SYMBOL INDEX (212 symbols across 38 files)

FILE: gradio_demo.py
  function answer_question (line 35) | def answer_question(img, prompt):
  function extract_floats (line 55) | def extract_floats(text):
  function extract_bbox (line 65) | def extract_bbox(text):
  function process_answer (line 73) | def process_answer(img, answer):

FILE: moondream/eval/chartqa.py
  function relaxed_correctness (line 16) | def relaxed_correctness(
  function eval_chartqa (line 59) | def eval_chartqa(model, debug=False):

FILE: moondream/eval/coco_map.py
  function calculate_iou (line 116) | def calculate_iou(
  function calculate_map (line 132) | def calculate_map(
  function get_total_map (line 210) | def get_total_map(results_by_label, frequency_by_label):
  function eval_coco_map (line 222) | def eval_coco_map(model, iou_threshold=0.5, debug=False):

FILE: moondream/eval/countbenchqa.py
  function eval_countbenchqa (line 14) | def eval_countbenchqa(model, debug=False):

FILE: moondream/eval/docvqa.py
  function get_anls (line 14) | def get_anls(s1, s2):
  function eval_docvqa (line 22) | def eval_docvqa(model, debug=False):

FILE: moondream/eval/eval_all.py
  function create_model (line 22) | def create_model(ckpt_path):
  function eval_all (line 30) | def eval_all(model, skip=[]):

FILE: moondream/eval/gazefollow.py
  function eval_gazefollow (line 12) | def eval_gazefollow(model, debug=False):

FILE: moondream/eval/mmstar.py
  function eval_mmstar (line 13) | def eval_mmstar(model, debug=False):

FILE: moondream/eval/naturalbench.py
  function eval_naturalbench (line 10) | def eval_naturalbench(model, debug=False):

FILE: moondream/eval/pope.py
  function evaluate_pope (line 11) | def evaluate_pope(model, debug=False):

FILE: moondream/eval/realworldqa.py
  function eval_realworldqa (line 12) | def eval_realworldqa(model, debug=False):

FILE: moondream/eval/tallyqa.py
  function eval_tallyqa (line 14) | def eval_tallyqa(model, debug=False):

FILE: moondream/eval/textvqa.py
  function eval_textvqa (line 15) | def eval_textvqa(model, debug=False):

FILE: moondream/eval/utils.py
  class VQAScorer (line 5) | class VQAScorer:
    method __init__ (line 6) | def __init__(self):
    method process_punctuation (line 175) | def process_punctuation(self, inText: str) -> str:
    method process_digit_article (line 188) | def process_digit_article(self, inText: str) -> str:
    method process_answer (line 201) | def process_answer(self, answer):
    method process_line (line 209) | def process_line(self, prediction: str, gt_answers: List[str]) -> float:
    method compute_score (line 225) | def compute_score(

FILE: moondream/eval/waste_detection.py
  function iou (line 18) | def iou(a: Box, b: Box) -> float:
  function match (line 28) | def match(gt: List[Box], pr: List[Box], iou_thr: float) -> Tuple[int, in...
  class WasteDetection (line 54) | class WasteDetection(torch.utils.data.Dataset):
    method __init__ (line 55) | def __init__(self, name: str = "moondream/waste_detection", split: str...
    method __len__ (line 58) | def __len__(self):
    method __getitem__ (line 61) | def __getitem__(self, idx: int) -> Dict:
  function evaluate (line 81) | def evaluate(
  function load_model (line 114) | def load_model(path: str, device: torch.device) -> MoondreamModel:
  function main (line 123) | def main():

FILE: moondream/torch/config.py
  class TextMoeConfig (line 6) | class TextMoeConfig:
  class TextConfig (line 14) | class TextConfig:
  class VisionConfig (line 28) | class VisionConfig:
  class RegionConfig (line 43) | class RegionConfig:
  class TokenizerConfig (line 53) | class TokenizerConfig:
  class MoondreamConfig (line 77) | class MoondreamConfig:
    method from_dict (line 84) | def from_dict(cls, config_dict: dict):
    method to_dict (line 96) | def to_dict(self):

