class TestIntegration (line 956) | class TestIntegration:
    method test_error_scenarios (line 959) | def test_error_scenarios(self, mock_launch_server_process, basic_adapt...

FILE: tests/workers/rollout/rollout_trtllm/test_adapter.py
  class TestAsyncTRTLLMHttpAdapter (line 27) | class TestAsyncTRTLLMHttpAdapter:
    method _build_async_session (line 28) | def _build_async_session(
    method test_make_async_request_get_method (line 48) | async def test_make_async_request_get_method(self):
    method test_make_async_request_post_method (line 70) | async def test_make_async_request_post_method(self):
    method test_make_async_request_http_error (line 94) | async def test_make_async_request_http_error(self):
    method test_make_async_request_max_attempts_exceeded (line 120) | async def test_make_async_request_max_attempts_exceeded(self):
  class TestTRTLLMServerAdapter (line 135) | class TestTRTLLMServerAdapter:
    method test_init_without_device_mesh (line 136) | def test_init_without_device_mesh(self):

FILE: tests/workers/rollout/rollout_trtllm/test_async_server.py
  class TestTRTLLMReplica (line 30) | class TestTRTLLMReplica:
    method test_placement_group_with_sub_ray_resource_pool (line 31) | def test_placement_group_with_sub_ray_resource_pool(self):
    method test_placement_group_with_ray_resource_pool (line 69) | def test_placement_group_with_ray_resource_pool(self):
  class TestTRTLLMHttpServer (line 110) | class TestTRTLLMHttpServer:
    method _build_rollout_config (line 112) | def _build_rollout_config(*, response_length: int | None = None, free_...
    method _create_server (line 137) | def _create_server(rollout_config, model_config, *, name: str):
    method test_async_generate (line 169) | def test_async_generate(self):
    method test_async_memory_management (line 215) | def test_async_memory_management(self):

FILE: tests/workers/rollout/rollout_trtllm/test_trtllm_rollout_utils.py
  function create_test_image (line 35) | def create_test_image(width: int = 224, height: int = 224) -> Image.Image:
  function create_rollout_config_dict (line 47) | def create_rollout_config_dict():
  function create_model_config_dict (line 77) | def create_model_config_dict(model_path: str):
  function get_tokenizer (line 87) | def get_tokenizer(model_path: str):
  function get_processor (line 91) | def get_processor(model_path: str):
  class TestUnimodalTRTLLMRollout (line 101) | class TestUnimodalTRTLLMRollout:
    method ray_context (line 103) | def ray_context(self):
    method trtllm_replica (line 111) | def trtllm_replica(self, ray_context):
    method tokenizer (line 134) | def tokenizer(self):
    method test_unimodal_generate (line 145) | def test_unimodal_generate(self, trtllm_replica, tokenizer, prompt):
    method test_unimodal_batch_generate (line 185) | def test_unimodal_batch_generate(self, trtllm_replica, tokenizer):
  class TestMultimodalTRTLLMRollout (line 230) | class TestMultimodalTRTLLMRollout:
    method ray_context (line 232) | def ray_context(self):
    method trtllm_vlm_replica (line 240) | def trtllm_vlm_replica(self, ray_context):
    method tokenizer (line 263) | def tokenizer(self):
    method processor (line 267) | def processor(self):
    method test_multimodal_generate_with_image (line 278) | def test_multimodal_generate_with_image(self, trtllm_vlm_replica, proc...
    method test_multimodal_different_image_sizes (line 336) | def test_multimodal_different_image_sizes(self, trtllm_vlm_replica, pr...
    method test_multimodal_text_only_fallback (line 376) | def test_multimodal_text_only_fallback(self, trtllm_vlm_replica, token...
  class TestTRTLLMServerLifecycle (line 413) | class TestTRTLLMServerLifecycle:
    method ray_context (line 415) | def ray_context(self):
    method trtllm_replica_lifecycle (line 423) | def trtllm_replica_lifecycle(self, ray_context):
    method tokenizer (line 446) | def tokenizer(self):
    method test_wake_sleep_cycle (line 449) | def test_wake_sleep_cycle(self, trtllm_replica_lifecycle, tokenizer):

FILE: tests/workers/rollout/rollout_vllm/run_fsdp_vllm.py
  function _pre_process_inputs (line 30) | def _pre_process_inputs(pad_token_id, prompt_token_ids: torch.Tensor) ->...
  function main (line 36) | def main():

FILE: tests/workers/rollout/rollout_vllm/test_vllm_abort.py
  function test_vllm_abort (line 29) | def test_vllm_abort():

FILE: tests/workers/rollout/test_hf_rollout.py
  function prepare_input_dataproto (line 48) | def prepare_input_dataproto(tokenizer, config, validate):
  function prepare_fsdp_model (line 75) | def prepare_fsdp_model(model, world_size):
  function test_hf_rollout (line 100) | def test_hf_rollout(n: int = 1, do_sample: bool = True, validate: bool =...

FILE: tests/workers/rollout/test_sglang_async_rollout_multimodal_delta.py
  function _test_add_tool_response_messages_image_delta (line 31) | def _test_add_tool_response_messages_image_delta(processor, image_list, ...
  function test_add_tool_response_messages_image_delta (line 157) | def test_add_tool_response_messages_image_delta():
  function test_add_tool_response_messages_image_delta_resize_image (line 179) | def test_add_tool_response_messages_image_delta_resize_image():

FILE: tests/workers/rollout/test_sglang_rollout_sharding_manager.py
  function test_get_named_tensor_buckets (line 50) | def test_get_named_tensor_buckets(named_tensors, bucket_size_mb, gt_grou...

FILE: tests/workers/rollout/test_vllm_cli_args_on_cpu.py
  class TestBuildCliArgsFromConfig (line 22) | class TestBuildCliArgsFromConfig:
    method test_string_value (line 25) | def test_string_value(self):
    method test_integer_value (line 31) | def test_integer_value(self):
    method test_float_value (line 37) | def test_float_value(self):
    method test_bool_true (line 43) | def test_bool_true(self):
    method test_bool_false (line 49) | def test_bool_false(self):
    method test_none_value (line 55) | def test_none_value(self):
    method test_list_values (line 61) | def test_list_values(self):
    method test_empty_list (line 67) | def test_empty_list(self):
    method test_list_with_strings (line 73) | def test_list_with_strings(self):
    method test_dict_value (line 79) | def test_dict_value(self):
    method test_mixed_config (line 87) | def test_mixed_config(self):
    method test_preserves_order (line 113) | def test_preserves_order(self):
    method test_empty_config (line 119) | def test_empty_config(self):
    method test_single_element_list (line 125) | def test_single_element_list(self):

FILE: tests/workers/test_fsdp_attn_implementation.py
  class TestFSDPAttnImplementation (line 43) | class TestFSDPAttnImplementation:
    method test_attn_implementation_extraction_logic (line 46) | def test_attn_implementation_extraction_logic(self):
    method test_attn_implementation_passed_to_autoconfig (line 71) | def test_attn_implementation_passed_to_autoconfig(self, mock_model_fro...
    method test_attn_implementation_passed_to_model (line 109) | def test_attn_implementation_passed_to_model(self, mock_model_from_pre...
    method test_override_config_integration (line 144) | def test_override_config_integration(self):
    method test_hydra_plus_prefix_config (line 165) | def test_hydra_plus_prefix_config(self):
    method test_backward_compatibility (line 194) | def test_backward_compatibility(self):
    method test_critic_attn_implementation_extraction_logic (line 214) | def test_critic_attn_implementation_extraction_logic(self):
    method test_critic_attn_implementation_passed_to_autoconfig (line 238) | def test_critic_attn_implementation_passed_to_autoconfig(self, mock_co...
    method test_critic_override_config_integration (line 277) | def test_critic_override_config_integration(self):
    method test_critic_hydra_plus_prefix_config (line 302) | def test_critic_hydra_plus_prefix_config(self):
    method test_both_actor_and_critic_configuration (line 331) | def test_both_actor_and_critic_configuration(self):
    method test_critic_backward_compatibility (line 358) | def test_critic_backward_compatibility(self):
  function test_attn_implementation_fix_integration (line 379) | def test_attn_implementation_fix_integration():
  function test_critic_attn_implementation_fix_integration (line 411) | def test_critic_attn_implementation_fix_integration():
  function test_complete_training_configuration (line 438) | def test_complete_training_configuration():

FILE: tests/workers/test_fsdp_workers.py
  function test_actor_rollout_ref_worker_actor_ref_model (line 21) | def test_actor_rollout_ref_worker_actor_ref_model():

FILE: verl/__init__.py
  function _sync_all_patch (line 90) | def _sync_all_patch(self):

FILE: verl/base_config.py
  class BaseConfig (line 22) | class BaseConfig(collections.abc.Mapping):
    method __setattr__ (line 33) | def __setattr__(self, name: str, value):
    method get (line 40) | def get(self, key: str, default: Any = None) -> Any:
    method __getitem__ (line 55) | def __getitem__(self, key: str):
    method __iter__ (line 70) | def __iter__(self):
    method __len__ (line 79) | def __len__(self):

FILE: verl/checkpoint_engine/base.py
  class TensorMeta (line 30) | class TensorMeta(TypedDict):
  class CheckpointEngineRegistry (line 37) | class CheckpointEngineRegistry:
    method register (line 42) | def register(backend: str):
    method get (line 56) | def get(cls, backend: str) -> type["CheckpointEngine"]:
    method new (line 68) | def new(cls, backend: str, *args, **kwargs) -> "CheckpointEngine":
  class CheckpointEngine (line 84) | class CheckpointEngine(ABC):
    method prepare (line 99) | def prepare(self) -> dict[str, Any]:
    method build_topology (line 116) | def build_topology(
    method init_process_group (line 143) | def init_process_group(self, **kwargs):
    method finalize (line 152) | def finalize(self):
    method send_weights (line 162) | async def send_weights(self, weights: Generator[tuple[str, torch.Tenso...
    method receive_weights (line 171) | async def receive_weights(self) -> Generator[tuple[str, torch.Tensor],...
  class CheckpointEngineWithCache (line 180) | class CheckpointEngineWithCache(CheckpointEngine):
    method get_weights (line 188) | async def get_weights(self) -> Generator[tuple[str, torch.Tensor], Non...
  class ColocatedCheckpointEngine (line 198) | class ColocatedCheckpointEngine(CheckpointEngine):
    method __init__ (line 209) | def __init__(self, bucket_size: int, is_master: bool = False) -> None:
    method prepare (line 213) | def prepare(self):
    method init_process_group (line 216) | def init_process_group(self, **kwargs):
    method finalize (line 219) | def finalize(self):
    method build_topology (line 223) | def build_topology(cls, *args, **kwargs):
    method send_weights (line 226) | def send_weights(self, weights: Generator[tuple[str, torch.Tensor], No...
    method receive_weights (line 234) | def receive_weights(self) -> Generator[tuple[str, torch.Tensor], None,...
  class CheckpointEngineWorker (line 244) | class CheckpointEngineWorker(Worker):
    method __init__ (line 253) | def __init__(
    method update_weights (line 286) | async def update_weights(self, global_steps: int = None):
    method execute_checkpoint_engine (line 291) | def execute_checkpoint_engine(self, method: str, *args, **kwargs):
    method get_replica_rank (line 295) | def get_replica_rank(self) -> int:
    method is_leader_rank (line 300) | def is_leader_rank(self) -> bool:
  class CheckpointEngineManager (line 308) | class CheckpointEngineManager:
    method __init__ (line 337) | def __init__(
    method build_process_group (line 349) | def build_process_group(self, rollout: RayWorkerGroup):
    method add_replicas (line 376) | def add_replicas(self, replicas: list[RolloutReplica]):
    method remove_replicas (line 384) | def remove_replicas(self, replicas: list[RolloutReplica]):
    method sleep_replicas (line 394) | async def sleep_replicas(self):
    method wake_up_replicas (line 399) | async def wake_up_replicas(self):
    method update_weights (line 404) | async def update_weights(self, global_steps: int = None):

FILE: verl/checkpoint_engine/hccl_checkpoint_engine.py
  class MasterMetadata (line 38) | class MasterMetadata:
  class BroadcastOperation (line 45) | class BroadcastOperation:
    method __init__ (line 57) | def __init__(
    method _run (line 75) | def _run(self):
    method wait_for_complete (line 87) | async def wait_for_complete(self) -> dict[str, TensorMeta]:
  class HCCLCheckpointEngine (line 97) | class HCCLCheckpointEngine(CheckpointEngine):
    method __init__ (line 109) | def __init__(
    method prepare (line 131) | def prepare(self) -> MasterMetadata:
    method finalize (line 141) | def finalize(self):
    method build_topology (line 155) | def build_topology(cls, trainer_world_size: int, rollout_world_size: i...
    method _start_zmq_server (line 168) | def _start_zmq_server(self):
    method _connect_zmq_client (line 182) | def _connect_zmq_client(self, metadata: MasterMetadata):
    method init_process_group (line 195) | def init_process_group(self, rank: int, world_size: int, master_metada...
    method send_weights (line 230) | async def send_weights(self, weights: Generator[tuple[str, torch.Tenso...
    method receive_weights (line 303) | async def receive_weights(self) -> AsyncGenerator[tuple[str, torch.Ten...

FILE: verl/checkpoint_engine/kimi_checkpoint_engine.py
  function ckpt_get_named_tensor_buckets (line 37) | def ckpt_get_named_tensor_buckets(
  function receive_tensor (line 66) | async def receive_tensor(
  class MasterMetadata (line 176) | class MasterMetadata:
  class BroadcastOperation (line 183) | class BroadcastOperation:
    method __init__ (line 193) | def __init__(
    method _run (line 208) | def _run(self):
    method wait_for_complete (line 212) | async def wait_for_complete(self) -> list[ParameterMeta]:
  class KIMICheckpointEngine (line 223) | class KIMICheckpointEngine(CheckpointEngine):
    method __init__ (line 234) | def __init__(
    method prepare (line 248) | def prepare(self) -> MasterMetadata:
    method finalize (line 259) | def finalize(self):
    method build_topology (line 268) | def build_topology(cls, trainer_world_size: int, rollout_world_size: i...
    method init_process_group (line 285) | def init_process_group(
    method send_weights (line 321) | async def send_weights(self, weights: Generator[tuple[str, torch.Tenso...
    method receive_weights (line 362) | async def receive_weights(self) -> AsyncGenerator[tuple[str, torch.Ten...

FILE: verl/checkpoint_engine/mooncake_checkpoint_engine.py
  class MooncakeCheckpointEngine (line 35) | class MooncakeCheckpointEngine(CheckpointEngine):
    method __init__ (line 45) | def __init__(
    method prepare (line 88) | def prepare(self) -> dict[str, Any]:
    method build_topology (line 98) | def build_topology(cls, trainer_world_size: int, rollout_world_size: i...
    method init_process_group (line 111) | def init_process_group(self, rank: int, world_size: int, metadata: dic...
    method finalize (line 135) | def finalize(self):
    method wait_for_complete (line 142) | async def wait_for_complete(self, buf: torch.Tensor):
    method send_weights (line 150) | async def send_weights(self, weights: Generator[tuple[str, torch.Tenso...
    method receive_weights (line 222) | async def receive_weights(self) -> AsyncGenerator[tuple[str, torch.Ten...

FILE: verl/checkpoint_engine/nccl_checkpoint_engine.py
  class MasterMetadata (line 38) | class MasterMetadata:
  class BroadcastOperation (line 43) | class BroadcastOperation:
    method __init__ (line 55) | def __init__(
    method _run (line 74) | def _run(self):
    method wait_for_complete (line 86) | async def wait_for_complete(self) -> dict[str, TensorMeta]:
  class NCCLCheckpointEngine (line 97) | class NCCLCheckpointEngine(CheckpointEngine):
    method __init__ (line 109) | def __init__(
    method prepare (line 128) | def prepare(self) -> MasterMetadata:
    method finalize (line 140) | def finalize(self):
    method build_topology (line 154) | def build_topology(cls, trainer_world_size: int, rollout_world_size: i...
    method _start_zmq_server (line 167) | def _start_zmq_server(self):
    method _connect_zmq_client (line 181) | def _connect_zmq_client(self, metadata: MasterMetadata):
    method init_process_group (line 194) | def init_process_group(self, rank: int, world_size: int, master_metada...
    method send_weights (line 224) | async def send_weights(self, weights: Generator[tuple[str, torch.Tenso...
    method receive_weights (line 297) | async def receive_weights(self) -> AsyncGenerator[tuple[str, torch.Ten...

