![\"Open](\"https://colab.research.google.com/assets/colab-badge.svg\"){kind=icon}
"
]
},
{
"cell_type": "code",
"source": [
"#@title Allowed Resources\n",
"import multiprocessing\n",
"import torch\n",
"from psutil import virtual_memory\n",
"\n",
"ram_gb = round(virtual_memory().total / 1024**3, 1)\n",
"\n",
"print('CPU:', multiprocessing.cpu_count())\n",
"print('RAM GB:', ram_gb)\n",
"print(\"PyTorch version:\", torch.__version__)\n",
"print(\"CUDA version:\", torch.version.cuda)\n",
"print(\"cuDNN version:\", torch.backends.cudnn.version())\n",
"device = torch.device(\"cuda:0\" if torch.cuda.is_available() else \"cpu\")\n",
"print(\"device:\", device.type)\n",
"\n",
"!nvidia-smi"
],
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"cellView": "form",
"id": "4gfq46gnYcnU",
"outputId": "41e2054a-e2e4-4bb5-ed39-8bd8bfc639c3"
},
"execution_count": 1,
"outputs": [
{
"output_type": "stream",
"name": "stdout",
"text": [
"CPU: 2\n",
"RAM GB: 12.7\n",
"PyTorch version: 1.10.0+cu111\n",
"CUDA version: 11.1\n",
"cuDNN version: 8005\n",
"device: cuda\n",
"Wed Jan 19 22:10:10 2022 \n",
"+-----------------------------------------------------------------------------+\n",
"| NVIDIA-SMI 495.46 Driver Version: 460.32.03 CUDA Version: 11.2 |\n",
"|-------------------------------+----------------------+----------------------+\n",
"| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |\n",
"| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |\n",
"| | | MIG M. |\n",
"|===============================+======================+======================|\n",
"| 0 Tesla T4 Off | 00000000:00:04.0 Off | 0 |\n",
"| N/A 41C P8 9W / 70W | 3MiB / 15109MiB | 0% Default |\n",
"| | | N/A |\n",
"+-------------------------------+----------------------+----------------------+\n",
" \n",
"+-----------------------------------------------------------------------------+\n",
"| Processes: |\n",
"| GPU GI CI PID Type Process name GPU Memory |\n",
"| ID ID Usage |\n",
"|=============================================================================|\n",
"| No running processes found |\n",
"+-----------------------------------------------------------------------------+\n"
]
}
]
},
{
"cell_type": "markdown",
"source": [
"## Restart colab session after installation\n",
"Reload the session if something doesn't work"
],
"metadata": {
"id": "whlsBiJgR8le"
}
},
{
"cell_type": "code",
"source": [
"%%capture\n",
"!pip install git+https://github.com/Lednik7/CLIP-ONNX.git\n",
"!pip install ruclip==0.0.1rc7\n",
"!pip install onnxruntime-gpu"
],
"metadata": {
"id": "HnbpAkvuR73L"
},
"execution_count": 2,
"outputs": []
},
{
"cell_type": "code",
"source": [
"%%capture\n",
"!wget -c -O CLIP.png https://github.com/openai/CLIP/blob/main/CLIP.png?raw=true"
],
"metadata": {
"id": "tqy0zKM4R-7M"
},
"execution_count": 3,
"outputs": []
},
{
"cell_type": "code",
"source": [
"import onnxruntime\n",
"\n",
"# priority device (if available)\n",
"print(onnxruntime.get_device())"
],
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "x8IN72OnSAIh",
"outputId": "3174cf2c-ace3-4e1f-a550-e16c72302d51"
},
"execution_count": 4,
"outputs": [
{
"output_type": "stream",
"name": "stdout",
"text": [
"GPU\n"
]
}
]
},
{
"cell_type": "markdown",
"source": [
"## RuCLIP\n",
"WARNING: specific RuCLIP like forward \"model(text, image)\" instead of classic(OpenAI CLIP) \"model(image, text)\""
],
"metadata": {
"id": "8_wSsSheT5mw"
}
},
{
"cell_type": "code",
"source": [
"import warnings\n",
"\n",
"warnings.filterwarnings(\"ignore\", category=UserWarning)"
],
"metadata": {
"id": "gZTxanR26knr"
},
"execution_count": 1,
"outputs": []
},
{
"cell_type": "code",
"source": [
"import ruclip\n",
"\n",
"# onnx cannot export with cuda\n",
"model, processor = ruclip.load(\"ruclip-vit-base-patch32-384\", device=\"cpu\")"
],
"metadata": {
"id": "FdTLuqsJUBFY"
},
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"source": [
"from PIL import Image\n",
"import numpy as np\n",
"\n",
"# simple input\n",
"pil_images = [Image.open(\"CLIP.png\")]\n",
"labels = ['диаграмма', 'собака', 'кошка']\n",
"dummy_input = processor(text=labels, images=pil_images,\n",
" return_tensors='pt', padding=True)\n",
"\n",
"# batch first\n",
"image = dummy_input[\"pixel_values\"] # torch tensor [1, 3, 384, 384]\n",
"image_onnx = dummy_input[\"pixel_values\"].cpu().detach().numpy().astype(np.float32)\n",
"\n",
"# batch first\n",
"text = dummy_input[\"input_ids\"] # torch tensor [3, 77]\n",
"text_onnx = dummy_input[\"input_ids\"].cpu().detach().numpy()[::-1].astype(np.int64)"
],
"metadata": {
"id": "rPwc6A2SSGyl"
},
"execution_count": 3,
"outputs": []
},
{
"cell_type": "code",
"source": [
"#RuCLIP output\n",
"logits_per_image, logits_per_text = model(text, image)\n",
"probs = logits_per_image.softmax(dim=-1).detach().cpu().numpy()\n",
"\n",
"print(\"Label probs:\", probs) # prints: [[0.9885839 0.00894288 0.0024732 ]]"
],
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "pv0mH626SdzO",
"outputId": "d563462f-b2a9-4d49-b491-17e88ffa81f0"
},
"execution_count": 4,
"outputs": [
{
"output_type": "stream",
"name": "stdout",
"text": [
"Label probs: [[0.9885839 0.00894288 0.0024732 ]]\n"
]
}
]
},
{
"cell_type": "markdown",
"source": [
"## Convert RuCLIP model to ONNX"
],
"metadata": {
"id": "R_e5OjJeXRiF"
}
},
{
"cell_type": "code",
"source": [
"from clip_onnx import clip_onnx\n",
"\n",
"visual_path = \"clip_visual.onnx\"\n",
"textual_path = \"clip_textual.onnx\"\n",
"\n",