FILE: moondream/torch/hf_moondream.py
  function extract_question (line 18) | def extract_question(text):
  class HfConfig (line 28) | class HfConfig(PretrainedConfig):
    method __init__ (line 32) | def __init__(self, **kwargs):
  class HfMoondream (line 37) | class HfMoondream(PreTrainedModel):
    method __init__ (line 41) | def __init__(self, config):
    method _setup_caches (line 48) | def _setup_caches(self):
    method encode_image (line 54) | def encode_image(self):
    method query (line 59) | def query(self):
    method caption (line 64) | def caption(self):
    method detect (line 69) | def detect(self):
    method point (line 74) | def point(self):
    method detect_gaze (line 79) | def detect_gaze(self):
    method answer_question (line 83) | def answer_question(
    method batch_answer (line 99) | def batch_answer(self, images, prompts, tokenizer=None, **kwargs):
    method _unsupported_exception (line 105) | def _unsupported_exception(self):
    method generate (line 112) | def generate(self, image_embeds, prompt, tokenizer, max_new_tokens=128...
    method get_input_embeddings (line 142) | def get_input_embeddings(self) -> nn.Embedding:
    method set_input_embeddings (line 155) | def set_input_embeddings(self, value: Union[nn.Embedding, nn.Module]) ...
    method input_embeds (line 165) | def input_embeds(

FILE: moondream/torch/image_crops.py
  function select_tiling (line 17) | def select_tiling(
  class OverlapCropOutput (line 53) | class OverlapCropOutput(TypedDict):
  function overlap_crop_image (line 58) | def overlap_crop_image(
  function reconstruct_from_crops (line 170) | def reconstruct_from_crops(

FILE: moondream/torch/layers.py
  function quantize_ (line 13) | def quantize_(model, quant_mode):
  function int4_weight_only (line 18) | def int4_weight_only(group_size):
  function gelu_approx (line 24) | def gelu_approx(x):
  class LinearWeights (line 29) | class LinearWeights:
  function linear (line 34) | def linear(x: torch.Tensor, w: LinearWeights) -> torch.Tensor:
  function dequantize_tensor (line 38) | def dequantize_tensor(W_q, scale, zero, orig_shape, dtype=torch.bfloat16):
  class QuantizedLinear (line 47) | class QuantizedLinear(nn.Module):
    method __init__ (line 48) | def __init__(
    method unpack (line 79) | def unpack(self):
    method forward (line 106) | def forward(self, x: torch.Tensor) -> torch.Tensor:
  class LayerNormWeights (line 113) | class LayerNormWeights:
  function layer_norm (line 118) | def layer_norm(x: torch.Tensor, w: LayerNormWeights) -> torch.Tensor:
  class MLPWeights (line 123) | class MLPWeights:
  function mlp (line 129) | def mlp(x: torch.Tensor, w: MLPWeights, lora: Optional[dict] = None) -> ...
  function moe_mlp (line 149) | def moe_mlp(
  class AttentionWeights (line 218) | class AttentionWeights:
  function attn (line 223) | def attn(x: torch.Tensor, w: AttentionWeights, n_heads: int) -> torch.Te...

FILE: moondream/torch/lora.py
  function variant_cache_dir (line 11) | def variant_cache_dir():
  function cached_variant_path (line 23) | def cached_variant_path(variant_id: str):
  function nest (line 46) | def nest(flat):
  function variant_state_dict (line 58) | def variant_state_dict(variant_id: Optional[str] = None, device: str = "...

FILE: moondream/torch/moondream.py
  class EncodedImage (line 58) | class EncodedImage:
  class KVCache (line 63) | class KVCache(nn.Module):
    method __init__ (line 65) | def __init__(self, n_heads, n_kv_heads, max_context, dim, device, dtype):
    method update (line 75) | def update(self, pos_ids, k, v):
  function causal_mask (line 82) | def causal_mask(b, h, q_idx, kv_idx):
  function get_mask_mod (line 86) | def get_mask_mod(mask_mod, offset):
  class MoondreamModel (line 93) | class MoondreamModel(nn.Module):
    method __init__ (line 95) | def __init__(
    method causal_block_mask (line 151) | def causal_block_mask(self):
    method point_gen_indices (line 164) | def point_gen_indices(self):
    method _setup_caches (line 172) | def _setup_caches(self):
    method device (line 185) | def device(self):
    method _vis_enc (line 188) | def _vis_enc(self, x: torch.Tensor):
    method _vis_proj (line 191) | def _vis_proj(self, g: torch.Tensor, r: torch.Tensor):
    method _prefill (line 194) | def _prefill(
    method _decode_one_tok (line 203) | def _decode_one_tok(
    method compile (line 232) | def compile(self):
    method _run_vision_encoder (line 272) | def _run_vision_encoder(self, image: Image.Image) -> torch.Tensor:
    method encode_image (line 296) | def encode_image(
    method _apply_top_p (line 338) | def _apply_top_p(self, probs: torch.Tensor, top_p: float):
    method _prefill_prompt (line 348) | def _prefill_prompt(
    method _generate_reasoning (line 393) | def _generate_reasoning(
    method _generate_answer (line 506) | def _generate_answer(
    method query (line 615) | def query(
    method load_encoded_image (line 691) | def load_encoded_image(self, encoded_image: EncodedImage):
    method caption (line 696) | def caption(
    method _generate_points (line 724) | def _generate_points(
    method detect (line 816) | def detect(
    method point (line 864) | def point(
    method _detect_gaze (line 912) | def _detect_gaze(
    method detect_gaze (line 968) | def detect_gaze(
  function _is_cjk_char (line 1060) | def _is_cjk_char(cp):