FILE: verl/checkpoint_engine/nixl_checkpoint_engine.py
  class NixlAgentMetadata (line 42) | class NixlAgentMetadata:
  class NixlAgent (line 49) | class NixlAgent:
    method __init__ (line 54) | def __init__(self):
    method __getattr__ (line 63) | def __getattr__(self, name):
    method get_agent_metadata (line 75) | def get_agent_metadata(self) -> NixlAgentMetadata:
    method start_zmq_server (line 83) | def start_zmq_server(self):
    method add_remote_agent (line 97) | def add_remote_agent(self, metadata: NixlAgentMetadata) -> str:
    method remove_remote_agent (line 113) | def remove_remote_agent(self, agent_name: str):
    method send_message (line 118) | def send_message(self, agent_name, message: dict):
    method read_message (line 122) | async def read_message(self, agent_name: str) -> dict:
    method get_notification (line 128) | async def get_notification(self, remote_name: str) -> bytes:
  class ReadableOperation (line 137) | class ReadableOperation:
    method __init__ (line 150) | def __init__(
    method wait_for_complete (line 164) | async def wait_for_complete(self):
  class ReadOperation (line 171) | class ReadOperation:
    method __init__ (line 184) | def __init__(self, agent: NixlAgent, remote_agent: str, local_descs: n...
    method read_metadata (line 194) | async def read_metadata(self) -> dict:
    method begin_read (line 205) | def begin_read(self):
    method wait_for_complete (line 215) | async def wait_for_complete(self):
  class NIXLCheckpointEngine (line 233) | class NIXLCheckpointEngine(CheckpointEngine):
    method __init__ (line 246) | def __init__(
    method prepare (line 259) | def prepare(self) -> NixlAgentMetadata:
    method build_topology (line 283) | def build_topology(cls, trainer_world_size: int, rollout_world_size: i...
    method init_process_group (line 301) | def init_process_group(
    method finalize (line 343) | def finalize(self):
    method send_weights (line 365) | async def send_weights(self, weights: Generator[tuple[str, torch.Tenso...
    method receive_weights (line 435) | async def receive_weights(self) -> AsyncGenerator[tuple[str, torch.Ten...

FILE: verl/experimental/agent_loop/agent_loop.py
  class GlobalRequestLoadBalancer (line 58) | class GlobalRequestLoadBalancer:
    method __init__ (line 61) | def __init__(self, server_actor_ids: list[str], max_cache_size: int = ...
    method acquire_server (line 68) | def acquire_server(self, request_id: str) -> str:
    method release_server (line 82) | def release_server(self, server_id: str) -> None:
  function _get_rollout_and_model_config (line 91) | def _get_rollout_and_model_config(config: DictConfig) -> tuple[DictConfi...
  class AsyncLLMServerManager (line 99) | class AsyncLLMServerManager:
    method __init__ (line 106) | def __init__(
    method _acquire_server (line 123) | async def _acquire_server(self, request_id: str) -> tuple[str, ray.act...
    method _release_server (line 130) | def _release_server(self, server_id: str) -> None:
    method generate (line 136) | async def generate(
  class AgentLoopMetrics (line 169) | class AgentLoopMetrics(BaseModel):
  class AgentLoopOutput (line 177) | class AgentLoopOutput(BaseModel):
  class _InternalAgentLoopOutput (line 202) | class _InternalAgentLoopOutput(AgentLoopOutput):
  class DictConfigWrap (line 229) | class DictConfigWrap:
    method __init__ (line 232) | def __init__(self, config: DictConfig):
  class AgentLoopBase (line 236) | class AgentLoopBase(ABC):
    method __init__ (line 249) | def __init__(
    method process_vision_info (line 270) | async def process_vision_info(self, messages: list[dict]) -> dict:
    method apply_chat_template (line 291) | async def apply_chat_template(
    method run (line 360) | async def run(self, sampling_params: dict[str, Any], **kwargs) -> Agen...
  function register (line 381) | def register(agent_name: str):
  class AgentLoopWorker (line 392) | class AgentLoopWorker:
    method __init__ (line 401) | def __init__(
    method generate_sequences (line 454) | async def generate_sequences(self, batch: DataProto) -> DataProto:
    method _run_agent_loop (line 535) | async def _run_agent_loop(
    method _agent_loop_postprocess (line 569) | async def _agent_loop_postprocess(self, output, **kwargs) -> _Internal...
    method _compute_multi_modal_inputs (line 693) | def _compute_multi_modal_inputs(self, output, input_ids) -> dict[str, ...
    method _compute_position_ids (line 728) | def _compute_position_ids(self, input_ids, attention_mask, multi_modal...
    method _compute_score (line 759) | async def _compute_score(self, output, prompts, responses, attention_m...
    method _postprocess (line 789) | def _postprocess(
  function get_trajectory_info (line 880) | async def get_trajectory_info(step, index, validate):
  class AgentLoopManager (line 902) | class AgentLoopManager:
    method __init__ (line 915) | def __init__(
    method create (line 938) | async def create(
    method _initialize_llm_servers (line 952) | async def _initialize_llm_servers(self):
    method _init_agent_loop_workers (line 999) | async def _init_agent_loop_workers(self):
    method _init_global_load_balancer (line 1023) | async def _init_global_load_balancer(self) -> None:
    method generate_sequences (line 1030) | async def generate_sequences(self, prompts: DataProto) -> DataProto:
    method _performance_metrics (line 1056) | def _performance_metrics(self, metrics: list[list[dict[str, str]]], ou...
    method clear_kv_cache (line 1084) | async def clear_kv_cache(self):
    method start_profile (line 1089) | async def start_profile(self, **kwargs):
    method stop_profile (line 1094) | async def stop_profile(self):

FILE: verl/experimental/agent_loop/prometheus_utils.py
  function update_prometheus_config (line 28) | def update_prometheus_config(config: PrometheusConfig, server_addresses:...

FILE: verl/experimental/agent_loop/single_turn_agent_loop.py
  class SingleTurnAgentLoop (line 28) | class SingleTurnAgentLoop(AgentLoopBase):
    method __init__ (line 31) | def __init__(self, *args, **kwargs):
    method run (line 36) | async def run(self, sampling_params: dict[str, Any], **kwargs) -> Agen...

FILE: verl/experimental/agent_loop/tool_agent_loop.py
  class AgentState (line 44) | class AgentState(Enum):
  class AgentData (line 52) | class AgentData:
    method __init__ (line 56) | def __init__(
  class ToolAgentLoop (line 96) | class ToolAgentLoop(AgentLoopBase):
    method __init__ (line 97) | def __init__(self, *args, **kwargs):
    method run (line 124) | async def run(self, sampling_params: dict[str, Any], **kwargs) -> Agen...
    method _handle_pending_state (line 203) | async def _handle_pending_state(self, agent_data: AgentData, sampling_...
    method _handle_generating_state (line 214) | async def _handle_generating_state(
    method _handle_processing_tools_state (line 281) | async def _handle_processing_tools_state(self, agent_data: AgentData) ...
    method _handle_interacting_state (line 384) | async def _handle_interacting_state(self, agent_data: AgentData) -> Ag...
    method _call_tool (line 421) | async def _call_tool(
    method _initialize_interactions (line 471) | def _initialize_interactions(self, interaction_config_file):

FILE: verl/experimental/agent_loop/tool_parser.py
  class FunctionCall (line 31) | class FunctionCall(BaseModel):
  class ToolParser (line 44) | class ToolParser(ABC):
    method __init__ (line 47) | def __init__(self, tokenizer) -> None:
    method extract_tool_calls (line 51) | async def extract_tool_calls(
    method get_tool_parser (line 66) | def get_tool_parser(cls, name: str, tokenizer):
    method register (line 72) | def register(cls, name: str):
  class HermesToolParser (line 81) | class HermesToolParser(ToolParser):
    method __init__ (line 84) | def __init__(self, tokenizer) -> None:
    method extract_tool_calls (line 92) | async def extract_tool_calls(
  class GptOssToolParser (line 117) | class GptOssToolParser(ToolParser):
    method __init__ (line 126) | def __init__(self, tokenizer) -> None:
    method extract_tool_calls (line 141) | async def extract_tool_calls(
  class Qwen3XMLToolParser (line 174) | class Qwen3XMLToolParser(ToolParser):
    method __init__ (line 183) | def __init__(self, tokenizer):
    method _parse_xml_function_call (line 195) | def _parse_xml_function_call(
    method _get_function_calls (line 299) | def _get_function_calls(self, model_output: str) -> list[str]:
    method extract_tool_calls (line 316) | async def extract_tool_calls(

FILE: verl/experimental/agent_loop/utils.py
  function resolve_config_path (line 19) | def resolve_config_path(config_path: str) -> str:
  function format_gpt_oss_tool_response_manually (line 78) | def format_gpt_oss_tool_response_manually(tool_response: str, tool_call_...
  function add_generation_prompt_for_gpt_oss (line 90) | def add_generation_prompt_for_gpt_oss(message_content: str) -> str:
  function build_gpt_oss_tool_response_text (line 101) | def build_gpt_oss_tool_response_text(messages: list[dict[str, Any]], too...

FILE: verl/experimental/dataset/sampler.py
  class AbstractSampler (line 23) | class AbstractSampler(Sampler[int]):
    method __init__ (line 27) | def __init__(
  class AbstractCurriculumSampler (line 35) | class AbstractCurriculumSampler(AbstractSampler):
    method update (line 39) | def update(self, batch: DataProto) -> None:

FILE: verl/experimental/dynamic_dataset/dynamicgen_dataset.py
  class AbstractDataGenerator (line 38) | class AbstractDataGenerator(ABC):
    method __init__ (line 39) | def __init__(self, config: DictConfig):
    method generate (line 43) | def generate(self, dataset: Dataset) -> datasets.Dataset:
  class MockDataGenerator (line 54) | class MockDataGenerator(AbstractDataGenerator):
    method __init__ (line 60) | def __init__(self, config: DictConfig = None):
    method generate (line 63) | def generate(self, dataset: Dataset) -> datasets.Dataset:
  class DynamicGenDataset (line 68) | class DynamicGenDataset(RLHFDataset):
    method __init__ (line 74) | def __init__(
    method append_dataframe (line 100) | def append_dataframe(self, new_dataframe: datasets.Dataset):
    method on_batch_end (line 106) | def on_batch_end(self, batch: DataProto) -> None:

FILE: verl/experimental/fully_async_policy/agent_loop/agent_loop.py
  class FullyAsyncLLMServerManager (line 40) | class FullyAsyncLLMServerManager(AsyncLLMServerManager):
    method generate (line 46) | async def generate(
  class FullyAsyncAgentLoopWorker (line 127) | class FullyAsyncAgentLoopWorker(AgentLoopWorker):
    method __init__ (line 128) | def __init__(
  class FullyAsyncAgentLoopManager (line 139) | class FullyAsyncAgentLoopManager(AgentLoopManager):
    method __init__ (line 140) | def __init__(
    method generate_sequences_single (line 151) | async def generate_sequences_single(self, prompts: DataProto) -> DataP...
    method _select_best_worker (line 163) | def _select_best_worker(self):

FILE: verl/experimental/fully_async_policy/detach_utils.py
  class RolloutSample (line 28) | class RolloutSample:
  class ValidateMetrics (line 43) | class ValidateMetrics:
  function prepare_single_generation_data (line 50) | def prepare_single_generation_data(batch_dict, config) -> DataProto:
  function addition_process (line 84) | def addition_process(output: DataProto):
  function assemble_batch_from_rollout_samples (line 94) | def assemble_batch_from_rollout_samples(
  class MetricsAggregator (line 189) | class MetricsAggregator:
    method __init__ (line 192) | def __init__(self, total_gpus: int):
    method _init_aggregation_rules (line 207) | def _init_aggregation_rules(self) -> dict[str, dict[str, list[str]]]:
    method add_step_metrics (line 225) | def add_step_metrics(self, metrics: dict[str, Any], sample_count: int,...
    method _get_aggregation_type (line 241) | def _get_aggregation_type(self, metric_name: str) -> str:
    method _aggregate_single_metric (line 263) | def _aggregate_single_metric(self, metric_name: str, values: list[floa...
    method get_aggregated_metrics (line 302) | def get_aggregated_metrics(self) -> dict[str, Any]:
    method _special_metrics_aggergate (line 321) | def _special_metrics_aggergate(self, aggregated: dict[str, Any]) -> di...
    method reset (line 341) | def reset(self):
    method get_current_stats (line 348) | def get_current_stats(self) -> dict[str, Any]:
  function task_exception_handler (line 358) | def task_exception_handler(task: asyncio.Task):
  function safe_create_task (line 369) | def safe_create_task(coro, name: str, task_set: set = None):

FILE: verl/experimental/fully_async_policy/fully_async_main.py
  class FullyAsyncTaskRunner (line 35) | class FullyAsyncTaskRunner:
    method __init__ (line 40) | def __init__(self):
    method run (line 45) | def run(self, config):
    method _initialize_components (line 50) | def _initialize_components(self, config) -> None:
    method _create_rollouter (line 118) | def _create_rollouter(self, config) -> None:
    method _create_trainer (line 136) | def _create_trainer(self, config) -> None:
    method _run_training_loop (line 158) | def _run_training_loop(self):
  function main (line 195) | def main(config):

FILE: verl/experimental/fully_async_policy/fully_async_rollouter.py
  class FullyAsyncRollouter (line 43) | class FullyAsyncRollouter(SeparateRayPPOTrainer):
    method __init__ (line 50) | def __init__(
    method _init_async_objects (line 183) | def _init_async_objects(self):
    method set_message_queue_client (line 193) | async def set_message_queue_client(self, message_queue_client: Message...
    method set_max_required_samples (line 198) | async def set_max_required_samples(self):
    method get_rollout_wg (line 223) | def get_rollout_wg(self):
    method get_replicas (line 227) | def get_replicas(self):
    method get_max_queue_size (line 231) | def get_max_queue_size(self):
    method get_total_train_steps (line 234) | def get_total_train_steps(self):
    method reset_staleness (line 237) | async def reset_staleness(self):
    method do_validate (line 263) | def do_validate(self) -> ValidateMetrics:
    method save_checkpoint (line 270) | async def save_checkpoint(self, local_global_step_folder: str):
    method load_checkpoint (line 286) | def load_checkpoint(self):
    method _validate_config (line 344) | def _validate_config(self):
    method init_workers (line 350) | async def init_workers(self):
    method _create_actor_rollout_classes (line 362) | def _create_actor_rollout_classes(self):
    method _init_models (line 366) | def _init_models(self):
    method _create_continuous_iterator (line 371) | def _create_continuous_iterator(self):
    method _init_async_rollout_manager (line 380) | async def _init_async_rollout_manager(self):
    method _feed_samples (line 400) | async def _feed_samples(self):
    method _processor_worker (line 433) | async def _processor_worker(self):
    method _process_single_sample_streaming (line 500) | async def _process_single_sample_streaming(self, rollout_sample: Rollo...
    method _streaming_generation_main (line 519) | async def _streaming_generation_main(self):
    method fit (line 578) | async def fit(self):
    method _async_monitor_loop (line 614) | async def _async_monitor_loop(self):
    method _should_pause_generation (line 643) | async def _should_pause_generation(self) -> bool:
    method get_statistics (line 667) | async def get_statistics(self) -> dict:

FILE: verl/experimental/fully_async_policy/fully_async_trainer.py
  class TrainingStopException (line 47) | class TrainingStopException(Exception):
  class FullyAsyncTrainer (line 54) | class FullyAsyncTrainer(SeparateRayPPOTrainer):
    method __init__ (line 60) | def __init__(
    method _setup_checkpoint_manager (line 189) | def _setup_checkpoint_manager(self, rollouter):
    method set_message_queue_client (line 198) | def set_message_queue_client(self, message_queue_client: MessageQueueC...
    method set_rollouter (line 202) | def set_rollouter(self, rollouter):
    method set_total_train_steps (line 208) | def set_total_train_steps(self, total_training_steps):
    method get_actor_wg (line 223) | def get_actor_wg(self):
    method _get_samples_from_queue (line 227) | async def _get_samples_from_queue(self) -> tuple[None, None] | tuple[i...
    method _create_actor_rollout_classes (line 286) | def _create_actor_rollout_classes(self):
    method _init_models (line 297) | def _init_models(self):
    method init_workers (line 314) | async def init_workers(self):
    method _init_reward_loop (line 327) | def _init_reward_loop(self):
    method _init_async_rollout_manager (line 332) | async def _init_async_rollout_manager(self):
    method fit (line 389) | async def fit(self):
    method fit_step (line 421) | async def fit_step(self, batch_dict: dict = None):
    method _fit_generate (line 462) | async def _fit_generate(self, batch: DataProto = None) -> DataProto | ...
    method _compute_old_log_prob (line 473) | def _compute_old_log_prob(self, batch: DataProto):
    method _fit_update_local_step (line 495) | def _fit_update_local_step(self):
    method _fit_update_weights (line 509) | async def _fit_update_weights(self):
    method _validate_process (line 535) | async def _validate_process(self):
    method _fit_validate (line 558) | async def _fit_validate(self, val_before_train=False):
    method _fit_save_checkpoint (line 599) | def _fit_save_checkpoint(self, force=False):
    method _fit_postprocess_step (line 626) | def _fit_postprocess_step(self):
    method _save_checkpoint (line 636) | def _save_checkpoint(self):
    method load_checkpoint (line 699) | async def load_checkpoint(self):
    method _collect_metrics_from_samples (line 756) | def _collect_metrics_from_samples(self, batch, metrics):