"onnx_model = clip_onnx(model, visual_path=visual_path, textual_path=textual_path)\n",
"onnx_model.convert2onnx(image, text, verbose=True)"
],
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "oYM5FDSGSJBW",
"outputId": "c647dc2e-946d-4769-c66e-77edfa98237f"
},
"execution_count": 5,
"outputs": [
{
"output_type": "stream",
"name": "stdout",
"text": [
"[CLIP ONNX] Start convert visual model\n",
"[CLIP ONNX] Start check visual model\n",
"[CLIP ONNX] Start convert textual model\n",
"[CLIP ONNX] Start check textual model\n",
"[CLIP ONNX] Models converts successfully\n"
]
}
]
},
{
"cell_type": "markdown",
"source": [
"## [ONNX] CPU inference mode"
],
"metadata": {
"id": "U1Pr-YTtSEhs"
}
},
{
"cell_type": "code",
"source": [
"# ['TensorrtExecutionProvider', 'CUDAExecutionProvider', 'CPUExecutionProvider']\n",
"onnx_model.start_sessions(providers=[\"CPUExecutionProvider\"]) # cpu mode"
],
"metadata": {
"id": "aY9wRe5kT3wG"
},
"execution_count": 6,
"outputs": []
},
{
"cell_type": "code",
"source": [
"image_features = onnx_model.encode_image(image_onnx)\n",
"text_features = onnx_model.encode_text(text_onnx)\n",
"\n",
"logits_per_image, logits_per_text = onnx_model(image_onnx, text_onnx)\n",
"probs = logits_per_image.softmax(dim=-1).detach().cpu().numpy()\n",
"\n",
"print(\"Label probs:\", probs) # prints: Label probs: [[0.90831375 0.07174418 0.01994203]]"
],
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "tYVuk72nSLw6",
"outputId": "75bf3803-6ed7-4516-ccd0-42f9cf7f22e0"
},
"execution_count": 7,
"outputs": [
{
"output_type": "stream",
"name": "stdout",
"text": [
"Label probs: [[0.90831375 0.07174418 0.01994203]]\n"
]
}
]
},
{
"cell_type": "code",
"source": [
"%timeit onnx_model.encode_text(text_onnx) # text representation"
],
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "Bpu4_HFRVeNk",
"outputId": "e8f1681b-40dc-495f-d382-f0348d87c412"
},
"execution_count": 8,
"outputs": [
{
"output_type": "stream",
"name": "stdout",
"text": [
"1 loop, best of 5: 285 ms per loop\n"
]
}
]
},
{
"cell_type": "code",
"source": [
"%timeit onnx_model.encode_image(image_onnx) # image representation"
],
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "JsOccP2gVmpo",
"outputId": "adb33860-b000-461b-959f-95126e2ac049"
},
"execution_count": 9,
"outputs": [
{
"output_type": "stream",
"name": "stdout",
"text": [
"1 loop, best of 5: 412 ms per loop\n"
]
}
]
},
{
"cell_type": "markdown",
"source": [
"## [ONNX] GPU inference mode"
],
"metadata": {
"id": "Zww0E-jIULug"
}
},
{
"cell_type": "code",
"source": [
"onnx_model.start_sessions(providers=[\"CUDAExecutionProvider\"]) # cuda mode"
],
"metadata": {
"id": "PBakYeiQUOAm"
},
"execution_count": 10,
"outputs": []
},
{
"cell_type": "code",
"source": [
"%timeit onnx_model.encode_text(text_onnx) # text representation"
],
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "EjvRBvCaWJBL",
"outputId": "07426652-1cc5-4713-c355-fb4f1bd138d4"
},
"execution_count": 11,
"outputs": [
{
"output_type": "stream",
"name": "stdout",
"text": [
"The slowest run took 5.07 times longer than the fastest. This could mean that an intermediate result is being cached.\n",
"100 loops, best of 5: 6.89 ms per loop\n"
]
}
]
},
{
"cell_type": "code",
"source": [
"%timeit onnx_model.encode_image(image_onnx) # image representation"
],
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "pmu4mQCsWJ8w",
"outputId": "5cb45026-dfd3-419d-e5d3-f5d0d9681cd0"
},
"execution_count": 12,
"outputs": [
{
"output_type": "stream",
"name": "stdout",
"text": [
"The slowest run took 699.84 times longer than the fastest. This could mean that an intermediate result is being cached.\n",
"1 loop, best of 5: 18.9 ms per loop\n"
]
}
]
}
]
}
================================================
FILE: examples/clip_onnx_example.ipynb
================================================
{
"nbformat": 4,
"nbformat_minor": 0,
"metadata": {
"colab": {
"name": "clip_onnx_example.ipynb",
"provenance": [],
"include_colab_link": true
},
"kernelspec": {
"name": "python3",
"display_name": "Python 3"
},
"language_info": {
"name": "python"
},
"accelerator": "GPU"
},
"cells": [
{
"cell_type": "markdown",
"metadata": {
"id": "view-in-github",
"colab_type": "text"
},
"source": [
""
]
},
{
"cell_type": "markdown",
"source": [
"## Restart colab session after installation\n",
"Reload the session if something doesn't work"
],
"metadata": {
"id": "fxPg_VvZuScV"
}
},
{
"cell_type": "code",
"execution_count": 1,
"metadata": {
"id": "al_QNjyFq6Jj"
},
"outputs": [],
"source": [
"%%capture\n",
"!pip install git+https://github.com/Lednik7/CLIP-ONNX.git\n",
"!pip install git+https://github.com/openai/CLIP.git\n",
"!pip install onnxruntime-gpu"
]
},
{
"cell_type": "code",
"source": [
"%%capture\n",
"!wget -c -O CLIP.png https://github.com/openai/CLIP/blob/main/CLIP.png?raw=true"
],
"metadata": {
"id": "42eeJz9lTdJ6"
},
"execution_count": 2,
"outputs": []
},
{
"cell_type": "code",
"source": [
"!nvidia-smi"
],
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "XuauIZIBSEUX",
"outputId": "2c7c2bd9-90dd-4b1a-e98a-79e1f2218644"
},
"execution_count": 3,
"outputs": [
{
"output_type": "stream",
"name": "stdout",
"text": [
"Thu Jan 6 16:36:44 2022 \n",
"+-----------------------------------------------------------------------------+\n",
"| NVIDIA-SMI 495.44 Driver Version: 460.32.03 CUDA Version: 11.2 |\n",
"|-------------------------------+----------------------+----------------------+\n",
"| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |\n",
"| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |\n",