FILE: moondream/torch/region.py
  function fourier_features (line 10) | def fourier_features(x: torch.Tensor, w: torch.Tensor) -> torch.Tensor:
  function encode_coordinate (line 30) | def encode_coordinate(coord: torch.Tensor, w: nn.Module) -> torch.Tensor:
  function decode_coordinate (line 44) | def decode_coordinate(hidden_state: torch.Tensor, w: nn.Module) -> torch...
  function encode_size (line 58) | def encode_size(size: torch.Tensor, w: nn.Module) -> torch.Tensor:
  function decode_size (line 72) | def decode_size(hidden_state: torch.Tensor, w: nn.Module) -> torch.Tensor:
  function encode_spatial_refs (line 94) | def encode_spatial_refs(spatial_refs: SpatialRefs, w: nn.Module) -> torc...

FILE: moondream/torch/rope.py
  function precompute_freqs_cis (line 6) | def precompute_freqs_cis(
  function apply_rotary_emb (line 19) | def apply_rotary_emb(

FILE: moondream/torch/text.py
  function text_encoder (line 13) | def text_encoder(input_ids: torch.Tensor, w: nn.Module):
  function attn (line 17) | def attn(
  function text_decoder (line 82) | def text_decoder(
  function lm_head (line 124) | def lm_head(
  function build_dense_mlp (line 137) | def build_dense_mlp(d_model, d_ffn, dtype, linear_cls):
  function build_moe_mlp (line 146) | def build_moe_mlp(d_model, d_ffn, n_experts, dtype):
  function build_text_model (line 169) | def build_text_model(config: TextConfig, dtype: torch.dtype) -> nn.Module:

FILE: moondream/torch/utils.py
  function remove_outlier_points (line 4) | def remove_outlier_points(points_tuples, k_nearest=2, threshold=2.0):

FILE: moondream/torch/vision.py
  function adaptive_avg_pool2d (line 16) | def adaptive_avg_pool2d(input, output_size):
  function prepare_crops (line 25) | def prepare_crops(
  function create_patches (line 44) | def create_patches(x, patch_size):
  function vision_encoder (line 64) | def vision_encoder(input_BCHW: torch.Tensor, w: nn.Module, config: Visio...
  function vision_projection (line 77) | def vision_projection(
  function build_vision_model (line 92) | def build_vision_model(config: VisionConfig, dtype: torch.dtype):

FILE: moondream/torch/weights.py
  function safetensors_open (line 10) | def safetensors_open(safetensors_file: str):
  function _load_weights (line 30) | def _load_weights(get_tensor: Callable[[str], torch.Tensor], model: nn.M...
  function load_weights_from_safetensors (line 130) | def load_weights_from_safetensors(weights_file: str, model: nn.Module) -...
  function load_weights_from_pt (line 150) | def load_weights_from_pt(weights_file: str, model: nn.Module) -> None:
  function load_weights_into_model (line 166) | def load_weights_into_model(weights_file: str, model: nn.Module) -> None:

FILE: recipes/gaze-detection-video/gaze-detection-video.py
  function initialize_model (line 20) | def initialize_model() -> Optional[AutoModelForCausalLM]:
  function video_handler (line 56) | def video_handler(
  function fig2rgb_array (line 81) | def fig2rgb_array(fig: plt.Figure) -> np.ndarray:
  function visualize_frame (line 91) | def visualize_frame(
  function process_video (line 200) | def process_video(

FILE: recipes/promptable-content-moderation/app.py
  function process_video_file (line 28) | def process_video_file(
  function create_visualization_plots (line 119) | def create_visualization_plots(json_path):

FILE: recipes/promptable-content-moderation/deep_sort_integration.py
  class DeepSORTTracker (line 7) | class DeepSORTTracker:
    method __init__ (line 8) | def __init__(self, max_age=5):
    method _create_tracker (line 13) | def _create_tracker(self):
    method reset (line 21) | def reset(self):
    method update (line 26) | def update(self, frame, detections):