FILE: verl/experimental/fully_async_policy/message_queue.py
  class MessageQueue (line 27) | class MessageQueue:
    method __init__ (line 32) | def __init__(self, config: DictConfig, max_queue_size: int = 1000):
    method put_sample (line 55) | async def put_sample(self, sample: Any) -> bool:
    method get_sample (line 85) | async def get_sample(self) -> Any | None:
    method get_queue_size (line 105) | async def get_queue_size(self) -> int:
    method get_statistics (line 110) | async def get_statistics(self) -> dict[str, Any]:
    method clear_queue (line 121) | async def clear_queue(self):
    method shutdown (line 128) | async def shutdown(self):
    method get_memory_usage (line 136) | async def get_memory_usage(self) -> dict:
    method put_validate (line 168) | async def put_validate(self, data):
    method get_validate (line 172) | async def get_validate(self):
  class MessageQueueClient (line 180) | class MessageQueueClient:
    method __init__ (line 183) | def __init__(self, queue_actor: Any):
    method put_sample (line 186) | async def put_sample(self, sample: Any) -> bool:
    method put_validate (line 191) | async def put_validate(self, data: Any) -> bool:
    method get_validate_sync (line 195) | def get_validate_sync(self) -> Any | None:
    method get_sample (line 198) | async def get_sample(self) -> Any | None:
    method get_queue_size (line 203) | async def get_queue_size(self) -> int:
    method get_statistics (line 208) | async def get_statistics(self) -> dict[str, Any]:
    method clear_queue (line 213) | async def clear_queue(self):
    method shutdown (line 218) | async def shutdown(self):
    method get_memory_usage (line 223) | async def get_memory_usage(self) -> dict:
    method get_sample_sync (line 228) | def get_sample_sync(self) -> Any | None:
    method get_statistics_sync (line 232) | def get_statistics_sync(self) -> dict[str, Any]:

FILE: verl/experimental/fully_async_policy/unittest/simple_streaming_demo.py
  class SimpleStreamingSystem (line 20) | class SimpleStreamingSystem:
    method __init__ (line 23) | def __init__(self, max_concurrent_tasks: int = 4):
    method data_stream (line 30) | async def data_stream(self):
    method add_data_stream (line 47) | async def add_data_stream(self, data_list: list[dict]):
    method _process_data_async (line 61) | async def _process_data_async(self, data_item: dict):
    method _submit_worker (line 85) | async def _submit_worker(self):
    method _consumer_worker (line 120) | async def _consumer_worker(self):
    method run_demo (line 140) | async def run_demo(self):
  function main (line 169) | async def main():

FILE: verl/experimental/one_step_off_policy/main_ppo.py
  class OneStepTaskRunner (line 35) | class OneStepTaskRunner:
    method run (line 36) | def run(self, config):
  function main (line 111) | def main(config):

FILE: verl/experimental/one_step_off_policy/ray_trainer.py
  class OneStepOffRayTrainer (line 48) | class OneStepOffRayTrainer(SeparateRayPPOTrainer):
    method __init__ (line 49) | def __init__(
    method _create_actor_rollout_classes (line 141) | def _create_actor_rollout_classes(self):
    method _init_models (line 151) | def _init_models(self):
    method _init_async_rollout_manager (line 169) | def _init_async_rollout_manager(self):
    method _create_continuous_iterator (line 188) | def _create_continuous_iterator(self):
    method _async_gen_next_batch (line 197) | async def _async_gen_next_batch(self, continuous_iterator):
    method _launch_individual_rewards (line 252) | def _launch_individual_rewards(batch, config, tokenizer):
    method fit (line 256) | async def fit(self):
    method fit_step (line 318) | async def fit_step(self, batch_data_future, continuous_iterator):
    method _fit_generate (line 383) | async def _fit_generate(self, batch_data_future, continuous_iterator):

FILE: verl/experimental/reward_loop/reward_loop.py
  function migrate_legacy_reward_impl (line 38) | def migrate_legacy_reward_impl(config):
  class RewardLoopWorker (line 92) | class RewardLoopWorker:
    method __init__ (line 108) | def __init__(self, config: DictConfig, reward_router_address: str = No...
    method _init_reward_fn (line 118) | def _init_reward_fn(self):
    method compute_score_batch (line 133) | async def compute_score_batch(self, data: DataProto) -> list[dict]:
    method compute_score (line 140) | async def compute_score(self, data: DataProto) -> dict:
    method _post_request (line 153) | async def _post_request(self, payload: dict, endpoint: str, max_retrie...
    method _preprocess_reward_inputs (line 193) | async def _preprocess_reward_inputs(self, data: DataProto) -> str:
    method compute_score_disrm (line 229) | async def compute_score_disrm(self, data: DataProto) -> dict:
  class RewardLoopManager (line 271) | class RewardLoopManager:
    method __init__ (line 277) | def __init__(self, config: DictConfig, rm_resource_pool: RayResourcePo...
    method _init_reward_loop_workers (line 289) | def _init_reward_loop_workers(self):
    method compute_rm_score (line 308) | def compute_rm_score(self, data: DataProto) -> DataProto:
    method _run_all (line 344) | def _run_all(self, tasks: list[asyncio.Task]):

FILE: verl/experimental/reward_loop/reward_manager/base.py
  class RewardManagerBase (line 33) | class RewardManagerBase(ABC):
    method __init__ (line 36) | def __init__(self, config: DictConfig, tokenizer: AutoTokenizer, compu...
    method init_class (line 50) | def init_class(cls, config: DictConfig, tokenizer: AutoTokenizer):
    method run_single (line 57) | async def run_single(self, data: DataProto):

FILE: verl/experimental/reward_loop/reward_manager/dapo.py
  class DAPORewardManager (line 24) | class DAPORewardManager(RewardManagerBase):
    method __init__ (line 27) | def __init__(self, config, tokenizer, compute_score, reward_router_add...
    method run_single (line 52) | async def run_single(self, data: DataProto) -> dict:

FILE: verl/experimental/reward_loop/reward_manager/gdpo.py
  class GDPORewardManager (line 24) | class GDPORewardManager(RewardManagerBase):
    method __init__ (line 27) | def __init__(self, config, tokenizer, compute_score, reward_router_add...
    method run_single (line 35) | async def run_single(self, data: DataProto) -> dict:

FILE: verl/experimental/reward_loop/reward_manager/limited.py
  class AsyncTokenBucket (line 32) | class AsyncTokenBucket:
    method __init__ (line 83) | def __init__(self, rate_limit: float, max_tokens: float = None):
    method acquire (line 90) | async def acquire(self, num_tokens: float = 1.0) -> None:
  class RateLimitedRewardManager (line 174) | class RateLimitedRewardManager(RewardManagerBase):
    method init_class (line 265) | def init_class(cls, config: DictConfig, tokenizer: AutoTokenizer):
    method __init__ (line 341) | def __init__(
    method _compute_reward (line 367) | async def _compute_reward(
    method run_single (line 398) | async def run_single(self, data: DataProto) -> dict:
    method __call__ (line 471) | def __call__(self, data: DataProto, return_dict: bool = False):

FILE: verl/experimental/reward_loop/reward_manager/naive.py
  class NaiveRewardManager (line 24) | class NaiveRewardManager(RewardManagerBase):
    method __init__ (line 27) | def __init__(self, config, tokenizer, compute_score, reward_router_add...
    method run_single (line 34) | async def run_single(self, data: DataProto) -> dict:

FILE: verl/experimental/reward_loop/reward_manager/registry.py
  function register (line 24) | def register(name: str) -> Callable[[type[RewardManagerBase]], type[Rewa...
  function get_reward_manager_cls (line 41) | def get_reward_manager_cls(name: str) -> type[RewardManagerBase]:

FILE: verl/experimental/reward_loop/reward_manager/remote.py
  class RewardComputeWorker (line 27) | class RewardComputeWorker:
    method __init__ (line 32) | def __init__(self, compute_score_fn):
    method compute_score (line 36) | def compute_score(self, **kwargs) -> dict:
  class RemoteRewardManager (line 41) | class RemoteRewardManager(RewardManagerBase):
    method __init__ (line 50) | def __init__(self, config, tokenizer, compute_score, reward_router_add...
    method choose_reward_worker (line 72) | def choose_reward_worker(self):
    method run_single (line 75) | async def run_single(self, data: DataProto) -> dict:

FILE: verl/experimental/reward_loop/reward_model.py
  class RewardModelManager (line 27) | class RewardModelManager:
    method __init__ (line 30) | def __init__(
    method _initialize_llm_servers (line 50) | def _initialize_llm_servers(self):
    method _initialize_router (line 87) | def _initialize_router(self):
    method get_router_address (line 100) | def get_router_address(self):
    method wake_up (line 103) | def wake_up(self):
    method sleep (line 107) | def sleep(self):
    method _run_all (line 111) | def _run_all(self, tasks: list[asyncio.Task]):

FILE: verl/experimental/reward_loop/router/inner_sglang_router.py
  function launch_router_process (line 30) | def launch_router_process(

FILE: verl/experimental/reward_loop/router/naive_router.py
  function _read_async_response (line 34) | async def _read_async_response(resp: aiohttp.ClientResponse) -> dict[str...
  function launch_router_process (line 51) | def launch_router_process(
  function run_router (line 77) | def run_router(router_ip: str, router_port: int, worker_urls: list[str]):
  class NaiveRouter (line 82) | class NaiveRouter:
    method __init__ (line 83) | def __init__(
    method _on_startup (line 115) | async def _on_startup(self):
    method _on_shutdown (line 128) | async def _on_shutdown(self):
    method _make_async_request (line 135) | async def _make_async_request(self, request: Request, endpoint: str):
    method _select_worker (line 175) | def _select_worker(self) -> str:
    method _release_worker (line 181) | def _release_worker(self, url: str) -> None:

FILE: verl/experimental/separation/engine_workers.py
  class DetachActorWorker (line 35) | class DetachActorWorker(ActorRolloutRefWorker):
    method __init__ (line 44) | def __init__(self, config: DictConfig, role: str):
    method _get_strategy_handlers (line 56) | def _get_strategy_handlers(self):
    method save_model_to_cpu (line 91) | def save_model_to_cpu(self, n):
    method restore_model_from_cpu (line 104) | def restore_model_from_cpu(self, n):
    method clear_cpu_model (line 121) | def clear_cpu_model(self, n):

FILE: verl/experimental/separation/ray_trainer.py
  class SeparateRayPPOTrainer (line 55) | class SeparateRayPPOTrainer(RayPPOTrainer):
    method __init__ (line 62) | def __init__(
    method init_workers (line 108) | def init_workers(self):
    method _init_resource_pools (line 128) | def _init_resource_pools(self):
    method _create_worker_classes (line 132) | def _create_worker_classes(self):
    method _create_actor_rollout_classes (line 138) | def _create_actor_rollout_classes(self):
    method _create_critic_class (line 141) | def _create_critic_class(self):
    method _create_reference_policy_class (line 171) | def _create_reference_policy_class(self):
    method _create_reward_model_class (line 183) | def _create_reward_model_class(self):
    method _init_worker_groups (line 193) | def _init_worker_groups(self):
    method _init_models (line 227) | def _init_models(self):
    method _init_reward_loop (line 254) | def _init_reward_loop(self):
    method _init_async_rollout_manager (line 266) | def _init_async_rollout_manager(self):
    method fit (line 269) | def fit(self):
    method fit_step (line 336) | def fit_step(self, batch_dict: Any = None):
    method _fit_prepare_step (line 378) | def _fit_prepare_step(self):
    method _fit_start_profile (line 383) | def _fit_start_profile(self):
    method _fit_get_batch (line 392) | def _fit_get_batch(self, batch_dict: dict) -> DataProto:
    method _fit_generate (line 399) | def _fit_generate(self, batch: DataProto = None) -> DataProto:
    method _fit_compute_reward (line 470) | def _fit_compute_reward(self, batch: DataProto) -> DataProto:
    method _fit_compute_log_prob (line 484) | def _fit_compute_log_prob(self, batch: DataProto) -> DataProto:
    method _fit_compute_ref_log_prob (line 535) | def _fit_compute_ref_log_prob(self, batch: DataProto) -> DataProto:
    method _fit_compute_critic (line 543) | def _fit_compute_critic(self, batch: DataProto) -> DataProto:
    method _fit_compute_advantage (line 551) | def _fit_compute_advantage(self, batch) -> DataProto:
    method _fit_update_critic (line 607) | def _fit_update_critic(self, batch: DataProto) -> DataProto:
    method _fit_update_actor (line 617) | def _fit_update_actor(self, batch: DataProto) -> DataProto:
    method _fit_update_weights (line 630) | def _fit_update_weights(self):
    method _fit_dump_data (line 637) | def _fit_dump_data(self, batch: DataProto):
    method _fit_validate (line 645) | def _fit_validate(self):
    method _fit_save_checkpoint (line 657) | def _fit_save_checkpoint(self):
    method _fit_stop_profile (line 684) | def _fit_stop_profile(self):
    method _fit_collect_metrics (line 700) | def _fit_collect_metrics(self, batch):
    method _fit_torch_memory (line 714) | def _fit_torch_memory(self):
    method _fit_experimental (line 723) | def _fit_experimental(self, batch):
    method _fit_postprocess_step (line 734) | def _fit_postprocess_step(self):

FILE: verl/experimental/separation/utils.py
  function create_resource_pool_manager (line 22) | def create_resource_pool_manager(config, roles: list) -> ResourcePoolMan...
  function create_role_worker_mapping (line 57) | def create_role_worker_mapping(config):

FILE: verl/experimental/vla/dp_rob.py
  class RobDataParallelPPOActor (line 39) | class RobDataParallelPPOActor(BasePPOActor):
    method __init__ (line 40) | def __init__(
    method process_tensor (line 57) | def process_tensor(self, tensor, pad_id):
    method generate_traj_mask (line 65) | def generate_traj_mask(self, end_step, traj_len):
    method apply_mask_with_grad_control (line 78) | def apply_mask_with_grad_control(self, log_probs, entropy, mask):
    method _forward_micro_batch (line 95) | def _forward_micro_batch(self, micro_batch, temperature) -> tuple[torc...
    method _forward_micro_batch_update (line 139) | def _forward_micro_batch_update(
    method _optimizer_step (line 165) | def _optimizer_step(self):
    method compute_log_prob (line 175) | def compute_log_prob(self, data: DataProto, calculate_entropy=False) -...
    method update_policy (line 231) | def update_policy(self, data: DataProto):

FILE: verl/experimental/vla/env_loop.py
  class EnvLoop (line 30) | class EnvLoop:
    method __init__ (line 34) | def __init__(self, env_wg: RayWorkerGroup, rollout_wg: RayWorkerGroup,...
    method generate_sequences (line 61) | def generate_sequences(self, prompts: DataProto, reset_future: asyncio...
    method run (line 80) | async def run(self, prompts: DataProto, reset_results: DataProto) -> D...
    method _restructure_obs_data (line 146) | def _restructure_obs_data(self, data_proto: DataProto) -> list[DataPro...
    method _collate_trajectories (line 164) | def _collate_trajectories(self, trajectories: dict, initial_state_ids:...