"| | | MIG M. |\n",
"|===============================+======================+======================|\n",
"| 0 Tesla K80 Off | 00000000:00:04.0 Off | 0 |\n",
"| N/A 35C P8 26W / 149W | 0MiB / 11441MiB | 0% Default |\n",
"| | | N/A |\n",
"+-------------------------------+----------------------+----------------------+\n",
" \n",
"+-----------------------------------------------------------------------------+\n",
"| Processes: |\n",
"| GPU GI CI PID Type Process name GPU Memory |\n",
"| ID ID Usage |\n",
"|=============================================================================|\n",
"| No running processes found |\n",
"+-----------------------------------------------------------------------------+\n"
]
}
]
},
{
"cell_type": "code",
"source": [
"import onnxruntime\n",
"print(onnxruntime.get_device())"
],
"metadata": {
"id": "gqvxpdajRX5_"
},
"execution_count": null,
"outputs": []
},
{
"cell_type": "markdown",
"source": [
"## CPU inference mode"
],
"metadata": {
"id": "010k-ksVTjAu"
}
},
{
"cell_type": "markdown",
"source": [
"### Torch CLIP"
],
"metadata": {
"id": "KdTz0IJWVBqE"
}
},
{
"cell_type": "code",
"source": [
"import clip\n",
"from PIL import Image\n",
"import numpy as np\n",
"\n",
"# onnx cannot work with cuda\n",
"model, preprocess = clip.load(\"ViT-B/32\", device=\"cpu\", jit=False)\n",
"\n",
"# batch first\n",
"image = preprocess(Image.open(\"CLIP.png\")).unsqueeze(0).cpu() # [1, 3, 224, 224]\n",
"image_onnx = image.detach().cpu().numpy().astype(np.float32)\n",
"\n",
"# batch first\n",
"text = clip.tokenize([\"a diagram\", \"a dog\", \"a cat\"]).cpu() # [3, 77]\n",
"text_onnx = text.detach().cpu().numpy().astype(np.int64)"
],
"metadata": {
"id": "9ROPwKYurOhP"
},
"execution_count": 1,
"outputs": []
},
{
"cell_type": "code",
"source": [
"%timeit model(image, text)"
],
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "1CrHQ8cYt8Cx",
"outputId": "4d98f85d-4b02-4ae2-b18f-fb3c7a2d6caf"
},
"execution_count": 2,
"outputs": [
{
"output_type": "stream",
"name": "stdout",
"text": [
"1 loop, best of 5: 636 ms per loop\n"
]
}
]
},
{
"cell_type": "markdown",
"source": [
"### CLIP-ONNX"
],
"metadata": {
"id": "Ao2MriaVVG6Y"
}
},
{
"cell_type": "code",
"source": [
"from clip_onnx import clip_onnx, attention\n",
"clip.model.ResidualAttentionBlock.attention = attention\n",
"\n",
"onnx_model = clip_onnx(model)\n",
"onnx_model.convert2onnx(image, text, verbose=True)\n",
"# ['TensorrtExecutionProvider', 'CUDAExecutionProvider', 'CPUExecutionProvider']\n",
"onnx_model.start_sessions(providers=[\"CPUExecutionProvider\"]) # cpu mode"
],
"metadata": {
"id": "nSeG9uAZrcph",
"colab": {
"base_uri": "https://localhost:8080/"
},
"outputId": "8c394684-d78e-49f6-a60f-872485d5f650"
},
"execution_count": 2,
"outputs": [
{
"output_type": "stream",
"name": "stdout",
"text": [
"[CLIP ONNX] Start convert visual model\n"
]
},
{
"output_type": "stream",
"name": "stderr",
"text": [
"/usr/local/lib/python3.7/dist-packages/clip_onnx/utils.py:40: UserWarning: __floordiv__ is deprecated, and its behavior will change in a future version of pytorch. It currently rounds toward 0 (like the 'trunc' function NOT 'floor'). This results in incorrect rounding for negative values. To keep the current behavior, use torch.div(a, b, rounding_mode='trunc'), or for actual floor division, use torch.div(a, b, rounding_mode='floor').\n",
" head_dim = q.shape[2] // num_heads\n",
"/usr/local/lib/python3.7/dist-packages/torch/onnx/symbolic_helper.py:716: UserWarning: allowzero=0 by default. In order to honor zero value in shape use allowzero=1\n",
" warnings.warn(\"allowzero=0 by default. In order to honor zero value in shape use allowzero=1\")\n"
]
},
{
"output_type": "stream",
"name": "stdout",
"text": [
"[CLIP ONNX] Start check visual model\n",
"[CLIP ONNX] Start convert textual model\n"
]
},
{
"output_type": "stream",
"name": "stderr",
"text": [
"/usr/local/lib/python3.7/dist-packages/torch/onnx/symbolic_opset9.py:2819: UserWarning: Exporting aten::index operator of advanced indexing in opset 14 is achieved by combination of multiple ONNX operators, including Reshape, Transpose, Concat, and Gather. If indices include negative values, the exported graph will produce incorrect results.\n",
" \"If indices include negative values, the exported graph will produce incorrect results.\")\n"
]
},
{
"output_type": "stream",
"name": "stdout",
"text": [
"[CLIP ONNX] Start check textual model\n",
"[CLIP ONNX] Models converts successfully\n"
]
}
]
},
{
"cell_type": "code",
"source": [
"%timeit onnx_model(image_onnx, text_onnx)"
],
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "B15dr51UrvMh",
"outputId": "7c5fbc64-61f5-4742-d5a1-24d123971515"
},
"execution_count": 5,
"outputs": [
{
"output_type": "stream",
"name": "stdout",
"text": [
"1 loop, best of 5: 550 ms per loop\n"
]
}
]
},
{
"cell_type": "markdown",
"source": [
"## GPU inference mode\n",
"Select a runtime GPU to continue:\n",
"\n",
"Click Runtime -> Change Runtime Type -> switch \"Harware accelerator\" to be GPU. Save it, and you maybe connect to GPU"
],
"metadata": {
"id": "Ahh_7CkTUb8y"
}
},
{
"cell_type": "markdown",
"source": [
"### CLIP-ONNX"
],
"metadata": {
"id": "B6M7yq7qceb5"
}
},
{
"cell_type": "code",
"source": [
"onnx_model.start_sessions(providers=[\"CUDAExecutionProvider\"]) # GPU mode"
],
"metadata": {
"id": "6LtPSZhfUd_m"
},
"execution_count": 6,
"outputs": []
},
{
"cell_type": "code",
"source": [
"onnx_model.visual_session.get_providers() # optional"