FILE: recipes/promptable-content-moderation/main.py
  function get_sam_model (line 50) | def get_sam_model(slim=False):
  function load_sam_model (line 72) | def load_sam_model(slim=False):
  function generate_color_pair (line 77) | def generate_color_pair():
  function create_mask_overlay (line 84) | def create_mask_overlay(image, masks, points=None, labels=None):
  function process_sam_detection (line 153) | def process_sam_detection(image, center_x, center_y, slim=False):
  function load_moondream (line 184) | def load_moondream():
  function get_video_properties (line 193) | def get_video_properties(video_path):
  function is_valid_bounding_box (line 204) | def is_valid_bounding_box(bounding_box):
  function split_frame_into_grid (line 221) | def split_frame_into_grid(frame, grid_rows, grid_cols):
  function convert_tile_coords_to_frame (line 243) | def convert_tile_coords_to_frame(box, tile_pos, frame_shape):
  function merge_tile_detections (line 273) | def merge_tile_detections(tile_detections, iou_threshold=0.5):
  function detect_objects_in_frame (line 330) | def detect_objects_in_frame(
  function detect_objects_in_frame_single (line 376) | def detect_objects_in_frame_single(model, tokenizer, image, target_object):
  function draw_hitmarker (line 401) | def draw_hitmarker(
  function draw_ad_boxes (line 501) | def draw_ad_boxes(frame, detected_objects, detect_keyword, model, box_st...
  function filter_temporal_outliers (line 767) | def filter_temporal_outliers(detections_dict):
  function describe_frames (line 805) | def describe_frames(
  function create_detection_video (line 875) | def create_detection_video(
  function process_video (line 1058) | def process_video(
  function main (line 1226) | def main():

FILE: recipes/promptable-content-moderation/persistence.py
  function save_detection_data (line 5) | def save_detection_data(data, output_file):
  function load_detection_data (line 26) | def load_detection_data(input_file):

FILE: recipes/promptable-content-moderation/video_visualization.py
  function create_frame_data (line 12) | def create_frame_data(json_path):
  function generate_frame_image (line 64) | def generate_frame_image(df, frame_num, temp_dir, max_y):
  function generate_gauge_frame (line 124) | def generate_gauge_frame(df, frame_num, temp_dir, detect_keyword="OBJECT"):
  function create_video_visualization (line 256) | def create_video_visualization(json_path, style="timeline"):

FILE: recipes/promptable-content-moderation/visualization.py
  function visualize_detections (line 7) | def visualize_detections(json_path):
  function main (line 97) | def main():

FILE: recipes/promptable-video-redaction/app.py
  function process_video_file (line 22) | def process_video_file(

FILE: recipes/promptable-video-redaction/main.py
  function load_moondream (line 35) | def load_moondream():
  function get_video_properties (line 44) | def get_video_properties(video_path):
  function is_valid_box (line 55) | def is_valid_box(box):
  function split_frame_into_tiles (line 72) | def split_frame_into_tiles(frame, rows, cols):
  function convert_tile_coords_to_frame (line 94) | def convert_tile_coords_to_frame(box, tile_pos, frame_shape):
  function merge_tile_detections (line 124) | def merge_tile_detections(tile_detections, iou_threshold=0.5):
  function detect_ads_in_frame (line 181) | def detect_ads_in_frame(model, tokenizer, image, detect_keyword, rows=1,...
  function detect_ads_in_frame_single (line 225) | def detect_ads_in_frame_single(model, tokenizer, image, detect_keyword):
  function draw_hitmarker (line 250) | def draw_hitmarker(
  function draw_ad_boxes (line 350) | def draw_ad_boxes(frame, detected_objects, detect_keyword, box_style="ce...
  function filter_temporal_outliers (line 427) | def filter_temporal_outliers(detections_dict):
  function describe_frames (line 455) | def describe_frames(
  function create_detection_video (line 503) | def create_detection_video(
  function process_video (line 617) | def process_video(
  function main (line 657) | def main():

FILE: tests/test_image_crops.py
  function test_overlap_crop_basic (line 6) | def test_overlap_crop_basic():
  function test_overlap_crop_small_image (line 21) | def test_overlap_crop_small_image():
  function test_reconstruction (line 32) | def test_reconstruction():

FILE: webcam_gradio_demo.py
  function answer_question (line 33) | def answer_question(img, prompt):
  function img_change (line 81) | def img_change(img):
  function prompt_change (line 86) | def prompt_change(prompt):
  function live_video (line 91) | def live_video():