FILE: verl/experimental/vla/envs/action_utils.py
  function prepare_actions_simplevla (line 28) | def prepare_actions_simplevla(
  function prepare_actions (line 38) | def prepare_actions(
  function to_tensor (line 54) | def to_tensor(array: dict | torch.Tensor | np.ndarray | list | Any, devi...
  function tile_images (line 87) | def tile_images(images: list[np.ndarray | torch.Tensor], nrows: int = 1)...
  function put_text_on_image (line 155) | def put_text_on_image(image: np.ndarray, lines: list[str], max_width: in...
  function put_info_on_image (line 203) | def put_info_on_image(
  function list_of_dict_to_dict_of_list (line 224) | def list_of_dict_to_dict_of_list(
  function save_rollout_video (line 247) | def save_rollout_video(rollout_images: list[np.ndarray], output_dir: str...
  function resize_image (line 265) | def resize_image(img: np.ndarray, resize_size: tuple[int, int]) -> np.nd...
  function center_crop_image (line 298) | def center_crop_image(image: Image.Image) -> Image.Image:

FILE: verl/experimental/vla/envs/isaac_env/isaac_env.py
  class IsaacEnv (line 34) | class IsaacEnv(gym.Env):
    method __init__ (line 35) | def __init__(self, cfg, rank, world_size):
    method _init_env (line 70) | def _init_env(self, task_id=0):
    method _init_metrics (line 116) | def _init_metrics(self):
    method _reset_metrics (line 120) | def _reset_metrics(self, env_idx=None):
    method _record_metrics (line 134) | def _record_metrics(self, step_reward, terminations, infos):
    method reset (line 151) | def reset(self, env_idx: Optional[int | list[int] | np.ndarray] = None...
    method step (line 163) | def step(self, actions=None, critic_values=None):
    method chunk_step (line 204) | def chunk_step(self, chunk_actions, chunk_values=None):
    method _calc_step_reward (line 241) | def _calc_step_reward(self, reward):
    method _wrap_obs (line 249) | def _wrap_obs(self, raw_obs):
    method _extract_image_and_state (line 258) | def _extract_image_and_state(self, obs):
    method add_new_frames (line 282) | def add_new_frames(self, obs, plot_infos):
    method flush_video (line 291) | def flush_video(self, video_sub_dir: Optional[str] = None):
    method close (line 303) | def close(self):
    method load_state (line 308) | def load_state(self, state_buffer: bytes):
    method get_state (line 311) | def get_state(self):
    method reset_envs_to_state_ids (line 314) | def reset_envs_to_state_ids(self, state_ids_list, task_ids_list):

FILE: verl/experimental/vla/envs/libero_env/libero_env.py
  function patched_get_task_init_states (line 42) | def patched_get_task_init_states(self, i):
  class LiberoEnv (line 55) | class LiberoEnv(gym.Env):
    method __init__ (line 56) | def __init__(self, cfg, rank, world_size, stage_id: int = 0):
    method _compose_seed (line 89) | def _compose_seed(self, env_id: int, rollout_id: Optional[int] = None,...
    method elapsed_steps (line 103) | def elapsed_steps(self):
    method get_all_state_ids (line 106) | def get_all_state_ids(self):
    method _init_env (line 110) | def _init_env(self):
    method get_env_fns (line 114) | def get_env_fns(self):
    method get_env_fn_params (line 128) | def get_env_fn_params(self, env_idx=None):
    method _compute_total_num_group_envs (line 158) | def _compute_total_num_group_envs(self):
    method _init_task_and_trial_ids (line 169) | def _init_task_and_trial_ids(self):
    method _get_random_reset_state_ids (line 172) | def _get_random_reset_state_ids(self, num_reset_states):
    method get_reset_state_ids_all (line 176) | def get_reset_state_ids_all(self):
    method _get_ordered_reset_state_ids (line 185) | def _get_ordered_reset_state_ids(self, num_reset_states):
    method _get_task_and_trial_ids_from_reset_state_ids (line 193) | def _get_task_and_trial_ids_from_reset_state_ids(self, reset_state_ids):
    method _get_reset_states (line 214) | def _get_reset_states(self, env_idx):
    method _init_metrics (line 222) | def _init_metrics(self):
    method _reset_metrics (line 227) | def _reset_metrics(self, env_idx=None):
    method _record_metrics (line 243) | def _record_metrics(self, step_reward, terminations, infos):
    method _extract_image_and_state (line 254) | def _extract_image_and_state(self, obs):
    method _wrap_obs (line 267) | def _wrap_obs(self, obs_list):
    method _reconfigure (line 279) | def _reconfigure(self, reset_state_ids, env_idx):
    method reset (line 297) | def reset(
    method step (line 325) | def step(self, actions=None, critic_values=None):
    method chunk_step (line 363) | def chunk_step(self, chunk_actions, chunk_values=None):
    method _calc_step_reward (line 398) | def _calc_step_reward(self, terminations):
    method add_new_frames (line 408) | def add_new_frames(self, raw_obs, plot_infos):
    method flush_video (line 418) | def flush_video(self, video_sub_dir: Optional[str] = None):
    method reset_envs_to_state_ids (line 430) | def reset_envs_to_state_ids(self, state_ids_list, task_ids_list):
    method load_state (line 440) | def load_state(self, state_buffer: bytes):

FILE: verl/experimental/vla/envs/libero_env/utils.py
  function get_libero_image (line 24) | def get_libero_image(obs: dict[str, np.ndarray]) -> np.ndarray:
  function get_libero_wrist_image (line 39) | def get_libero_wrist_image(obs: dict[str, np.ndarray]) -> np.ndarray:
  function quat2axisangle (line 54) | def quat2axisangle(quat: np.ndarray) -> np.ndarray:
  function normalize_gripper_action (line 81) | def normalize_gripper_action(action: np.ndarray, binarize: bool = True) ...
  function invert_gripper_action (line 112) | def invert_gripper_action(action: np.ndarray) -> np.ndarray:

FILE: verl/experimental/vla/envs/libero_env/venv.py
  function _worker (line 35) | def _worker(
  class ReconfigureSubprocEnvWorker (line 121) | class ReconfigureSubprocEnvWorker(SubprocEnvWorker):
    method __init__ (line 122) | def __init__(self, env_fn: Callable[[], gym.Env], share_memory: bool =...
    method reconfigure_env_fn (line 143) | def reconfigure_env_fn(self, env_fn_param):
  class ReconfigureSubprocEnv (line 148) | class ReconfigureSubprocEnv(SubprocVectorEnv):
    method __init__ (line 149) | def __init__(self, env_fns: list[Callable[[], gym.Env]], **kwargs: Any...
    method reconfigure_env_fns (line 155) | def reconfigure_env_fns(self, env_fns, id=None):

FILE: verl/experimental/vla/fsdp_workers.py
  class RobActorRolloutRefWorker (line 52) | class RobActorRolloutRefWorker(ActorRolloutRefWorker):
    method _build_rollout (line 60) | def _build_rollout(self, trust_remote_code=False):
    method switch_to_rollout (line 111) | def switch_to_rollout(self):
    method switch_to_train (line 117) | def switch_to_train(self):
    method rollout_mode (line 122) | async def rollout_mode(self):
    method trainer_mode (line 168) | async def trainer_mode(self):
    method generate_sequences (line 201) | def generate_sequences(self, prompts: DataProto):
    method init_model (line 240) | def init_model(self):

FILE: verl/experimental/vla/main_ppo.py
  function calculate_reward (line 35) | def calculate_reward(data: DataProto, return_dict: bool = False) -> torc...
  function main (line 48) | def main(config):
  function main_task (line 84) | def main_task(config):

FILE: verl/experimental/vla/main_sac.py
  function calculate_reward (line 36) | def calculate_reward(data: DataProto, return_dict: bool = False) -> torc...
  function main (line 46) | def main(config):
  function main_task (line 59) | def main_task(config):

FILE: verl/experimental/vla/models/modules/mlp.py
  class MLP (line 19) | class MLP(nn.Module):
    method __init__ (line 36) | def __init__(
    method _get_activation (line 68) | def _get_activation(self, name: str):
    method init_weights (line 90) | def init_weights(self, m: nn.Module):
    method forward (line 122) | def forward(self, x):

FILE: verl/experimental/vla/models/openvla_oft/configuration_prismatic.py
  class PrismaticConfig (line 88) | class PrismaticConfig(PretrainedConfig):
    method __init__ (line 92) | def __init__(
  class OpenVLAConfig (line 145) | class OpenVLAConfig(PrismaticConfig):
    method __init__ (line 148) | def __init__(

FILE: verl/experimental/vla/models/openvla_oft/constants.py
  class NormalizationType (line 35) | class NormalizationType(str, Enum):
  function detect_robot_platform (line 67) | def detect_robot_platform():

FILE: verl/experimental/vla/models/openvla_oft/modeling_prismatic.py
  function unpack_tuple (line 61) | def unpack_tuple(fn: Callable[[Any], tuple[Any]]) -> Callable[[Any], Any]:
  function _ls_new_forward (line 72) | def _ls_new_forward(self, x: torch.Tensor) -> torch.Tensor:
  function ls_apply_patch (line 76) | def ls_apply_patch(ls_module: LayerScale):
  class PrismaticVisionBackbone (line 83) | class PrismaticVisionBackbone(nn.Module):
    method __init__ (line 91) | def __init__(
    method _create_featurizer (line 131) | def _create_featurizer(self, model_id: str, img_size: int, act_layer: ...
    method _patch_layer_scales (line 157) | def _patch_layer_scales(self) -> None:
    method get_num_patches (line 175) | def get_num_patches(self) -> int:
    method get_num_images_in_input (line 184) | def get_num_images_in_input(self) -> int:
    method set_num_images_in_input (line 193) | def set_num_images_in_input(self, num_images_in_input: int) -> None:
    method forward (line 202) | def forward(self, pixel_values: torch.Tensor) -> torch.Tensor:
  class PrismaticProjector (line 247) | class PrismaticProjector(nn.Module):
    method __init__ (line 248) | def __init__(self, use_fused_vision_backbone: bool, vision_dim: int, l...
    method forward (line 266) | def forward(self, img_patches: torch.Tensor) -> torch.Tensor:
  class PrismaticCausalLMOutputWithPast (line 283) | class PrismaticCausalLMOutputWithPast(ModelOutput):
  class PrismaticPreTrainedModel (line 296) | class PrismaticPreTrainedModel(PreTrainedModel):
    method _init_weights (line 305) | def _init_weights(self, module: nn.Module) -> None:
    method _supports_sdpa (line 328) | def _supports_sdpa(self) -> bool:
  class PrismaticForConditionalGeneration (line 333) | class PrismaticForConditionalGeneration(PrismaticPreTrainedModel):
    method __init__ (line 334) | def __init__(self, config: PrismaticConfig) -> None:
    method get_input_embeddings (line 379) | def get_input_embeddings(self) -> nn.Module:
    method set_input_embeddings (line 382) | def set_input_embeddings(self, value: nn.Module) -> None:
    method get_output_embeddings (line 385) | def get_output_embeddings(self) -> nn.Module:
    method set_output_embeddings (line 388) | def set_output_embeddings(self, new_embeddings: nn.Module) -> None:
    method get_decoder (line 391) | def get_decoder(self) -> nn.Module:
    method set_decoder (line 394) | def set_decoder(self, decoder: nn.Module) -> None:
    method tie_weights (line 397) | def tie_weights(self) -> None:
    method resize_token_embeddings (line 400) | def resize_token_embeddings(
    method _replace_input_embeddings (line 411) | def _replace_input_embeddings(self, input_embeddings, all_actions_mask...
    method _process_action_masks (line 447) | def _process_action_masks(self, labels):
    method _process_vision_features (line 454) | def _process_vision_features(self, pixel_values, language_embeddings=N...
    method _process_proprio_features (line 465) | def _process_proprio_features(self, projected_patch_embeddings, propri...
    method _build_multimodal_attention (line 477) | def _build_multimodal_attention(self, input_embeddings, projected_patc...
    method _build_multimodal_labels (line 502) | def _build_multimodal_labels(self, labels, projected_patch_embeddings):
    method prepare_inputs_for_generation (line 701) | def prepare_inputs_for_generation(
    method _reorder_cache (line 739) | def _reorder_cache(self, *args, **kwargs) -> Any:
    method _prepare_input_for_action_prediction_verl (line 742) | def _prepare_input_for_action_prediction_verl(self, input_ids, attenti...
    method _prepare_labels_for_action_prediction_verl (line 765) | def _prepare_labels_for_action_prediction_verl(self, labels, input_ids):
    method _verl_discrete_compute_logits (line 780) | def _verl_discrete_compute_logits(
    method forward (line 1091) | def forward(
  class OpenVLAForActionPrediction (line 1336) | class OpenVLAForActionPrediction(PrismaticForConditionalGeneration):
    method __init__ (line 1340) | def __init__(self, config: OpenVLAConfig) -> None:
    method _prepare_input_for_action_prediction (line 1351) | def _prepare_input_for_action_prediction(self, input_ids, attention_ma...
    method _prepare_labels_for_action_prediction (line 1374) | def _prepare_labels_for_action_prediction(self, labels, input_ids):
    method _unnormalize_actions (line 1389) | def _unnormalize_actions(self, normalized_actions, unnorm_key=None):
    method _run_diffusion_prediction (line 1410) | def _run_diffusion_prediction(
    method _regression_or_discrete_prediction (line 1496) | def _regression_or_discrete_prediction(
    method _verl_discrete_prediction (line 1563) | def _verl_discrete_prediction(
    method predict_action (line 1715) | def predict_action(
    method generate_action_verl (line 1833) | def generate_action_verl(
    method _check_unnorm_key (line 1976) | def _check_unnorm_key(norm_stats: dict[str, dict[str, Any]], unnorm_ke...
    method get_action_dim (line 1992) | def get_action_dim(self, unnorm_key: Optional[str] = None) -> int:
    method get_action_stats (line 1997) | def get_action_stats(self, unnorm_key: Optional[str] = None) -> dict[s...

FILE: verl/experimental/vla/models/openvla_oft/processing_prismatic.py
  function letterbox_pad_transform (line 40) | def letterbox_pad_transform(image: Image.Image, padding_fill_value: tupl...
  class PrismaticImageProcessor (line 49) | class PrismaticImageProcessor(ImageProcessingMixin):
    method __init__ (line 52) | def __init__(
    method apply_transform (line 145) | def apply_transform(self, img: Image.Image) -> torch.Tensor:
    method preprocess (line 164) | def preprocess(
    method __call__ (line 186) | def __call__(self, images: Image.Image | list[Image.Image], **kwargs) ...
  class PrismaticProcessor (line 192) | class PrismaticProcessor(ProcessorMixin):
    method __init__ (line 197) | def __init__(
    method __call__ (line 204) | def __call__(
    method batch_decode (line 236) | def batch_decode(
    method decode (line 250) | def decode(
    method model_input_names (line 265) | def model_input_names(self) -> list[str]:

FILE: verl/experimental/vla/models/openvla_oft/train_utils.py
  function get_current_action_mask (line 24) | def get_current_action_mask(token_ids):
  function get_next_actions_mask (line 41) | def get_next_actions_mask(token_ids):
  function compute_token_accuracy (line 58) | def compute_token_accuracy(predicted_token_ids, ground_truth_token_ids, ...
  function compute_actions_l1_loss (line 64) | def compute_actions_l1_loss(action_tokenizer, predicted_token_ids, groun...

FILE: verl/experimental/vla/models/pi0_torch/configuration_pi0_torch.py
  class PI0TorchConfig (line 18) | class PI0TorchConfig(PretrainedConfig):
    method __init__ (line 21) | def __init__(self, **kwargs):

FILE: verl/experimental/vla/models/pi0_torch/model/modeling_pi0.py
  function get_safe_dtype (line 30) | def get_safe_dtype(dtype: torch.dtype, device: str | torch.device) -> to...
  function create_sinusoidal_pos_embedding (line 40) | def create_sinusoidal_pos_embedding(
  function make_att_2d_masks (line 62) | def make_att_2d_masks(pad_masks: torch.Tensor, att_masks: torch.Tensor) ...
  class PI0Model (line 98) | class PI0Model(ModelMixin, ConfigMixin):
    method __init__ (line 124) | def __init__(
    method forward (line 161) | def forward(
    method sample_noise (line 209) | def sample_noise(self, shape: tuple[int, ...], device: torch.device | ...
    method embed_prefix (line 228) | def embed_prefix(
    method embed_suffix (line 304) | def embed_suffix(
    method sample_actions (line 384) | def sample_actions(
    method denoise_step (line 443) | def denoise_step(

FILE: verl/experimental/vla/models/pi0_torch/model/paligemma_with_expert.py
  function get_transformers_siglip_vision_config (line 38) | def get_transformers_siglip_vision_config() -> SiglipVisionConfig:
  class GemmaRMSNorm (line 54) | class GemmaRMSNorm(nn.Module):
    method __init__ (line 55) | def __init__(self, dim: int, eps: float = 1e-6, use_ada_rms_norm: bool...
    method _norm (line 65) | def _norm(self, x):
    method forward (line 68) | def forward(self, x, cond: torch.Tensor | None = None):
    method extra_repr (line 82) | def extra_repr(self):
  class SiglipVisionTransformer (line 89) | class SiglipVisionTransformer(nn.Module):
    method __init__ (line 90) | def __init__(self, config: SiglipVisionConfig):
    method forward (line 105) | def forward(
  class PaliGemmaMultiModalProjector (line 150) | class PaliGemmaMultiModalProjector(nn.Module):
    method __init__ (line 151) | def __init__(self, vision_hidden_size: int = 1152, projection_dim: int...
    method forward (line 155) | def forward(self, image_features: torch.Tensor) -> torch.Tensor:
  class RoPEEmbedding (line 161) | class RoPEEmbedding(nn.Module):
    method __init__ (line 168) | def __init__(self, dim: int, max_wavelength: int = 10_000, max_seq_len...
    method forward (line 193) | def forward(self, x: torch.Tensor, positions: torch.LongTensor) -> tor...
  class GemmaAttentionWithExpert (line 223) | class GemmaAttentionWithExpert(nn.Module):
    method __init__ (line 224) | def __init__(
    method forward (line 300) | def forward(
  class GemmaMLP (line 414) | class GemmaMLP(nn.Module):
    method __init__ (line 415) | def __init__(self, hidden_size: int = 1024, intermediate_size: int = 4...
    method forward (line 424) | def forward(self, x: torch.Tensor) -> torch.Tensor:
  class GemmaDecoderLayerWithExpert (line 430) | class GemmaDecoderLayerWithExpert(nn.Module):
    method __init__ (line 431) | def __init__(
    method gated_residual (line 496) | def gated_residual(self, x, y, gate):
    method forward (line 503) | def forward(
  class PaliGemmaWithExpertModel (line 574) | class PaliGemmaWithExpertModel(nn.Module):
    method __init__ (line 575) | def __init__(
    method embed_image (line 654) | def embed_image(self, image: torch.Tensor) -> torch.Tensor:
    method embed_language_tokens (line 661) | def embed_language_tokens(self, tokens: torch.Tensor) -> torch.Tensor:
    method forward (line 665) | def forward(