],
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "xE0VGt9sQwrf",
"outputId": "6feb4701-7b7f-437e-dc2f-c95c504dbb89"
},
"execution_count": 7,
"outputs": [
{
"output_type": "execute_result",
"data": {
"text/plain": [
"['CUDAExecutionProvider', 'CPUExecutionProvider']"
]
},
"metadata": {},
"execution_count": 7
}
]
},
{
"cell_type": "code",
"source": [
"%timeit onnx_model(image_onnx, text_onnx)"
],
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "iPUVzqmgcYas",
"outputId": "3e7c1526-6e38-4982-ca36-eabfc95c2ab9"
},
"execution_count": 9,
"outputs": [
{
"output_type": "stream",
"name": "stdout",
"text": [
"The slowest run took 79.70 times longer than the fastest. This could mean that an intermediate result is being cached.\n",
"1 loop, best of 5: 60.8 ms per loop\n"
]
}
]
},
{
"cell_type": "markdown",
"source": [
"### Torch CLIP"
],
"metadata": {
"id": "jb58mrkbch2V"
}
},
{
"cell_type": "code",
"source": [
"import clip\n",
"from PIL import Image\n",
"\n",
"device = \"cuda\"\n",
"# onnx cannot work with cuda\n",
"model, preprocess = clip.load(\"ViT-B/32\", device=device, jit=False)\n",
"# batch first\n",
"image = preprocess(Image.open(\"CLIP.png\")).unsqueeze(0).to(device) # [1, 3, 224, 224]\n",
"text = clip.tokenize([\"a diagram\", \"a dog\", \"a cat\"]).to(device) # [3, 77]"
],
"metadata": {
"id": "gidR99GOckyF"
},
"execution_count": 10,
"outputs": []
},
{
"cell_type": "code",
"source": [
"%timeit model(image, text)"
],
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "XpBrtjlOcwOC",
"outputId": "56375401-18a0-499b-f29b-c6e2d4d07e42"
},
"execution_count": 11,
"outputs": [
{
"output_type": "stream",
"name": "stdout",
"text": [
"10 loops, best of 5: 72.2 ms per loop\n"
]
}
]
}
]
}
================================================
FILE: examples/dev/clip_onnx_benchmark_cpu.ipynb
================================================
{
"nbformat": 4,
"nbformat_minor": 0,
"metadata": {
"colab": {
"name": "clip-onnx-benchmark-cpu.ipynb",
"provenance": [],
"authorship_tag": "ABX9TyNUvpypuYYk54s1lZecP8Pf",
"include_colab_link": true
},
"kernelspec": {
"name": "python3",
"display_name": "Python 3"
},
"language_info": {
"name": "python"
},
"accelerator": "GPU"
},
"cells": [
{
"cell_type": "markdown",
"metadata": {
"id": "view-in-github",
"colab_type": "text"
},
"source": [
""
]
},
{
"cell_type": "markdown",
"source": [
"## Restart colab session after installation\n",
"Reload the session if something doesn't work"
],
"metadata": {
"id": "fxPg_VvZuScV"
}
},
{
"cell_type": "code",
"execution_count": 1,
"metadata": {
"id": "al_QNjyFq6Jj"
},
"outputs": [],
"source": [
"%%capture\n",
"!pip install git+https://github.com/Lednik7/CLIP-ONNX.git@dev\n",
"!pip install git+https://github.com/openai/CLIP.git\n",
"!pip install onnxruntime-gpu"
]
},
{
"cell_type": "code",
"source": [
"%%capture\n",
"!wget -c -O CLIP.png https://github.com/openai/CLIP/blob/main/CLIP.png?raw=true"
],
"metadata": {
"id": "42eeJz9lTdJ6"
},
"execution_count": 1,
"outputs": []
},
{
"cell_type": "code",
"source": [
"!nvidia-smi"
],
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "XuauIZIBSEUX",
"outputId": "7e3fa9a5-2970-4bc1-81e5-9ec997a267a1"
},
"execution_count": 2,
"outputs": [
{
"output_type": "stream",
"name": "stdout",
"text": [
"Tue May 3 06:56:57 2022 \n",
"+-----------------------------------------------------------------------------+\n",
"| NVIDIA-SMI 460.32.03 Driver Version: 460.32.03 CUDA Version: 11.2 |\n",
"|-------------------------------+----------------------+----------------------+\n",
"| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |\n",
"| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |\n",
"| | | MIG M. |\n",
"|===============================+======================+======================|\n",
"| 0 Tesla T4 Off | 00000000:00:04.0 Off | 0 |\n",
"| N/A 47C P8 9W / 70W | 0MiB / 15109MiB | 0% Default |\n",
"| | | N/A |\n",
"+-------------------------------+----------------------+----------------------+\n",
" \n",
"+-----------------------------------------------------------------------------+\n",
"| Processes: |\n",
"| GPU GI CI PID Type Process name GPU Memory |\n",
"| ID ID Usage |\n",
"|=============================================================================|\n",
"| No running processes found |\n",
"+-----------------------------------------------------------------------------+\n"
]
}
]
},
{
"cell_type": "code",
"source": [
"import onnxruntime\n",
"print(onnxruntime.get_device())"
],
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "gqvxpdajRX5_",
"outputId": "4ad23904-186a-4e19-af9a-66538a70a3c8"
},
"execution_count": 3,
"outputs": [
{
"output_type": "stream",
"name": "stdout",
"text": [
"GPU\n"
]
}
]
},
{
"cell_type": "markdown",
"source": [
"## GPU inference mode\n",
"Select a runtime GPU to continue:\n",
"\n",
"Click Runtime -> Change Runtime Type -> switch \"Harware accelerator\" to be GPU. Save it, and you maybe connect to GPU"
],
"metadata": {
"id": "010k-ksVTjAu"
}
},
{
"cell_type": "markdown",
"source": [
"### Torch CLIP"
],
"metadata": {
"id": "KdTz0IJWVBqE"
}
},
{
"cell_type": "code",
"source": [
"import clip\n",
"from PIL import Image\n",
"import numpy as np\n",
"\n",
"# onnx cannot work with cuda\n",
"model, preprocess = clip.load(\"ViT-B/32\", device=\"cpu\", jit=False)\n",
"\n",
"# batch first\n",
"image = preprocess(Image.open(\"CLIP.png\")).unsqueeze(0) # [1, 3, 224, 224]\n",
"image_onnx = image.detach().cpu().numpy().astype(np.float32)\n",