FILE: verl/experimental/vla/models/pi0_torch/modeling_pi0_torch.py
  function beta_schedule (line 43) | def beta_schedule(step, beta0, beta_min, T):
  class PI0ForActionPrediction (line 49) | class PI0ForActionPrediction(PreTrainedModel, SupportSACTraining):
    method __init__ (line 53) | def __init__(self, config: PI0TorchConfig):
    method _to (line 136) | def _to(self, device: torch.device | str):
    method forward (line 143) | def forward(
    method sample_actions (line 182) | def sample_actions(
    method from_pretrained (line 254) | def from_pretrained(cls, pretrained_model_name_or_path, *model_args, *...
    method freeze_vision_tower (line 272) | def freeze_vision_tower(self) -> None:
    method bc_loss (line 281) | def bc_loss(
    method _multi_heads_value (line 321) | def _multi_heads_value(
    method _cross_attention_pool_prefix (line 334) | def _cross_attention_pool_prefix(
    method _gaussian_log_prob (line 356) | def _gaussian_log_prob(
    method flow_sde_beta (line 366) | def flow_sde_beta(self) -> torch.Tensor:
    method _sample_actions_flow_sde (line 375) | def _sample_actions_flow_sde(
    method _build_kv_cache_from_prefix (line 460) | def _build_kv_cache_from_prefix(
    method sac_init (line 482) | def sac_init(self):
    method sac_forward_actor (line 494) | def sac_forward_actor(
    method sac_forward_critic (line 519) | def sac_forward_critic(
    method sac_get_critic_parameters (line 558) | def sac_get_critic_parameters(self) -> list[torch.nn.Parameter]:
    method sac_get_named_actor_parameters (line 564) | def sac_get_named_actor_parameters(self) -> list[tuple[str, torch.nn.P...
    method sac_forward_state_features (line 569) | def sac_forward_state_features(
    method sac_update_target_network (line 583) | def sac_update_target_network(self, tau: float):

FILE: verl/experimental/vla/models/pi0_torch/pi0_utils.py
  class Normalize (line 26) | class Normalize:
    method __init__ (line 34) | def __init__(self, stats: dict[str, Any], *, use_quantiles: bool = Fal...
    method to (line 54) | def to(self, device: torch.device | str) -> None:
    method __call__ (line 62) | def __call__(self, x: torch.Tensor) -> torch.Tensor:
  class Unnormalize (line 72) | class Unnormalize:
    method __init__ (line 73) | def __init__(self, stats, *, use_quantiles: bool = False):
    method to (line 85) | def to(self, device: torch.device | str) -> None:
    method __call__ (line 93) | def __call__(self, x: torch.Tensor) -> torch.Tensor:
  class DeltaActions (line 103) | class DeltaActions:
    method __init__ (line 106) | def __init__(self):
    method to (line 110) | def to(self, device: torch.device | str) -> None:
    method __call__ (line 113) | def __call__(self, data: dict[str, Any]) -> dict[str, Any]:
  class AbsoluteActions (line 125) | class AbsoluteActions:
    method __init__ (line 128) | def __init__(self):
    method to (line 132) | def to(self, device: torch.device | str) -> None:
    method __call__ (line 135) | def __call__(self, data: dict[str, Any]) -> dict[str, Any]:
  class AlohaInputs (line 147) | class AlohaInputs:
    method __init__ (line 150) | def __init__(self, adapt_to_pi: bool = True) -> None:
    method to (line 154) | def to(self, device: torch.device | str) -> None:
    method _gripper_from_angular_inv (line 157) | def _gripper_from_angular_inv(self, value: torch.Tensor) -> torch.Tensor:
    method _gripper_to_angular (line 162) | def _gripper_to_angular(self, value: torch.Tensor) -> torch.Tensor:
    method _encode_actions_inv (line 184) | def _encode_actions_inv(self, actions: torch.Tensor) -> torch.Tensor:
    method _decode_state (line 190) | def _decode_state(self, state: torch.Tensor) -> torch.Tensor:
    method _decode_aloha (line 198) | def _decode_aloha(self, state: torch.Tensor) -> torch.Tensor:
    method __call__ (line 204) | def __call__(self, data: dict[str, Any]) -> dict[str, Any]:
    method _encode_actions_inv_batch (line 218) | def _encode_actions_inv_batch(self, actions: torch.Tensor) -> torch.Te...
    method _decode_state_batch (line 224) | def _decode_state_batch(self, state: torch.Tensor) -> torch.Tensor:
    method call_batch (line 230) | def call_batch(self, data: dict[str, Any]) -> dict[str, Any]:
  class AlohaOutputs (line 240) | class AlohaOutputs:
    method __init__ (line 243) | def __init__(self, original_action_dim: int, adapt_to_pi: bool = True):
    method to (line 255) | def to(self, device: torch.device | str) -> None:
    method _gripper_from_angular (line 258) | def _gripper_from_angular(self, value: torch.Tensor) -> torch.Tensor:
    method _encode_actions (line 270) | def _encode_actions(self, actions: torch.Tensor) -> torch.Tensor:
    method __call__ (line 277) | def __call__(self, data: dict[str, Any]) -> dict[str, Any]:
    method _encode_actions_batch (line 283) | def _encode_actions_batch(self, actions: torch.Tensor) -> torch.Tensor:
    method call_batch (line 289) | def call_batch(self, data: dict[str, Any]) -> dict[str, Any]:
  class PadStatesAndActions (line 294) | class PadStatesAndActions:
    method __init__ (line 297) | def __init__(self, action_dim: int) -> None:
    method _pad_to_dim (line 300) | def _pad_to_dim(self, x: torch.Tensor, target_dim: int, axis: int = -1...
    method __call__ (line 312) | def __call__(self, data: dict[str, Any]) -> dict[str, Any]:
  function _normalize (line 319) | def _normalize(x: torch.Tensor, min_val: float, max_val: float) -> torch...
  function _unnormalize (line 323) | def _unnormalize(x: torch.Tensor, min_val: float, max_val: float) -> tor...
  function resize_with_pad (line 327) | def resize_with_pad(img: torch.Tensor, width: int, height: int, pad_valu...
  class ImageTransform (line 366) | class ImageTransform:
    method __init__ (line 367) | def __init__(
    method __call__ (line 397) | def __call__(self, data: dict[str, torch.Tensor]) -> tuple[list[torch....
    method call_batch (line 437) | def call_batch(self, data: dict[str, torch.Tensor]) -> tuple[list[torc...
  class PromptTokenizerTransform (line 483) | class PromptTokenizerTransform:
    method __init__ (line 484) | def __init__(self, max_length: int, discrete_state_input: bool = False...
    method __call__ (line 489) | def __call__(self, data: dict[str, Any], tokenizer) -> tuple[torch.Ten...
    method call_batch (line 528) | def call_batch(self, data: dict[str, Any], tokenizer) -> tuple[torch.T...

FILE: verl/experimental/vla/models/pi0_torch/policy/base.py
  class Pi0Input (line 20) | class Pi0Input(ABC):
    method __init__ (line 21) | def __init__(self):
    method from_env_obs (line 42) | def from_env_obs(cls, env_obs) -> "Pi0Input": ...
  class Pi0Output (line 45) | class Pi0Output:
    method __init__ (line 46) | def __init__(self):
    method from_model_output (line 51) | def from_model_output(cls, model_output) -> "Pi0Output": ...

FILE: verl/experimental/vla/models/pi0_torch/policy/libero_policy.py
  class LiberoPi0Input (line 27) | class LiberoPi0Input(Pi0Input):
    method from_env_obs (line 30) | def from_env_obs(cls, env_obs: DataProto) -> "LiberoPi0Input":
  class LiberoPi0Output (line 68) | class LiberoPi0Output(Pi0Output):
    method from_model_output (line 71) | def from_model_output(cls, model_output: dict) -> "LiberoPi0Output":

FILE: verl/experimental/vla/models/register_vla_models.py
  function register_openvla_oft (line 34) | def register_openvla_oft() -> None:
  function register_pi0_torch_model (line 47) | def register_pi0_torch_model() -> None:
  function register_vla_models (line 58) | def register_vla_models() -> None:

FILE: verl/experimental/vla/naive_rollout_rob.py
  function pad_sequence_to_length (line 45) | def pad_sequence_to_length(tensors, max_seq_len, pad_token_id, left_pad=...
  function process_input (line 58) | def process_input(task_descriptions, images_and_states, processor):
  class NaiveRolloutRob (line 112) | class NaiveRolloutRob(BaseRollout):
    method __init__ (line 113) | def __init__(
    method _generate_one_step (line 136) | def _generate_one_step(self, prompts: dict, do_sample, temperature, ma...
    method generate_sequences (line 181) | def generate_sequences(self, prompts: DataProto) -> DataProto:
    method update_weights (line 197) | async def update_weights(self, weights_iterator, **kwargs):
    method release (line 214) | async def release(self):
    method resume (line 221) | async def resume(self, **kwargs):

FILE: verl/experimental/vla/prepare_libero_dataset.py
  function patched_get_task_init_states (line 29) | def patched_get_task_init_states(self, i):
  function compute_total_num_group_envs (line 42) | def compute_total_num_group_envs(task_suite: Benchmark):
  function build_dataset_for_suite (line 55) | def build_dataset_for_suite(task_suite_name: str, local_save_dir: str):
  function resolve_task_suites (line 160) | def resolve_task_suites(task_suite_name: str) -> list[str]:

FILE: verl/experimental/vla/rob_ray_trainer.py
  function compute_response_mask (line 51) | def compute_response_mask(config, data: DataProto) -> torch.Tensor:
  function flatten_trajectories (line 85) | def flatten_trajectories(data: DataProto) -> DataProto:
  class RobRayPPOTrainer (line 105) | class RobRayPPOTrainer(RayPPOTrainer):
    method _start_profiling (line 113) | def _start_profiling(self, do_profile: bool) -> None:
    method _stop_profiling (line 119) | def _stop_profiling(self, do_profile: bool) -> None:
    method init_workers (line 125) | def init_workers(self):
    method _get_gen_batch (line 195) | def _get_gen_batch(self, batch: DataProto) -> DataProto:
    method _reset_envs (line 206) | def _reset_envs(self, gen_batch: DataProto) -> asyncio.Future:
    method fit (line 213) | def fit(self):
    method _validate (line 561) | def _validate(self):

FILE: verl/experimental/vla/sac/base.py
  class SupportSACTraining (line 23) | class SupportSACTraining:
    method sac_init (line 39) | def sac_init(self):
    method sac_get_critic_parameters (line 42) | def sac_get_critic_parameters(self) -> list[torch.nn.Parameter]:
    method sac_get_named_actor_parameters (line 51) | def sac_get_named_actor_parameters(self) -> list[tuple[str, torch.nn.P...
    method sac_forward_critic (line 60) | def sac_forward_critic(
    method sac_forward_actor (line 85) | def sac_forward_actor(
    method sac_forward_state_features (line 106) | def sac_forward_state_features(self, s: dict[str, torch.Tensor]) -> Any:
    method bc_loss (line 122) | def bc_loss(
    method sac_update_target_network (line 132) | def sac_update_target_network(self, tau: float):
  class BaseSACActor (line 142) | class BaseSACActor(ABC):
    method update_policy (line 144) | def update_policy(self, data: DataProto) -> dict:

FILE: verl/experimental/vla/sac/naive_rollout_pi05.py
  class PI0RolloutRob (line 37) | class PI0RolloutRob(NaiveRolloutRob):
    method __init__ (line 38) | def __init__(
    method generate_sequences (line 55) | def generate_sequences(self, prompts: DataProto) -> DataProto:

FILE: verl/experimental/vla/sac/replay_pool.py
  class _DualPoolState (line 28) | class _DualPoolState:
  class SACReplayPool (line 37) | class SACReplayPool:
    method __init__ (line 47) | def __init__(
    method add_batch (line 65) | def add_batch(self, batch: TensorDict, task_ids: Sequence[Any]):
    method sample_batch (line 110) | def sample_batch(
    method insert_and_resample (line 193) | def insert_and_resample(
    method save (line 203) | def save(self, directory: str):
    method load (line 242) | def load(self, directory: str):
    method from_path (line 275) | def from_path(
    method _insert_block_to_pool (line 299) | def _insert_block_to_pool(
    method _get_or_create_task_pool (line 331) | def _get_or_create_task_pool(self, task_id: str, sample: TensorDict) -...
    method _extract_positive_mask (line 361) | def _extract_positive_mask(self, batch: TensorDict) -> torch.Tensor:
    method _pad_sampled_batch (line 367) | def _pad_sampled_batch(self, sampled_batch: TensorDict, target_batch_s...
    method _index_select_batch (line 389) | def _index_select_batch(self, batch: TensorDict, idx: torch.Tensor) ->...
    method _sample_from_task_