"\n",
"# batch first\n",
"text = clip.tokenize([\"a diagram\", \"a dog\", \"a cat\"]) # [3, 77]\n",
"text_onnx = text.detach().cpu().numpy().astype(np.int32)"
],
"metadata": {
"id": "9ROPwKYurOhP"
},
"execution_count": 4,
"outputs": []
},
{
"cell_type": "markdown",
"source": [
"### CLIP-ONNX"
],
"metadata": {
"id": "Ao2MriaVVG6Y"
}
},
{
"cell_type": "code",
"source": [
"from clip_onnx import clip_onnx\n",
"\n",
"onnx_model = clip_onnx(model)\n",
"onnx_model.convert2onnx(image, text, verbose=True)\n",
"# ['TensorrtExecutionProvider', 'CUDAExecutionProvider', 'CPUExecutionProvider']\n",
"onnx_model.start_sessions(providers=[\"CPUExecutionProvider\"]) # GPU mode"
],
"metadata": {
"id": "nSeG9uAZrcph",
"colab": {
"base_uri": "https://localhost:8080/"
},
"outputId": "32e7fb6e-191a-4c3a-a8be-42ddf41ee62d"
},
"execution_count": 6,
"outputs": [
{
"output_type": "stream",
"name": "stdout",
"text": [
"[CLIP ONNX] Start convert visual model\n",
"[CLIP ONNX] Start check visual model\n",
"[CLIP ONNX] Start convert textual model\n"
]
},
{
"output_type": "stream",
"name": "stderr",
"text": [
"/usr/local/lib/python3.7/dist-packages/torch/onnx/symbolic_opset9.py:2909: UserWarning: Exporting aten::index operator of advanced indexing in opset 12 is achieved by combination of multiple ONNX operators, including Reshape, Transpose, Concat, and Gather. If indices include negative values, the exported graph will produce incorrect results.\n",
" \"If indices include negative values, the exported graph will produce incorrect results.\")\n"
]
},
{
"output_type": "stream",
"name": "stdout",
"text": [
"[CLIP ONNX] Start check textual model\n",
"[CLIP ONNX] Models converts successfully\n"
]
}
]
},
{
"cell_type": "code",
"source": [
"onnx_model = clip_onnx(model)\n",
"onnx_model.load_onnx(\"/content/clip_visual.onnx\",\n",
" \"/content/clip_textual.onnx\",\n",
" model.logit_scale.exp())\n",
"onnx_model.start_sessions(providers=[\"CPUExecutionProvider\"]) # GPU mode"
],
"metadata": {
"id": "PsDS7ty79zZf"
},
"execution_count": 7,
"outputs": []
},
{
"cell_type": "code",
"source": [
"onnx_model.visual_session.get_providers()"
],
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "aZsGJNrbNCYe",
"outputId": "27eec69c-6535-46e1-d98a-15836459149e"
},
"execution_count": 8,
"outputs": [
{
"output_type": "execute_result",
"data": {
"text/plain": [
"['CPUExecutionProvider']"
]
},
"metadata": {},
"execution_count": 8
}
]
},
{
"cell_type": "markdown",
"source": [
"## Benchmark"
],
"metadata": {
"id": "J5IcOG_6jAFz"
}
},
{
"cell_type": "code",
"source": [
"model, preprocess = clip.load(\"ViT-B/32\", device=\"cpu\", jit=False)"
],
"metadata": {
"id": "SJ_5_x7vLepK"
},
"execution_count": 9,
"outputs": []
},
{
"cell_type": "code",
"source": [
"model.eval()\n",
"for x in model.parameters():\n",
" x.requires_grad = False"
],
"metadata": {
"id": "OnOzZ3LMuubW"
},
"execution_count": 10,
"outputs": []
},
{
"cell_type": "code",
"source": [
"import numpy, random, torch"
],
"metadata": {
"id": "wDwqRRrTGKUS"
},
"execution_count": 11,
"outputs": []
},
{
"cell_type": "code",
"source": [
"def set_seed():\n",
" torch.manual_seed(12)\n",
" torch.cuda.manual_seed(12)\n",
" np.random.seed(12)\n",
" random.seed(12)\n",
"\n",
" torch.backends.cudnn.deterministic=True"
],
"metadata": {
"id": "9H17n_6gGJgT"
},
"execution_count": 12,
"outputs": []
},
{
"cell_type": "code",
"source": [
"import torch\n",
"import time\n",
"\n",
"n = 5\n",
"clip_results = {\"encode_image\": [],\n",
" \"encode_text\": []}\n",
"onnx_results = {\"encode_image\": [],\n",
" \"encode_text\": []}\n",
"for batch in [2, 8, 16, 32, 64]:\n",
" set_seed()\n",
" t_mean = []\n",
" for _ in range(n):\n",
" image_input = torch.randint(1, 255, (batch, 3, 224, 224))\n",
" image_input_onnx = image_input.detach().cpu().numpy().astype(np.float32)\n",
" t = time.time()\n",
" onnx_model.encode_image(image_input_onnx)\n",
" t_mean.append(time.time() - t)\n",
" print(\"onnx\", batch, \"encode_image\", round(sum(t_mean) / n, 3))\n",
" torch.cuda.empty_cache()\n",
" onnx_results[\"encode_image\"].append([batch, round(sum(t_mean) / n, 3)])\n",
"\n",
" set_seed()\n",
" with torch.inference_mode():\n",
" t_mean = []\n",
" for _ in range(n):\n",
" image_input = torch.randint(1, 255, (batch, 3, 224, 224))\n",
" t = time.time()\n",
" model.encode_image(image_input)\n",
" t_mean.append(time.time() - t)\n",
" print(\"torch\", batch, \"encode_image\", round(sum(t_mean) / n, 3))\n",
" torch.cuda.empty_cache()\n",
" clip_results[\"encode_image\"].append([batch, round(sum(t_mean) / n, 3)])\n",
"\n",
" set_seed()\n",
" t_mean = []\n",
" for _ in range(n):\n",
" text_input = torch.randint(320, 49407, (batch, 77))\n",
" text_input_onnx = text_input.detach().cpu().numpy().astype(np.int32)\n",
" t = time.time()\n",
" onnx_model.encode_text(text_input_onnx)\n",
" t_mean.append(time.time() - t)\n",
" print(\"onnx\", batch, \"encode_text\", round(sum(t_mean) / n, 3))\n",
" torch.cuda.empty_cache()\n",
" onnx_results[\"encode_text\"].append([batch, round(sum(t_mean) / n, 3)])\n",
"\n",
" set_seed()\n",
" with torch.inference_mode():\n",
" t_mean = []\n",
" for _ in range(n):\n",
" text_input = torch.randint(320, 49407, (batch, 77))\n",
" t = time.time()\n",
" model.encode_text(text_input)\n",
" t_mean.append(time.time() - t)\n",
" print(\"torch\", batch, \"encode_text\", round(sum(t_mean) / n, 3))\n",
" torch.cuda.empty_cache()\n",
" clip_results[\"encode_text\"].append([batch, round(sum(t_mean) / n, 3)])\n",