[
  {
    "path": ".gemini/config.yaml",
    "chars": 207,
    "preview": "have_fun: false\ncode_review:\n  disable: false\n  comment_severity_threshold: HIGH\n  max_review_comments: -1\n  pull_reques"
  },
  {
    "path": ".git-blame-ignore-revs",
    "chars": 556,
    "preview": "# Local uasge: git config blame.ignoreRevsFile .git-blame-ignore-revs\n\n# [dev] feat: immigrate from yapf & pylint to ruf"
  },
  {
    "path": ".github/CODEOWNERS",
    "chars": 1347,
    "preview": "/docs @eric-haibin-lin @zhaochenyang20 @hongpeng-guo\n/docs/amd_tutorial @yushengsu-thu\n/docs/slang_multiturn @zhaochenya"
  },
  {
    "path": ".github/ISSUE_TEMPLATE/bug-report.yml",
    "chars": 2301,
    "preview": "# modified from https://github.com/huggingface/transformers/blob/main/.github/ISSUE_TEMPLATE/bug-report.yml?plain=1\nname"
  },
  {
    "path": ".github/ISSUE_TEMPLATE/config.yml",
    "chars": 40,
    "preview": "blank_issues_enabled: true\nversion: 0.1\n"
  },
  {
    "path": ".github/ISSUE_TEMPLATE/feature-request.yml",
    "chars": 1203,
    "preview": "# modified from https://github.com/huggingface/transformers/blob/main/.github/ISSUE_TEMPLATE/feature-request.yml?plain=1"
  },
  {
    "path": ".github/PULL_REQUEST_TEMPLATE.md",
    "chars": 2906,
    "preview": "### What does this PR do?\n\n> Add **concise** overview of what this PR aims to achieve or accomplish. Reference related G"
  },
  {
    "path": ".github/dependabot.yml",
    "chars": 245,
    "preview": "## Enabled the dependabot to check the dependencies of the project\n## Dependabot will open pull requests to update depen"
  },
  {
    "path": ".github/workflows/README.md",
    "chars": 2406,
    "preview": "### Adding a New Workflow\n\nWhen adding a new workflow for continuous integration (CI), you have two runner options: a fi"
  },
  {
    "path": ".github/workflows/check-pr-title.yml",
    "chars": 2494,
    "preview": "# # Tests layout\n\n# Each folder under tests/ corresponds to a test category for a sub-namespace in verl. For instance:\n#"
  },
  {
    "path": ".github/workflows/cpu_unit_tests.yml",
    "chars": 4619,
    "preview": "# # Tests layout\n\n# Each folder under tests/ corresponds to a test category for a sub-namespace in verl. For instance:\n#"
  },
  {
    "path": ".github/workflows/doc.yml",
    "chars": 4196,
    "preview": "# # Tests layout\n\n# Each folder under tests/ corresponds to a test category for a sub-namespace in verl. For instance:\n#"
  },
  {
    "path": ".github/workflows/docker-build-ascend-a2.yml",
    "chars": 2926,
    "preview": "name: docker-build-ascend-a2\n\non:\n  workflow_dispatch:\n  push:\n    branches: [\"main\"]\n    paths:\n      - \"docker/ascend/"
  },
  {
    "path": ".github/workflows/docker-build-ascend-a3.yml",
    "chars": 2926,
    "preview": "name: docker-build-ascend-a3\n\non:\n  workflow_dispatch:\n  push:\n    branches: [\"main\"]\n    paths:\n      - \"docker/ascend/"
  },
  {
    "path": ".github/workflows/e2e_ascend.yml",
    "chars": 6483,
    "preview": "# # Tests layout\n\n# Each folder under tests/ corresponds to a test category for a sub-namespace in verl. For instance:\n#"
  },
  {
    "path": ".github/workflows/e2e_fully_async_policy.yml",
    "chars": 6411,
    "preview": "# # Tests layout\n\n# Each folder under tests/ corresponds to a test category for a sub-namespace in verl. For instance:\n#"
  },
  {
    "path": ".github/workflows/e2e_fully_async_policy_ascend.yml",
    "chars": 6352,
    "preview": "# # Tests layout\n\n# Each folder under tests/ corresponds to a test category for a sub-namespace in verl. For instance:\n#"
  },
  {
    "path": ".github/workflows/e2e_one_step_off_policy.yml",
    "chars": 6463,
    "preview": "# # Tests layout\n\n# Each folder under tests/ corresponds to a test category for a sub-namespace in verl. For instance:\n#"
  },
  {
    "path": ".github/workflows/e2e_one_step_off_policy_ascend.yml",
    "chars": 6416,
    "preview": "# # Tests layout\n\n# Each folder under tests/ corresponds to a test category for a sub-namespace in verl. For instance:\n#"
  },
  {
    "path": ".github/workflows/e2e_ppo_grpo_trainer_trtllm.yml",
    "chars": 12297,
    "preview": "# # Tests layout\n\n# Each folder under tests/ corresponds to a test category for a sub-namespace in verl. For instance:\n#"
  },
  {
    "path": ".github/workflows/e2e_ppo_trainer.yml",
    "chars": 2196,
    "preview": "name: e2e_ppo_trainer\n\non:\n  # Trigger the workflow on push or pull request,\n  # but only for the main branch\n  # For pu"
  },
  {
    "path": ".github/workflows/e2e_ppo_trainer_megatron_sglang.yml",
    "chars": 8541,
    "preview": "# # Tests layout\n\n# Each folder under tests/ corresponds to a test category for a sub-namespace in verl. For instance:\n#"
  },
  {
    "path": ".github/workflows/e2e_ppo_trainer_megatron_sglang_2.yml",
    "chars": 8679,
    "preview": "# # Tests layout\n\n# Each folder under tests/ corresponds to a test category for a sub-namespace in verl. For instance:\n#"
  },
  {
    "path": ".github/workflows/e2e_ppo_trainer_megatron_vllm.yml",
    "chars": 9674,
    "preview": "# # Tests layout\n\n# Each folder under tests/ corresponds to a test category for a sub-namespace in verl. For instance:\n#"
  },
  {
    "path": ".github/workflows/e2e_ppo_trainer_megatron_vllm_2.yml",
    "chars": 16565,
    "preview": "# # Tests layout\n\n# Each folder under tests/ corresponds to a test category for a sub-namespace in verl. For instance:\n#"
  },
  {
    "path": ".github/workflows/e2e_ppo_trainer_megatron_vllm_2_ascend.yml",
    "chars": 11103,
    "preview": "# # Tests layout\n\n# Each folder under tests/ corresponds to a test category for a sub-namespace in verl. For instance:\n#"
  },
  {
    "path": ".github/workflows/e2e_ppo_trainer_veomni_vllm.yml",
    "chars": 5943,
    "preview": "# # Tests layout\n\n# Each folder under tests/ corresponds to a test category for a sub-namespace in verl. For instance:\n#"
  },
  {
    "path": ".github/workflows/e2e_sft_llm.yml",
    "chars": 6075,
    "preview": "# # Tests layout\n\n# Each folder under tests/ corresponds to a test category for a sub-namespace in verl. For instance:\n#"
  },
  {
    "path": ".github/workflows/e2e_sft_llm_ascend.yml",
    "chars": 7299,
    "preview": "# # Tests layout\n\n# Each folder under tests/ corresponds to a test category for a sub-namespace in verl. For instance:\n#"
  },
  {
    "path": ".github/workflows/e2e_sft_vlm.yml",
    "chars": 5013,
    "preview": "# # Tests layout\n\n# Each folder under tests/ corresponds to a test category for a sub-namespace in verl. For instance:\n#"
  },
  {
    "path": ".github/workflows/gpu_unit_tests.yml",
    "chars": 5882,
    "preview": "# # Tests layout\n\n# Each folder under tests/ corresponds to a test category for a sub-namespace in verl. For instance:\n#"
  },
  {
    "path": ".github/workflows/model.yml",
    "chars": 7532,
    "preview": "# # Tests layout\n\n# Each folder under tests/ corresponds to a test category for a sub-namespace in verl. For instance:\n#"
  },
  {
    "path": ".github/workflows/model_ascend.yml",
    "chars": 5358,
    "preview": "# # Tests layout\n\n# Each folder under tests/ corresponds to a test category for a sub-namespace in verl. For instance:\n#"
  },
  {
    "path": ".github/workflows/nightly_ascend.yml",
    "chars": 6970,
    "preview": "# # Tests layout\n\n# Each folder under tests/ corresponds to a test category for a sub-namespace in verl. For instance:\n#"
  },
  {
    "path": ".github/workflows/npu_unit_tests.yml",
    "chars": 5686,
    "preview": "# # Tests layout\n\n# Each folder under tests/ corresponds to a test category for a sub-namespace in verl. For instance:\n#"
  },
  {
    "path": ".github/workflows/pre-commit.yml",
    "chars": 1204,
    "preview": "# c.f. https://github.com/pre-commit/action?tab=readme-ov-file#using-this-action\nname: pre-commit\n\n# No need to avoid / "
  },
  {
    "path": ".github/workflows/precommit-autofix.yml",
    "chars": 1257,
    "preview": "name: scheduled pre-commit autofix\n\non:\n  schedule:\n    # Every hour\n    - cron: \"0 * * * *\"\n  workflow_dispatch:\n\npermi"
  },
  {
    "path": ".github/workflows/reward_model_sglang.yml",
    "chars": 5563,
    "preview": "# # Tests layout\n\n# Each folder under tests/ corresponds to a test category for a sub-namespace in verl. For instance:\n#"
  },
  {
    "path": ".github/workflows/reward_model_vllm.yml",
    "chars": 5497,
    "preview": "# # Tests layout\n\n# Each folder under tests/ corresponds to a test category for a sub-namespace in verl. For instance:\n#"
  },
  {
    "path": ".github/workflows/reward_model_vllm_ascend.yml",
    "chars": 4728,
    "preview": "# # Tests layout\n\n# Each folder under tests/ corresponds to a test category for a sub-namespace in verl. For instance:\n#"
  },
  {
    "path": ".github/workflows/sanity.yml",
    "chars": 4736,
    "preview": "# # Tests layout\n\n# Each folder under tests/ corresponds to a test category for a sub-namespace in verl. For instance:\n#"
  },
  {
    "path": ".github/workflows/scorecard.yml",
    "chars": 2759,
    "preview": "# This workflow uses actions that are not certified by GitHub. They are provided\n# by a third-party and are governed by "
  },
  {
    "path": ".github/workflows/secrets_scan.yml",
    "chars": 486,
    "preview": "on:\n  push:\n    branches:\n      - main\n      - v0.*\n  pull_request:\n\npermissions:\n  contents: read\n\njobs:\n  test:\n    ru"
  },
  {
    "path": ".github/workflows/sgl.yml",
    "chars": 6170,
    "preview": "# # Tests layout\n\n# Each folder under tests/ corresponds to a test category for a sub-namespace in verl. For instance:\n#"
  },
  {
    "path": ".github/workflows/type-coverage-check.yml",
    "chars": 934,
    "preview": "name: Type Annotation and Docstring Coverage\n\non:\n  pull_request:\n    paths:\n      - '**/*.py'\n      - '.github/workflow"
  },
  {
    "path": ".github/workflows/vllm.yml",
    "chars": 6747,
    "preview": "# # Tests layout\n\n# Each folder under tests/ corresponds to a test category for a sub-namespace in verl. For instance:\n#"
  },
  {
    "path": ".gitignore",
    "chars": 1376,
    "preview": "**/*.pt\n**/checkpoints\n**/wget-log\n**/_build/\n**/*.ckpt\n**/outputs\n**/*.tar.gz\n**/playground\n**/wandb\n\n/pyrightconfig.js"
  },
  {
    "path": ".gitmodules",
    "chars": 91,
    "preview": "[submodule \"recipe\"]\n\tpath = recipe\n\turl = https://github.com/verl-project/verl-recipe.git\n"
  },
  {
    "path": ".pre-commit-config.yaml",
    "chars": 1290,
    "preview": "repos:\n  - repo: https://github.com/astral-sh/ruff-pre-commit\n    rev: \"v0.12.2\"\n    hooks:\n      - id: ruff\n        arg"
  },
  {
    "path": ".readthedocs.yaml",
    "chars": 333,
    "preview": "# Read the Docs configuration file\n# See https://docs.readthedocs.io/en/stable/config-file/v2.html for details\n\nversion:"
  },
  {
    "path": "CONTRIBUTING.md",
    "chars": 3774,
    "preview": "# Contributing to verl\n\nThank you for considering a contribution to verl! We welcome contributions of any kind - bug fix"
  },
  {
    "path": "LICENSE",
    "chars": 11358,
    "preview": "\n                                 Apache License\n                           Version 2.0, January 2004\n                  "
  },
  {
    "path": "Notice.txt",
    "chars": 57,
    "preview": "Copyright 2023-2024 Bytedance Ltd. and/or its affiliates "
  },
  {
    "path": "README.md",
    "chars": 36359,
    "preview": "<div align=\"center\">\n 👋 Hi, everyone!\n    verl is a RL training library initiated by <b>ByteDance Seed team</b> and main"
  },
  {
    "path": "docker/Dockerfile.isaaclab230",
    "chars": 5325,
    "preview": "\n#FROM nvcr.nju.edu.cn/nvidia/isaac-lab:2.3.0\nFROM isaac-lab-base:latest\n\nENV ACCEPT_EULA=Y\nENTRYPOINT []\n\n# desktop\nRUN"
  },
  {
    "path": "docker/Dockerfile.stable.sglang",
    "chars": 2279,
    "preview": "# sgl059\n\nFROM lmsysorg/sglang:v0.5.9\n\nARG PIP_NO_CACHE_DIR=1\n\nRUN pip install pybind11\n\nRUN pip install nvidia-mathdx\n\n"
  },
  {
    "path": "docker/Dockerfile.stable.trtllm",
    "chars": 2680,
    "preview": "# Base image from NGC TensorRT-LLM, which includes a pre-installed TensorRT-LLM.\n# For available images, visit: https://"
  },
  {
    "path": "docker/Dockerfile.stable.vllm",
    "chars": 4058,
    "preview": "# vllm017\n\nFROM nvidia/cuda:12.9.1-devel-ubuntu22.04\n\nARG DEBIAN_FRONTEND=noninteractive\nARG PIP_NO_CACHE_DIR=1\n\nRUN apt"
  },
  {
    "path": "docker/README.md",
    "chars": 2697,
    "preview": "# Dockerfiles of verl\n\nWe provide pre-built Docker images for quick setup. And from this version, we utilize a new image"
  },
  {
    "path": "docker/ascend/Dockerfile.ascend.sglang_8.3.rc1_a2",
    "chars": 3760,
    "preview": "# Pull base image\nFROM swr.cn-south-1.myhuaweicloud.com/ascendhub/cann:8.3.rc1-910b-ubuntu22.04-py3.11\n\nARG ASCEND_CANN_"
  },
  {
    "path": "docker/ascend/Dockerfile.ascend.sglang_8.3.rc1_a3",
    "chars": 3545,
    "preview": "# Pull base image\nFROM swr.cn-south-1.myhuaweicloud.com/ascendhub/cann:8.3.rc1-a3-ubuntu22.04-py3.11\n\nARG ASCEND_CANN_PA"
  },
  {
    "path": "docker/ascend/Dockerfile.ascend_8.2.rc1_a2",
    "chars": 2578,
    "preview": "# Pull base image\nFROM swr.cn-south-1.myhuaweicloud.com/ascendhub/cann:8.2.rc1-910b-ubuntu22.04-py3.11\n\n# Prepare requir"
  },
  {
    "path": "docker/ascend/Dockerfile.ascend_8.2.rc1_a3",
    "chars": 2576,
    "preview": "# Pull base image\nFROM swr.cn-south-1.myhuaweicloud.com/ascendhub/cann:8.2.rc1-a3-ubuntu22.04-py3.11\n\n# Prepare required"
  },
  {
    "path": "docker/ascend/Dockerfile.ascend_8.3.rc1_a2",
    "chars": 2797,
    "preview": "# Pull base image\nFROM swr.cn-south-1.myhuaweicloud.com/ascendhub/cann:8.3.rc1-910b-ubuntu22.04-py3.11\n\n# Prepare requir"
  },
  {
    "path": "docker/ascend/Dockerfile.ascend_8.3.rc1_a3",
    "chars": 2795,
    "preview": "# Pull base image\nFROM swr.cn-south-1.myhuaweicloud.com/ascendhub/cann:8.3.rc1-a3-ubuntu22.04-py3.11\n\n# Prepare required"
  },
  {
    "path": "docker/ascend/Dockerfile.ascend_8.5.0_a2",
    "chars": 2822,
    "preview": "# Pull base image\nFROM swr.cn-south-1.myhuaweicloud.com/ascendhub/cann:8.5.0-910b-ubuntu22.04-py3.11\n\nARG SOC_VERSION=\"a"
  },
  {
    "path": "docker/ascend/Dockerfile.ascend_8.5.0_a3",
    "chars": 2824,
    "preview": "# Pull base image\nFROM swr.cn-south-1.myhuaweicloud.com/ascendhub/cann:8.5.0-a3-ubuntu22.04-py3.11\n\nARG SOC_VERSION=\"asc"
  },
  {
    "path": "docker/aws/Dockerfile.extention.awsefa",
    "chars": 2105,
    "preview": "# Base Image support aws EFA\n# Build Image with frameworks based on this\nFROM verlai/verl:app-verl0.6-transformers4.56.1"
  },
  {
    "path": "docker/aws/Dockerfile.ngc.vllm0.8.sagemaker",