"\n",
" print(\"-\" * 78)"
],
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "4lFL6tzWjiWL",
"outputId": "45819718-619e-429c-9aa4-7e28b068b9a3"
},
"execution_count": 13,
"outputs": [
{
"output_type": "stream",
"name": "stdout",
"text": [
"onnx 2 encode_image 0.234\n",
"torch 2 encode_image 0.343\n",
"onnx 2 encode_text 0.162\n",
"torch 2 encode_text 0.243\n",
"------------------------------------------------------------------------------\n",
"onnx 8 encode_image 0.923\n",
"torch 8 encode_image 1.093\n",
"onnx 8 encode_text 0.656\n",
"torch 8 encode_text 0.831\n",
"------------------------------------------------------------------------------\n",
"onnx 16 encode_image 2.079\n",
"torch 16 encode_image 1.952\n",
"onnx 16 encode_text 1.288\n",
"torch 16 encode_text 1.523\n",
"------------------------------------------------------------------------------\n",
"onnx 32 encode_image 3.937\n",
"torch 32 encode_image 4.079\n",
"onnx 32 encode_text 2.658\n",
"torch 32 encode_text 3.015\n",
"------------------------------------------------------------------------------\n",
"onnx 64 encode_image 7.944\n",
"torch 64 encode_image 8.07\n",
"onnx 64 encode_text 5.567\n",
"torch 64 encode_text 6.212\n",
"------------------------------------------------------------------------------\n"
]
}
]
},
{
"cell_type": "code",
"source": [
"import pandas as pd"
],
"metadata": {
"id": "P2YhbE9v_4ci"
},
"execution_count": 14,
"outputs": []
},
{
"cell_type": "code",
"source": [
"pd.DataFrame({\"backend\": [\"onnx\", \"torch\"] * 5,\n",
" \"batch\": [2, 2, 8, 8, 16, 16, 32, 32, 64, 64],\n",
" \"encode_image\": [j[1] for i in zip(onnx_results[\"encode_image\"],\n",
" clip_results[\"encode_image\"]) for j in i],\n",
" \"encode_text\": [j[1] for i in zip(onnx_results[\"encode_text\"],\n",
" clip_results[\"encode_text\"]) for j in i]})"
],
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/",
"height": 362
},
"id": "WfZfDk4PAlqm",
"outputId": "38710ad6-09ae-4c48-fc20-1cdabf4c2a50"
},
"execution_count": 15,
"outputs": [
{
"output_type": "execute_result",
"data": {
"text/plain": [
" backend batch encode_image encode_text\n",
"0 onnx 2 0.234 0.162\n",
"1 torch 2 0.343 0.243\n",
"2 onnx 8 0.923 0.656\n",
"3 torch 8 1.093 0.831\n",
"4 onnx 16 2.079 1.288\n",
"5 torch 16 1.952 1.523\n",
"6 onnx 32 3.937 2.658\n",
"7 torch 32 4.079 3.015\n",
"8 onnx 64 7.944 5.567\n",
"9 torch 64 8.070 6.212"
],
"text/html": [
"\n",
" | \n", " | backend | \n", "batch | \n", "encode_image | \n", "encode_text | \n", "
|---|---|---|---|---|
| 0 | \n", "onnx | \n", "2 | \n", "0.234 | \n", "0.162 | \n", "
| 1 | \n", "torch | \n", "2 | \n", "0.343 | \n", "0.243 | \n", "
| 2 | \n", "onnx | \n", "8 | \n", "0.923 | \n", "0.656 | \n", "
| 3 | \n", "torch | \n", "8 | \n", "1.093 | \n", "0.831 | \n", "
| 4 | \n", "onnx | \n", "16 | \n", "2.079 | \n", "1.288 | \n", "
| 5 | \n", "torch | \n", "16 | \n", "1.952 | \n", "1.523 | \n", "
| 6 | \n", "onnx | \n", "32 | \n", "3.937 | \n", "2.658 | \n", "
| 7 | \n", "torch | \n", "32 | \n", "4.079 | \n", "3.015 | \n", "
| 8 | \n", "onnx | \n", "64 | \n", "7.944 | \n", "5.567 | \n", "
| 9 | \n", "torch | \n", "64 | \n", "8.070 | \n", "6.212 | \n", "
| \n", " | ONNX | \n", "batch | \n", "encode_image | \n", "encode_text | \n", "total | \n", "
|---|---|---|---|---|---|
| 0 | \n", "ViT-B/32 | \n", "2 | \n", "0.234 | \n", "0.162 | \n", "0.396 | \n", "
| 1 | \n", "ViT-B/32 | \n", "8 | \n", "0.923 | \n", "0.656 | \n", "1.579 | \n", "
| 2 | \n", "ViT-B/32 | \n", "16 | \n", "2.079 | \n", "1.288 | \n", "3.367 | \n", "
| 3 | \n", "ViT-B/32 | \n", "32 | \n", "3.937 | \n", "2.658 | \n", "6.595 | \n", "
| 4 | \n", "ViT-B/32 | \n", "64 | \n", "7.944 | \n", "5.567 | \n", "13.511 | \n", "
| \n", " | backend | \n", "batch | \n", "encode_image | \n", "encode_text | \n", "
|---|---|---|---|---|
| 0 | \n", "onnx | \n", "2 | \n", "0.073 | \n", "0.032 | \n", "
| 1 | \n", "torch | \n", "2 | \n", "0.041 | \n", "0.033 | \n", "
| 2 | \n", "onnx | \n", "8 | \n", "0.088 | \n", "0.052 | \n", "
| 3 | \n", "torch | \n", "8 | \n", "0.128 | \n", "0.102 | \n", "
| 4 | \n", "onnx | \n", "16 | \n", "0.123 | \n", "0.080 | \n", "
| 5 | \n", "torch | \n", "16 | \n", "0.258 | \n", "0.201 | \n", "
| 6 | \n", "onnx | \n", "32 | \n", "0.196 | \n", "0.138 | \n", "
| 7 | \n", "torch | \n", "32 | \n", "0.505 | \n", "0.386 | \n", "
| 8 | \n", "onnx | \n", "64 | \n", "0.352 | \n", "0.252 | \n", "
| 9 | \n", "torch | \n", "64 | \n", "0.995 | \n", "0.754 | \n", "
| \n", " | ONNX | \n", "batch | \n", "encode_image | \n", "encode_text | \n", "summary | \n", "
|---|---|---|---|---|---|
| 0 | \n", "ViT-B/32 | \n", "2 | \n", "0.073 | \n", "0.032 | \n", "0.105 | \n", "
| 1 | \n", "ViT-B/32 | \n", "8 | \n", "0.088 | \n", "0.052 | \n", "0.140 | \n", "
| 2 | \n", "ViT-B/32 | \n", "16 | \n", "0.123 | \n", "0.080 | \n", "0.203 | \n", "
| 3 | \n", "ViT-B/32 | \n", "32 | \n", "0.196 | \n", "0.138 | \n", "0.334 | \n", "
| 4 | \n", "ViT-B/32 | \n", "64 | \n", "0.352 | \n", "0.252 | \n", "0.604 | \n", "
"
]
},
{
"cell_type": "markdown",
"source": [
"## Restart colab session after installation\n",
"Reload the session if something doesn't work"
],
"metadata": {
"id": "fxPg_VvZuScV"
}
},
{
"cell_type": "code",
"execution_count": 2,
"metadata": {
"id": "al_QNjyFq6Jj"
},
"outputs": [],
"source": [
"%%capture\n",
"!pip install git+https://github.com/Lednik7/CLIP-ONNX.git@dev\n",
"!pip install git+https://github.com/openai/CLIP.git\n",
"!pip install onnxruntime-gpu"
]
},
{
"cell_type": "code",
"source": [
"%%capture\n",
"!wget -c -O CLIP.png https://github.com/openai/CLIP/blob/main/CLIP.png?raw=true"
],
"metadata": {
"id": "42eeJz9lTdJ6"
},
"execution_count": 3,
"outputs": []
},
{
"cell_type": "code",
"source": [
"!nvidia-smi"
],
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "XuauIZIBSEUX",
"outputId": "3bfb5833-272d-4aa0-f296-edab8122547c"
},
"execution_count": 1,
"outputs": [
{
"output_type": "stream",
"name": "stdout",
"text": [
"Tue May 3 07:20:58 2022 \n",
"+-----------------------------------------------------------------------------+\n",
"| NVIDIA-SMI 460.32.03 Driver Version: 460.32.03 CUDA Version: 11.2 |\n",
"|-------------------------------+----------------------+----------------------+\n",
"| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |\n",
"| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |\n",
"| | | MIG M. |\n",
"|===============================+======================+======================|\n",
"| 0 Tesla K80 Off | 00000000:00:04.0 Off | 0 |\n",
"| N/A 56C P8 29W / 149W | 0MiB / 11441MiB | 0% Default |\n",
"| | | N/A |\n",
"+-------------------------------+----------------------+----------------------+\n",
" \n",
"+-----------------------------------------------------------------------------+\n",
"| Processes: |\n",
"| GPU GI CI PID Type Process name GPU Memory |\n",
"| ID ID Usage |\n",
"|=============================================================================|\n",
"| No running processes found |\n",
"+-----------------------------------------------------------------------------+\n"
]
}
]
},
{
"cell_type": "code",
"source": [
"import onnxruntime\n",
"print(onnxruntime.get_device())"
],
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "gqvxpdajRX5_",
"outputId": "bb8e9195-fe9c-421c-e27b-d76da7136b82"
},
"execution_count": 2,
"outputs": [
{
"output_type": "stream",
"name": "stdout",
"text": [
"GPU\n"
]
}
]
},
{
"cell_type": "markdown",
"source": [
"## GPU inference mode\n",
"Select a runtime GPU to continue:\n",
"\n",
"Click Runtime -> Change Runtime Type -> switch \"Harware accelerator\" to be GPU. Save it, and you maybe connect to GPU"
],
"metadata": {
"id": "010k-ksVTjAu"
}
},
{
"cell_type": "markdown",
"source": [
"### Torch CLIP"
],
"metadata": {
"id": "KdTz0IJWVBqE"
}
},
{
"cell_type": "code",
"source": [
"import clip\n",
"from PIL import Image\n",
"import numpy as np\n",
"\n",
"# onnx cannot work with cuda\n",
"model, preprocess = clip.load(\"ViT-B/32\", device=\"cpu\", jit=False)\n",
"\n",
"# batch first\n",
"image = preprocess(Image.open(\"CLIP.png\")).unsqueeze(0) # [1, 3, 224, 224]\n",
"image_onnx = image.detach().cpu().numpy().astype(np.float32)\n",
"\n",
"# batch first\n",
"text = clip.tokenize([\"a diagram\", \"a dog\", \"a cat\"]) # [3, 77]\n",
"text_onnx = text.detach().cpu().numpy().astype(np.int32)"
],
"metadata": {
"id": "9ROPwKYurOhP"
},
"execution_count": 3,
"outputs": []
},
{
"cell_type": "markdown",
"source": [
"### CLIP-ONNX"
],
"metadata": {
"id": "Ao2MriaVVG6Y"
}
},
{
"cell_type": "code",
"source": [
"from clip_onnx import clip_onnx\n",
"from clip_onnx.utils import DEFAULT_EXPORT\n",
"\n",
"DEFAULT_EXPORT[\"opset_version\"] = 15\n",
"\n",
"onnx_model = clip_onnx(model)\n",
"onnx_model.convert2onnx(image, text, verbose=True)\n",
"# ['TensorrtExecutionProvider', 'CUDAExecutionProvider', 'CPUExecutionProvider']\n",
"onnx_model.start_sessions(providers=[\"CPUExecutionProvider\"]) # GPU mode"
],
"metadata": {
"id": "nSeG9uAZrcph",
"colab": {
"base_uri": "https://localhost:8080/"
},
"outputId": "1d4a8404-104f-4107-f2c4-e7e1f7b1d104"
},
"execution_count": 5,
"outputs": [
{
"output_type": "stream",
"name": "stdout",
"text": [
"[CLIP ONNX] Start convert visual model\n"
]
},
{
"output_type": "stream",
"name": "stderr",
"text": [
"/usr/local/lib/python3.7/dist-packages/torch/onnx/symbolic_helper.py:719: UserWarning: allowzero=0 by default. In order to honor zero value in shape use allowzero=1\n",
" warnings.warn(\"allowzero=0 by default. In order to honor zero value in shape use allowzero=1\")\n"
]
},
{
"output_type": "stream",
"name": "stdout",
"text": [
"[CLIP ONNX] Start check visual model\n",
"[CLIP ONNX] Start convert textual model\n"
]
},
{
"output_type": "stream",
"name": "stderr",
"text": [
"/usr/local/lib/python3.7/dist-packages/torch/onnx/symbolic_opset9.py:2909: UserWarning: Exporting aten::index operator of advanced indexing in opset 15 is achieved by combination of multiple ONNX operators, including Reshape, Transpose, Concat, and Gather. If indices include negative values, the exported graph will produce incorrect results.\n",
" \"If indices include negative values, the exported graph will produce incorrect results.\")\n"
]
},
{
"output_type": "stream",
"name": "stdout",
"text": [
"[CLIP ONNX] Start check textual model\n",
"[CLIP ONNX] Models converts successfully\n"
]
}
]
},
{
"cell_type": "code",
"source": [
"onnx_model = clip_onnx(model)\n",
"onnx_model.load_onnx(\"/content/clip_visual.onnx\",\n",
" \"/content/clip_textual.onnx\",\n",
" model.logit_scale.exp())\n",
"onnx_model.start_sessions(providers=[\"CUDAExecutionProvider\"]) # GPU mode"
],
"metadata": {
"id": "PsDS7ty79zZf"
},
"execution_count": 6,
"outputs": []
},
{
"cell_type": "code",
"source": [
"onnx_model.visual_session.get_providers()"
],
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "aZsGJNrbNCYe",
"outputId": "b0ee40a7-2ece-4e88-9e35-9ed0a735c533"
},
"execution_count": 7,
"outputs": [
{