    "chars": 1791,
    "preview": "# Using a pre-built image from AWS DLC which contains the current version of python (3.10) and supported cuda version (1"
  },
  {
    "path": "docker/rocm/Apptainerfile.rocm",
    "chars": 1519,
    "preview": "Bootstrap: docker\n\n# Support - Traing: fsdp; Inference: vllm\n# FROM: rocm/vllm:rocm6.2_mi300_ubuntu20.04_py3.9_vllm_0.6."
  },
  {
    "path": "docker/rocm/Dockerfile.rocm",
    "chars": 10458,
    "preview": "# FROM \"compute-artifactory.amd.com:5000/rocm-plus-docker/framework/compute-rocm-rel-6.4:94_ubuntu22.04_py3.10_pytorch_r"
  },
  {
    "path": "docker/rocm/Dockerfile.rocm7",
    "chars": 4795,
    "preview": "# default base image\nARG REMOTE_VLLM=\"1\"\nARG COMMON_WORKDIR=/app\nARG BASE_IMAGE=rocm/vllm-dev:base\n\nFROM ${BASE_IMAGE} A"
  },
  {
    "path": "docker/rocm/Dockerfile.rocm_verl-0.3.0.post1",
    "chars": 1499,
    "preview": "#  Build the docker in the repo dir:\n# docker build -f docker/Dockerfile.rocm -t verl-rocm:03.04.2015 .\n# docker images "
  },
  {
    "path": "docker/rocm/Dockerfile.rocm_verl-0.4.1",
    "chars": 10480,
    "preview": "# FROM \"compute-artifactory.amd.com:5000/rocm-plus-docker/framework/compute-rocm-rel-6.4:94_ubuntu22.04_py3.10_pytorch_r"
  },
  {
    "path": "docker/verl0.4-cu124-torch2.6-fa2.7.4/Dockerfile.app.sglang.vllm.mcore0.12",
    "chars": 1926,
    "preview": "# Start from the verl base image\n# Dockerfile.base\nFROM verlai/verl:base-verl0.4-cu124-cudnn9.8-torch2.6-fa2.7.4\n\n# Defi"
  },
  {
    "path": "docker/verl0.4-cu124-torch2.6-fa2.7.4/Dockerfile.app.sglang.vllm.mcore0.12.deepep",
    "chars": 3447,
    "preview": "# Start from the verl base image\n# Dockerfile.base\nFROM verlai/verl:base-verl0.4-cu124-cudnn9.8-torch2.6-fa2.7.4\n\n# Defi"
  },
  {
    "path": "docker/verl0.4-cu124-torch2.6-fa2.7.4/Dockerfile.app.sglang.vllm.mcore0.13.preview",
    "chars": 3453,
    "preview": "# Start from the verl base image\n# Dockerfile.base\nFROM verlai/verl:base-verl0.4-cu124-cudnn9.8-torch2.6-fa2.7.4\n\n# Defi"
  },
  {
    "path": "docker/verl0.4-cu124-torch2.6-fa2.7.4/Dockerfile.app.vllm.mcore0.12",
    "chars": 2229,
    "preview": "# Start from the verl base image\n# Dockerfile.base\nFROM verlai/verl:base-verl0.4-cu124-cudnn9.8-torch2.6-fa2.7.4\n\n# Defi"
  },
  {
    "path": "docker/verl0.4-cu124-torch2.6-fa2.7.4/Dockerfile.app.vllm.mcore0.12.deepep",
    "chars": 3750,
    "preview": "# Start from the verl base image\n# Dockerfile.base\nFROM verlai/verl:base-verl0.4-cu124-cudnn9.8-torch2.6-fa2.7.4\n\n# Defi"
  },
  {
    "path": "docker/verl0.4-cu124-torch2.6-fa2.7.4/Dockerfile.app.vllm.mcore0.13.preview",
    "chars": 3663,
    "preview": "# Start from the verl base image\n# Dockerfile.base\nFROM verlai/verl:base-verl0.4-cu124-cudnn9.8-torch2.6-fa2.7.4\n\n# Defi"
  },
  {
    "path": "docker/verl0.4-cu124-torch2.6-fa2.7.4/Dockerfile.base",
    "chars": 5477,
    "preview": "# Base Docker Image of verl, with CUDA/Torch/FlashAttn/Apex/TransformerEngine, without other frameworks\n# Target: verlai"
  },
  {
    "path": "docker/verl0.4-cu124-torch2.6-fa2.7.4/README.md",
    "chars": 1005,
    "preview": "# verl image with verl v0.4.x\n\n## Important packages version\n\n```txt\ncuda==12.4\ncudnn==9.8.0\ntorch==2.6.0\nflash_attn=2.7"
  },
  {
    "path": "docker/verl0.5-cu126-torch2.7-fa2.7.4/Dockerfile.app.sglang0.4.10.post2.mcore0.13",
    "chars": 1680,
    "preview": "# Start from the verl base image\n# Dockerfile.base\nFROM verlai/verl:base-verl0.5-cu126-cudnn9.8-torch2.7.1-fa2.7.4\n\n# De"
  },
  {
    "path": "docker/verl0.5-cu126-torch2.7-fa2.7.4/Dockerfile.app.sglang0.4.9.post6.mcore0.13",
    "chars": 1679,
    "preview": "# Start from the verl base image\n# Dockerfile.base\nFROM verlai/verl:base-verl0.5-cu126-cudnn9.8-torch2.7.1-fa2.7.4\n\n# De"
  },
  {
    "path": "docker/verl0.5-cu126-torch2.7-fa2.7.4/Dockerfile.app.vllm.mcore0.13",
    "chars": 1575,
    "preview": "# Start from the verl base image\n# Dockerfile.base\nFROM verlai/verl:base-verl0.5-cu126-cudnn9.8-torch2.7.1-fa2.7.4\n\n# De"
  },
  {
    "path": "docker/verl0.5-cu126-torch2.7-fa2.7.4/Dockerfile.app.vllm.mcore0.15",
    "chars": 1627,
    "preview": "# Start from the verl base image\n# Dockerfile.base\nFROM iseekyan/verl:base-verl0.5-cu126-cudnn9.8-torch2.7.1-fa2.7.4-h10"
  },
  {
    "path": "docker/verl0.5-cu126-torch2.7-fa2.7.4/Dockerfile.base.torch2.7.1",
    "chars": 6184,
    "preview": "# Base Docker Image of verl, with CUDA/Torch/FlashAttn/Apex/TransformerEngine, without other frameworks\n# Target: verlai"
  },
  {
    "path": "docker/verl0.5-cu126-torch2.7-fa2.7.4/README.md",
    "chars": 721,
    "preview": "# verl image with verl v0.5\n\n## Important packages version\n\n```txt\ncuda==12.6\ncudnn==9.8.0\ntorch==2.7.1\nflash_attn=2.7.4"
  },
  {
    "path": "docker/verl0.5-cu126-torch2.7.1-fa2.8.0/Dockerfile.app.sglang.mcore0.12",
    "chars": 1724,
    "preview": "# Start from the verl base image\n# Dockerfile.base\nFROM verlai/verl:base-verl0.5-cu126-cudnn9.8-torch2.7.1-fa2.8.0\n\n# De"
  },
  {
    "path": "docker/verl0.5-cu126-torch2.7.1-fa2.8.0/Dockerfile.app.sglang.mcore0.13.preview",
    "chars": 1732,
    "preview": "# Start from the verl base image\n# Dockerfile.base\nFROM verlai/verl:base-verl0.5-cu126-cudnn9.8-torch2.7.1-fa2.8.0\n\n# De"
  },
  {
    "path": "docker/verl0.5-cu126-torch2.7.1-fa2.8.0/Dockerfile.base",
    "chars": 6053,
    "preview": "# Base Docker Image of verl, with CUDA/Torch/FlashAttn/Apex/TransformerEngine, without other frameworks\n# Target: verlai"
  },
  {
    "path": "docker/verl0.5-cu126-torch2.7.1-fa2.8.0/README.md",
    "chars": 621,
    "preview": "# verl image with verl v0.5\n\n## Important packages version\n\n```txt\ncuda==12.6\ncudnn==9.8.0\ntorch==2.7.1\nflash_attn=2.8.0"
  },
  {
    "path": "docker/verl0.5-preview-cu128-torch2.7.1-fa2.8.0/Dockerfile.app.sglang.megatron",
    "chars": 1691,
    "preview": "# Start from the verl base image\n# Dockerfile.base\nFROM verlai/verl:base-verl0.5-preview-cu128-cudnn9.8-torch2.7.1-fa2.8"
  },
  {
    "path": "docker/verl0.5-preview-cu128-torch2.7.1-fa2.8.0/Dockerfile.base",
    "chars": 4587,
    "preview": "# Base Docker Image of verl, with CUDA/Torch/FlashAttn/Apex/TransformerEngine, without other frameworks\n# Target: verlai"
  },
  {
    "path": "docker/verl0.5-preview-cu128-torch2.7.1-fa2.8.0/README.md",
    "chars": 657,
    "preview": "# verl image with verl v0.5\n\n## Important packages version\n\n```txt\ncuda==12.8\ncudnn==9.8.0\ntorch==2.7.1\nflash_attn=2.8.0"
  },
  {
    "path": "docker/verl0.6-cu128-torch2.8.0-fa2.7.4/Dockerfile.app.sglang",
    "chars": 179,
    "preview": "FROM verlai/verl:base-verl0.6-cu128-cudnn9.8-torch2.8.0-fa2.7.4\n\nRUN pip install --no-cache-dir \"sglang[all]==0.5.2\"\nRUN"
  },
  {
    "path": "docker/verl0.6-cu128-torch2.8.0-fa2.7.4/Dockerfile.base",
    "chars": 4068,
    "preview": "# Start from the NVIDIA official image (ubuntu-24.04 + cuda-12.8 + python-3.12)\n# https://docs.nvidia.com/deeplearning/f"
  },
  {
    "path": "docker/verl0.6-cu128-torch2.8.0-fa2.7.4/Dockerfile.vllm011.mcore_gpt-oss",
    "chars": 514,
    "preview": "FROM nvcr.io/nvidia/nemo:25.07.gpt_oss\n\nRUN git clone -b v0.11.0 --depth 1 https://github.com/vllm-project/vllm.git /opt"
  },
  {
    "path": "docker/verl0.6.1-experimental/Dockerfile.sglang056exp",
    "chars": 2694,
    "preview": "# Dockerfile for verlai/verl:sgl056.exp\nFROM lmsysorg/sglang:v0.5.6.post1\n\nRUN pip install pybind11\n\nRUN pip install nvi"
  },
  {
    "path": "docker/verl0.6.1-experimental/Dockerfile.vllm012exp",
    "chars": 2665,
    "preview": "# dockerfile for verlai/verl:vll012.exp\nFROM nvcr.io/nvidia/pytorch:25.11-py3\n\nRUN git clone -b v0.12.0 --depth 1 https:"
  },
  {
    "path": "docs/Makefile",
    "chars": 602,
    "preview": "# Minimal makefile for Sphinx documentation\n#\n\n# You can set these variables from the command line.\nSPHINXOPTS    =\nSPHI"
  },
  {
    "path": "docs/README.md",
    "chars": 618,
    "preview": "# verl documentations\n\n## Build the docs\n\n```bash\n# If you want to view auto-generated API docstring, please make sure v"
  },
  {
    "path": "docs/README_vllm0.7.md",
    "chars": 3027,
    "preview": "# Upgrading to vllm >= 0.7\n\nNote: verl+vllm 0.8.3 is now stable. Please see ``docs/README_vllm0.8.md`` for upgrade guide"
  },
  {
    "path": "docs/README_vllm0.8.md",
    "chars": 1527,
    "preview": "# Upgrading to vLLM >= 0.8\n\nLast updated: 05/04/2025.\n\n## Installation\n\nNote: This version of verl+vLLM 0.8+ supports **"
  },
  {
    "path": "docs/_static/custom.css",
    "chars": 4891,
    "preview": "/* Make the documentation use full screen width */\n.wy-nav-content {\n    max-width: none !important;\n    width: 100% !im"
  },
  {
    "path": "docs/_static/js/resizable-sidebar.js",
    "chars": 8973,
    "preview": "// Resizable sidebar functionality\ndocument.addEventListener('DOMContentLoaded', function() {\n    const sidebar = docume"
  },
  {
    "path": "docs/_static/js/runllm-widget.js",
    "chars": 618,
    "preview": "document.addEventListener(\"DOMContentLoaded\", function () {\n    var script = document.createElement(\"script\");\n    scrip"
  },
  {
    "path": "docs/advance/agent_loop.rst",
    "chars": 9961,
    "preview": "Agent Loop\n==========\n\nLast updated: 07/17/2025.\n\n.. versionadded:: 0.4.2\n   [status: alpha]\n\n.. warning::\n   Agent Loop"
  },
  {
    "path": "docs/advance/async-on-policy-distill.md",
    "chars": 13439,
    "preview": "# Recipe: Async On-Policy Knowledge Distillation Trainer\n\n**Authors:** Brilliant Hanabi, furunding\n\n**Last updated:** 20"
  },
  {
    "path": "docs/advance/attention_implementation.rst",
    "chars": 4358,
    "preview": ".. _attention-implementation-override:\n\nAttention Implementation Override\n==================================\n\nLast updat"
  },
  {
    "path": "docs/advance/checkpoint.rst",
    "chars": 7652,
    "preview": ".. _checkpoint-page:\n\nUsing Checkpoints to Support Fault Tolerance Training\n============================================"
  },
  {
    "path": "docs/advance/dpo_extension.rst",
    "chars": 9704,
    "preview": "Extend to other RL(HF) algorithms\n=================================\n\nLast updated: 02/25/2025.\n\nWe already implemented t"
  },
  {
    "path": "docs/advance/fp8.md",
    "chars": 6549,
    "preview": "# FP8 RL in verl\n\nLast updated: 03/05/2026\n\nverl supports two FP8 modes for accelerating RL training:\n\n| Mode | Training"
  },
  {
    "path": "docs/advance/fsdp_extension.rst",
    "chars": 4264,
    "preview": "\nAdd models with the FSDP backend\n==================================\n\nLast updated: 02/09/2025.\n\nModel\n-----------------"
  },
  {
    "path": "docs/advance/fully_async.md",
    "chars": 39106,
    "preview": "# Recipe: Fully Async Policy Trainer\n\n**Author:** `https://github.com/meituan-search`\n\nLast updated: 02/05/2026.\n\nThis d"
  },
  {
    "path": "docs/advance/grafana_prometheus.md",
    "chars": 8492,
    "preview": "# Use Prometheus and Grafana to Monitor Rollout\n\n**Author:** `https://github.com/meituan-search`\n\nLast updated: 12/05/20"
  },
  {
    "path": "docs/advance/megatron_extension.rst",
    "chars": 766,
    "preview": "Add models with the Megatron-LM backend\n=========================================\n\nLast updated: 04/25/2025.\n\nModel\n----"
  },
  {
    "path": "docs/advance/mtp.md",
    "chars": 7179,
    "preview": "# Guide to Using MTP in SFT/RL Training and Inference\n\n**Author**: `https://github.com/meituan-search`\n\nLast updated: 02"
  },
  {
    "path": "docs/advance/one_step_off.md",
    "chars": 14897,
    "preview": "# Recipe: One Step Off Policy Async Trainer\n\n**Author:** `https://github.com/meituan-search`\n\nLast updated: 07/17/2025.\n"
  },
  {
    "path": "docs/advance/placement.rst",
    "chars": 456,
    "preview": "Ray API Design Tutorial\n=======================================\n\nLast updated: 10/30/2024.\n\nWe provide a tutorial for ou"
  },
  {
    "path": "docs/advance/ppo_lora.rst",
    "chars": 11369,
    "preview": "RL(HF) algorithms with LoRA Support\n===========================================\n\nLast updated: 02/03/2026.\n\nWe support L"
  },
  {
    "path": "docs/advance/reward_loop.rst",
    "chars": 13236,
    "preview": "Reward Loop\n===========\n\n.. _yyding: https://yyding1.github.io\n\nAuthor: `Yuyang Ding <https://yyding1.github.io>`_\n\nLast"
  },
  {
    "path": "docs/advance/rollout_skip.rst",
    "chars": 2285,
    "preview": "RolloutSkip Function Usage Documentation\n========================================\n\nLast updated: 08/01/2025.\n\nApplicable"
  },
  {
    "path": "docs/advance/rollout_trace.rst",
    "chars": 8833,
    "preview": "Trace Function Usage Instructions\n========================================\n\nLast updated: 07/10/2025.\n\nApplicable Scenar"
  },
  {
    "path": "docs/advance/rope.rst",
    "chars": 1168,
    "preview": "RoPE Scaling override\n=======================================\n\nLast updated: 05/14/2025.\n\nSome models such as `Qwen/Qwen"
  },
  {
    "path": "docs/algo/baseline.md",
    "chars": 12036,
    "preview": "# Algorithm Baselines\n\nLast updated: 06/18/2025.\n\n## Math related datasets\n\n### GSM8k\n\nAssuming GSM8k/math dataset is pr"
  },
  {
    "path": "docs/algo/collabllm.md",
    "chars": 6215,
    "preview": "# Recipe: CollabLLM \n\nLast updated: 09/22/2025.\n\n> Open-Source Algorithm Implementation & Expriement Running: [Haiquan C"
  },
  {
    "path": "docs/algo/dapo.md",
    "chars": 10546,
    "preview": "# Recipe: Decoupled Clip and Dynamic Sampling Policy Optimization (DAPO)\n\nLast updated: 06/19/2025.\n\n> Open-Source Algor"
  },
  {
    "path": "docs/algo/dppo.md",
    "chars": 4958,
    "preview": "# Divergence Proximal Policy Optimization (DPPO)\n\nLast updated: 02/25/2026.\n\n\n<div align=\"center\">\n\n## Rethinking the Tr"
  },
  {
    "path": "docs/algo/entropy.md",
    "chars": 7904,
    "preview": "# Recipe: Entropy Mechanism\n\nLast updated: 06/27/2025.\n\n\n<div align=\"center\">\n\n  The Entropy Mechanism of Reinforcement "
  },
  {
    "path": "docs/algo/gpg.md",
    "chars": 1563,
    "preview": "# GPG: Group Policy Gradient\n\nLast updated: 07/03/2025.\n\nGroup Policy Gradient (GPG) is a minimalist reinforcement learn"
  },
  {
    "path": "docs/algo/grpo.md",
    "chars": 5761,
    "preview": "# Group Relative Policy Optimization (GRPO)\n\nLast updated: 05/31/2025.\n\nIn reinforcement learning, classic algorithms li"
  },
  {
    "path": "docs/algo/opo.md",
    "chars": 2250,
    "preview": "# On-Policy RL with Optimal Reward Baseline (OPO)\n\nLast updated: 06/02/2025.\n\nLoose on-policy constraints and suboptimal"
  },
  {
    "path": "docs/algo/otb.md",
    "chars": 4773,
    "preview": "# Optimal Token Baseline (OTB)\n\nLast updated: 02/23/2026.\n\n📝 [ArXiv](https://www.arxiv.org/abs/2602.07078) | 📒 [Blog](ht"
  },
  {
    "path": "docs/algo/ppo.md",
    "chars": 6886,
    "preview": "# Proximal Policy Optimization (PPO)\n\nLast updated: 06/19/2025.\n\nProximal Policy Optimization (PPO) is a family of polic"
  },
  {
    "path": "docs/algo/rollout_corr.md",
    "chars": 53018,
    "preview": "# Rollout Correction\n\n**Author:** [Yingru Li](https://richardli.xyz/)\n\nLast updated: 10/30/2025.\n\n---\n\n> **📖 Documentati"
  },
  {
    "path": "docs/algo/rollout_corr_math.md",
    "chars": 47622,
    "preview": "# Mathematical Formulations of Rollout Correction Methods in `verl`\n\n**Author:** [Yingru Li](https://richardli.xyz)\n**La"
  },
  {
    "path": "docs/algo/spin.md",
    "chars": 11476,
    "preview": "# Recipe: Self-Play Fine-Tuning (SPIN)\n\nLast updated: 05/31/2025.\n\n`verl` provides a recipe inspired by the paper **\"Sel"
  },
  {
    "path": "docs/algo/sppo.md",
    "chars": 3214,
    "preview": "# Recipe: Self-Play Preference Optimization (SPPO)\n\nLast updated: 05/28/2025.\n\nverl provides a community recipe implemen"
  },
  {
    "path": "docs/amd_tutorial/amd_build_dockerfile_page.rst",
    "chars": 28799,
    "preview": "Getting started with AMD (ROCM Kernel)\n=====================================================\n\nLast updated: 07/06/2025.\n"
  },
  {
    "path": "docs/amd_tutorial/amd_vllm_page.rst",
    "chars": 2579,
    "preview": "verl performance tuning for AMD (ROCm Kernel)\n=====================================================\n\nLast updated: 11/13"
  },
  {
    "path": "docs/api/data.rst",
    "chars": 2276,
    "preview": "Data interface\n=========================\n\nLast updated: 05/19/2025 (API docstrings are auto-generated).\n\nDataProto is th"
  },
  {
    "path": "docs/api/single_controller.rst",
    "chars": 1040,
    "preview": "Single Controller interface\n============================\n\nLast updated: 05/27/2025 (API docstrings are auto-generated).\n"
  },
  {
    "path": "docs/api/trainer.rst",
    "chars": 855,
    "preview": "Trainer Interface\n================================\n\nLast updated: 06/08/2025 (API docstrings are auto-generated).\n\nTrain"
  },
  {
    "path": "docs/api/utils.rst",
    "chars": 1843,
    "preview": "Utilities\n============\n\nLast updated: 05/19/2025 (API docstrings are auto-generated).\n\nThis section documents the utilit"
  },
  {
    "path": "docs/ascend_tutorial/contribution_guide/ascend_ci_guide_zh.rst",
    "chars": 5981,
    "preview": "NPU-CI 添加指导\n===========\n\nLast updated: 02/02/2026.\n\n我们在 verl 上增加基于华为昇腾设备的CI用例添加指导。\n\nverl 仓库使用 GitHub Actions 作为 CI 平台，通过"
  },
  {
    "path": "docs/ascend_tutorial/examples/ascend_performance_analysis_guide.md",
    "chars": 6925,
    "preview": "# Ascend Performance Analysis Guide\n\nLast updated: 02/24/2026.\n\n## 背景介绍\n\n随着DeepSeek-R1的发布，大模型强化学习（RL）训练受到广泛关注。在昇腾NPU环境下，"
  },
  {
    "path": "docs/ascend_tutorial/examples/ascend_retool_best_pratice.rst",
    "chars": 8330,
    "preview": "Ascend Retool Best Practice\n===================================\n\nLast updated: 03/01/2026.\n\n引言\n-------------------------"
  },
  {
    "path": "docs/ascend_tutorial/examples/ascend_sglang_best_practices.rst",
    "chars": 9118,
    "preview": "Ascend SGLang Best Practice\n===================================\n\nLast updated: 01/27/2026.\n\n.. _Qwen3-30B: https://githu"
  },
  {
    "path": "docs/ascend_tutorial/examples/dapo_multi_model_optimization_practice.md",
    "chars": 9330,
    "preview": "# DAPO multi model optimization practice\n\n## DAPO 介绍\n\nLast updated: 03/04/2026.\n\nDAPO的论文可以参考：[DAPO](https://arxiv.org/pd"
  },
  {
    "path": "docs/ascend_tutorial/examples/gspo_optimization_practice.md",
    "chars": 12902,
    "preview": "# NPU Qwen3-32B GSPO Optimization Practice\r\n\r\nLast updated: 02/26/2026.\r\n\r\n本文章对应脚本地址：[qwen3_32b_gspo_npu](https://github"
  },
  {
    "path": "docs/ascend_tutorial/examples/run_qwen3_32B_megatron_1k_256k_npu.md",
    "chars": 5773,
    "preview": "# Long Sequence Qwen3-32B 1k-to-256k Example\n\nLast updated: 6/3/2026.\n\n本章对Qwen3-32B进行了长序列开发。Qwen3-32B的模型能力为最长推到40k\n\n## 全"
  },
  {
    "path": "docs/ascend_tutorial/faq/faq.rst",
    "chars": 27,
    "preview": "Last updated: 03/16/2026.\r\n"
  },
  {
    "path": "docs/ascend_tutorial/features/ascend_backend_features.md",
    "chars": 13689,
    "preview": "# Ascend Backend Features Guide\n==================================================================================\n\nLast"
  },
  {
    "path": "docs/ascend_tutorial/features/ascend_consistency.rst",
    "chars": 1211,
    "preview": "推理一致性指导\n====================================\n\n在昇腾设备上对齐verl和vLLM两个框架下的推理结果。\n\nLast updated: 11/17/2025.\n\n这是一份在昇腾设备上对齐verl和"
  },
  {
    "path": "docs/ascend_tutorial/profiling/ascend_profiling_en.rst",
    "chars": 18428,
    "preview": "Profiling Data Collection Guide\n========================================================================================"
  },
  {
    "path": "docs/ascend_tutorial/profiling/ascend_profiling_zh.rst",
    "chars": 13045,
    "preview": "Profiling采集指导\n==================================================================================\n\nLast updated: 12/20/20"
  },
  {
    "path": "docs/ascend_tutorial/quick_start/ascend_quick_start.rst",
    "chars": 22615,
    "preview": "Ascend Quickstart\n===================================\n\nLast updated: 03/03/2026.\n\n\n\n关键更新\n-------------------------------"
  },
  {
    "path": "docs/ascend_tutorial/quick_start/ascend_sglang_quick_start.rst",
    "chars": 5447,
    "preview": "Ascend Quickstart with SGLang Backend\n===================================\n\nLast updated: 01/27/2026.\n\n我们在 verl 上增加对华为昇腾设"
  },
  {
    "path": "docs/ascend_tutorial/quick_start/dockerfile_build_guidance.rst",
    "chars": 3266,
    "preview": "Ascend Dockerfile Build Guidance\n===================================\n\nLast updated: 03/03/2025.\n\n\n镜像获取 & 公开镜像地址\n--------"
  },
  {
    "path": "docs/blog/v0.7.md",
    "chars": 16929,
    "preview": "# verl 0.7 release blog\n\n**Author:** verl team\n\nLast updated: 01/03/2026.\n\n## Overview\nverl adopts a Hybrid-Controller a"
  },
  {
    "path": "docs/conf.py",
    "chars": 3870,
    "preview": "# Copyright 2024 Bytedance Ltd. and/or its affiliates\r\n#\r\n# Licensed under the Apache License, Version 2.0 (the \"License"
  },
  {
    "path": "docs/data/transfer_queue.md",
    "chars": 18128,
    "preview": "# TransferQueue Data System\n\nLast updated: 01/07/2026.\n\nThis doc introduce [TransferQueue](https://gitcode.com/Ascend/Tr"
  },
  {
    "path": "docs/examples/config.rst",
    "chars": 34061,
    "preview": ".. _config-explain-page:\n\nConfig Explanation\n===================\n\nLast updated: 06/18/2025.\n\nppo_trainer.yaml for RL FSD"
  },
  {
    "path": "docs/examples/gsm8k_example.rst",
    "chars": 6919,
    "preview": "GSM8K Example\n=============\n\nLast updated: 03/25/2025.\n\nIntroduction\n------------\n\nIn this example, we train an LLM to t"
  },
  {
    "path": "docs/examples/multi_modal_example.rst",
    "chars": 944,
    "preview": "Multi-Modal Example Architecture\n=================================\n\nLast updated: 04/28/2025.\n\nIntroduction\n------------"
  },
  {
    "path": "docs/examples/ppo_code_architecture.rst",
    "chars": 9013,
    "preview": "PPO Example Architecture\n========================\n\nLast updated: 02/17/2025.\n\nLet's start with the Proximal Policy Optim"
  },
  {
    "path": "docs/examples/sandbox_fusion_example.rst",
    "chars": 2848,
    "preview": "Sandbox Fusion Example\n============================\n\nLast updated: 06/27/2025.\n\nIntroduction\n------------\n\nSandbox Fusio"
  },
  {
    "path": "docs/examples/skypilot_examples.rst",
    "chars": 4211,
    "preview": "SkyPilot Examples\n=================\n\nLast updated: 09/04/2025.\n\nThis guide provides examples of running VERL reinforceme"
  },
  {
    "path": "docs/faq/faq.rst",
    "chars": 10846,
    "preview": "Frequently Asked Questions\n====================================\n\nLast updated: 09/24/2025.\n\nRay related\n------------\n\nHo"
  },
  {
    "path": "docs/hybrid_flow.rst",
    "chars": 15395,
    "preview": "=========================================================\nHybridFlow Programming Guide\n================================="
  },
  {
    "path": "docs/index.rst",
    "chars": 6918,
    "preview": "Welcome to verl's documentation!\n================================================\n\nverl is a flexible, efficient and pro"
  },
  {
    "path": "docs/perf/best_practices.rst",
    "chars": 16759,
    "preview": "Verl LLM Best Practices (DAPO + Qwen3-235B)\n===========================================\n\nLast updated: 11/03/2025.\n\nPurp"
  },
  {
    "path": "docs/perf/device_tuning.rst",
    "chars": 7561,
    "preview": "Hardware Resource Needed for RL\n===============================\n\nLast updated: 06/25/2025.\n\nSince RL requires more resou"
  },
  {
    "path": "docs/perf/dpsk.md",
    "chars": 5053,
    "preview": "# Training DeepSeek 671b\n\nLast updated: 08/20/2025.\n\nverl integrates Megatron to support large MoE models such as `Qwen3"
  },
  {
    "path": "docs/perf/nsight_profiling.md",
    "chars": 5667,
    "preview": "# NVIDIA Nsight Systems profiling in verl\n\nLast updated: 06/20/2025.\n\nThis guide explains how to use NVIDIA Nsight Syste"
  },
  {
    "path": "docs/perf/perf_tuning.rst",
    "chars": 12777,
    "preview": "Performance Tuning Guide\n==============================\n\nLast updated: 07/17/2025.\n\nAuthor: `Guangming Sheng <https://gi"
  },
  {
    "path": "docs/perf/perf_tuning_on_ascend.rst",
    "chars": 8962,
    "preview": "Performance Tuning Guide on Ascend\n====================================\n\nLast updated:  01/29/2026.\n\nAuthor:  `Xiaobo Hu"
  },
  {
    "path": "docs/perf/torch_profiling.md",
    "chars": 4100,
    "preview": "# PyTorch Profiling in verl\n\nLast updated: 01/13/2026.\n\nThis guide explains how to use the native [PyTorch Profiler](htt"
  },
  {
    "path": "docs/perf/verl_profiler_system.md",
    "chars": 1909,
    "preview": "# verl Profiler System\n\nLast updated: 08/18/2025.\n\n## Architecture\n\nThe architecture of verl profiler system is like bel"
  },
  {
    "path": "docs/preparation/prepare_data.rst",
    "chars": 4352,
    "preview": "Prepare Data for Post-Training\n========================================\n\nLast updated: 02/09/2025.\n\nBefore starting the "
  },
  {
    "path": "docs/preparation/reward_function.rst",
    "chars": 3816,
    "preview": "Implement Reward Function for Dataset\n======================================\n\nLast updated: 06/02/2025.\n\nFor each datase"
  },
  {
    "path": "docs/requirements-docs.txt",
    "chars": 238,
    "preview": "# markdown support\r\nrecommonmark\r\nmyst_parser\r\n# markdown table support\r\nsphinx-markdown-tables\r\n\r\n# theme default rtd\r\n"
  },
  {
    "path": "docs/sglang_multiturn/interaction_system.rst",
    "chars": 14892,
    "preview": "Interaction System for Multi-turn RL Training\n=============================================\n\nLast updated: 06/25/2025.\n\n"
  },
  {
    "path": "docs/sglang_multiturn/multiturn.rst",
    "chars": 16054,
    "preview": "Multi-turn Rollout Support\n==========================\n\nLast updated: 06/27/2025.\n\nBasic Configuration\n~~~~~~~~~~~~~~~~~~"
  },
  {
    "path": "docs/sglang_multiturn/sandbox_fusion.rst",
    "chars": 16637,
    "preview": "===============================\nSandbox Fusion Tool Integration\n===============================\n\nLast updated: 06/10/202"
  },
  {
    "path": "docs/sglang_multiturn/search_tool_example.rst",
    "chars": 7927,
    "preview": "=======================\r\nSearch Tool Integration\r\n=======================\r\n\r\nLast updated: 05/30/2025.\r\n\r\nIntroduction\r\n"
  },
  {
    "path": "docs/single_controller.rst",
    "chars": 12896,
    "preview": "The Design of ``verl.single_controller``\n==============================================\n\nLast updated: 05/21/2025.\n\n**Au"
  },
  {
    "path": "docs/start/agentic_rl.rst",
    "chars": 7414,
    "preview": "Agentic RL Training\n===================\n\nLast updated: 07/15/2025.\n\nOverview\n----------\nThe goal of Agentic RL is to imp"
  },
  {
    "path": "docs/start/install.rst",
    "chars": 11704,
    "preview": "Installation\n============\n\nRequirements\n------------\n\n- **Python**: Version >= 3.10\n- **CUDA**: Version >= 12.8\n\nverl su"
  },
  {
    "path": "docs/start/more_resources.rst",
    "chars": 295,
    "preview": "More Resources\n==============\n\nLast updated: 06/30/2025.\n\n- Introduction to verl (`Slides <https://tongyx361.github.io/b"
  },
  {
    "path": "docs/start/multinode.rst",
    "chars": 31618,
    "preview": "Multinode Training\n==================\n\nLast updated: 06/10/2025.\n\n.. _wuxibin89: https://github.com/wuxibin89\n\nAuthor: `"
  },
  {
    "path": "docs/start/quickstart.rst",
    "chars": 8485,
    "preview": ".. _quickstart:\n\n=========================================================\nQuickstart: PPO training on GSM8K dataset\n==="
  },
  {
    "path": "docs/start/ray_debug_tutorial.rst",
    "chars": 3536,
    "preview": "Ray Debug Tutorial\r\n==================\r\n\r\nLast updated: 04/23/2025\r\n\r\n\r\n.. _wuxibin89: https://github.com/wuxibin89\r\n\r\nA"
  },
  {
    "path": "docs/workers/automodel_workers.rst",
    "chars": 1953,
    "preview": "Automodel Backend\n=================\n\nLast updated: 03/07/2026.\n\nWe support the Automodel (nemo_automodel) backend by imp"
  },
  {
    "path": "docs/workers/fsdp_workers.rst",
    "chars": 3867,
    "preview": "PyTorch FSDP Backend\n======================\n\nLast updated: 12/01/2025.\n\nWe support PyTorch FSDP Backend by implementing "
  },
  {
    "path": "docs/workers/megatron_workers.rst",
    "chars": 10855,
    "preview": "Megatron-LM Backend\n===================\n\nLast updated: 12/01/2025.\n\nWe support Megatron Backend by implementing various "
  },
  {
    "path": "docs/workers/model_engine.rst",
    "chars": 5016,
    "preview": "Model Engine\n============\n\n.. _vermouth: https://github.com/vermouth1992\n\nAuthor: `Chi Zhang <https://github.com/vermout"
  },
  {
    "path": "docs/workers/ray_trainer.rst",
    "chars": 11740,
    "preview": "PPO Ray Trainer\n===============\n\nLast updated: 02/12/2025.\n\nWe implement the RayPPOTrainer, which is a trainer runs on t"
  },
  {
    "path": "docs/workers/sglang_worker.rst",
    "chars": 10820,
    "preview": "SGLang Backend\n==============\n\nLast updated: 05/31/2025.\n\n**Authored By SGLang RL Team and listed alphabetically by last"
  }
]

// ... and 928 more files (download for full content)