"output_type": "execute_result",
"data": {
"text/plain": [
"['CUDAExecutionProvider', 'CPUExecutionProvider']"
]
},
"metadata": {},
"execution_count": 7
}
]
},
{
"cell_type": "markdown",
"source": [
"## Benchmark"
],
"metadata": {
"id": "J5IcOG_6jAFz"
}
},
{
"cell_type": "code",
"source": [
"model, preprocess = clip.load(\"ViT-B/32\", device=\"cuda\", jit=False)"
],
"metadata": {
"id": "SJ_5_x7vLepK"
},
"execution_count": 8,
"outputs": []
},
{
"cell_type": "code",
"source": [
"model.eval()\n",
"for x in model.parameters():\n",
" x.requires_grad = False"
],
"metadata": {
"id": "OnOzZ3LMuubW"
},
"execution_count": 9,
"outputs": []
},
{
"cell_type": "code",
"source": [
"import numpy, random, torch"
],
"metadata": {
"id": "wDwqRRrTGKUS"
},
"execution_count": 10,
"outputs": []
},
{
"cell_type": "code",
"source": [
"def set_seed():\n",
" torch.manual_seed(12)\n",
" torch.cuda.manual_seed(12)\n",
" np.random.seed(12)\n",
" random.seed(12)\n",
"\n",
" torch.backends.cudnn.deterministic=True"
],
"metadata": {
"id": "9H17n_6gGJgT"
},
"execution_count": 11,
"outputs": []
},
{
"cell_type": "code",
"source": [
"import torch\n",
"import time\n",
"\n",
"n = 5\n",
"clip_results = {\"encode_image\": [],\n",
" \"encode_text\": []}\n",
"onnx_results = {\"encode_image\": [],\n",
" \"encode_text\": []}\n",
"for batch in [2, 8, 16, 32, 64]:\n",
" set_seed()\n",
" t_mean = []\n",
" for _ in range(n):\n",
" image_input = torch.randint(1, 255, (batch, 3, 224, 224))\n",
" image_input_onnx = image_input.detach().cpu().numpy().astype(np.float32)\n",
" t = time.time()\n",
" onnx_model.encode_image(image_input_onnx)\n",
" t_mean.append(time.time() - t)\n",
" print(\"onnx\", batch, \"encode_image\", round(sum(t_mean) / n, 3))\n",
" torch.cuda.empty_cache()\n",
" onnx_results[\"encode_image\"].append([batch, round(sum(t_mean) / n, 3)])\n",
"\n",
" set_seed()\n",
" with torch.inference_mode():\n",
" t_mean = []\n",
" for _ in range(n):\n",
" image_input = torch.randint(1, 255, (batch, 3, 224, 224)).cuda()\n",
" t = time.time()\n",
" model.encode_image(image_input)\n",
" t_mean.append(time.time() - t)\n",
" print(\"torch\", batch, \"encode_image\", round(sum(t_mean) / n, 3))\n",
" torch.cuda.empty_cache()\n",
" clip_results[\"encode_image\"].append([batch, round(sum(t_mean) / n, 3)])\n",
"\n",
" set_seed()\n",
" t_mean = []\n",
" for _ in range(n):\n",
" text_input = torch.randint(320, 49407, (batch, 77))\n",
" text_input_onnx = text_input.detach().cpu().numpy().astype(np.int32)\n",
" t = time.time()\n",
" onnx_model.encode_text(text_input_onnx)\n",
" t_mean.append(time.time() - t)\n",
" print(\"onnx\", batch, \"encode_text\", round(sum(t_mean) / n, 3))\n",
" torch.cuda.empty_cache()\n",
" onnx_results[\"encode_text\"].append([batch, round(sum(t_mean) / n, 3)])\n",
"\n",
" set_seed()\n",
" with torch.inference_mode():\n",
" t_mean = []\n",
" for _ in range(n):\n",
" text_input = torch.randint(320, 49407, (batch, 77)).cuda()\n",
" t = time.time()\n",
" model.encode_text(text_input)\n",
" t_mean.append(time.time() - t)\n",
" print(\"torch\", batch, \"encode_text\", round(sum(t_mean) / n, 3))\n",
" torch.cuda.empty_cache()\n",
" clip_results[\"encode_text\"].append([batch, round(sum(t_mean) / n, 3)])\n",
"\n",
" print(\"-\" * 78)"
],
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "4lFL6tzWjiWL",
"outputId": "ccaa7e0a-96f3-4a51-c4bd-c442aa13763c"
},
"execution_count": 12,
"outputs": [
{
"output_type": "stream",
"name": "stdout",
"text": [
"onnx 2 encode_image 0.136\n",
"torch 2 encode_image 0.02\n",
"onnx 2 encode_text 0.021\n",
"torch 2 encode_text 0.035\n",
"------------------------------------------------------------------------------\n",
"onnx 8 encode_image 0.054\n",
"torch 8 encode_image 0.081\n",
"onnx 8 encode_text 0.04\n",
"torch 8 encode_text 0.098\n",
"------------------------------------------------------------------------------\n",
"onnx 16 encode_image 0.089\n",
"torch 16 encode_image 0.207\n",
"onnx 16 encode_text 0.071\n",
"torch 16 encode_text 0.196\n",
"------------------------------------------------------------------------------\n",
"onnx 32 encode_image 0.158\n",
"torch 32 encode_image 0.44\n",
"onnx 32 encode_text 0.134\n",
"torch 32 encode_text 0.374\n",
"------------------------------------------------------------------------------\n",
"onnx 64 encode_image 0.325\n",
"torch 64 encode_image 0.919\n",
"onnx 64 encode_text 0.258\n",
"torch 64 encode_text 0.719\n",
"------------------------------------------------------------------------------\n"
]
}
]
},
{
"cell_type": "code",
"source": [
"import pandas as pd"
],
"metadata": {
"id": "P2YhbE9v_4ci"
},
"execution_count": 13,
"outputs": []
},
{
"cell_type": "code",
"source": [
"pd.DataFrame({\"backend\": [\"onnx\", \"torch\"] * 5,\n",
" \"batch\": [2, 2, 8, 8, 16, 16, 32, 32, 64, 64],\n",
" \"encode_image\": [j[1] for i in zip(onnx_results[\"encode_image\"],\n",
" clip_results[\"encode_image\"]) for j in i],\n",
" \"encode_text\": [j[1] for i in zip(onnx_results[\"encode_text\"],\n",
" clip_results[\"encode_text\"]) for j in i]})"
],
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/",
"height": 362
},
"id": "WfZfDk4PAlqm",
"outputId": "78a5cae8-68ee-4edd-f34d-ccf7d3d8a23b"
},
"execution_count": 14,
"outputs": [
{
"output_type": "execute_result",
"data": {
"text/plain": [
" backend batch encode_image encode_text\n",
"0 onnx 2 0.136 0.021\n",
"1 torch 2 0.020 0.035\n",
"2 onnx 8 0.054 0.040\n",
"3 torch 8 0.081 0.098\n",
"4 onnx 16 0.089 0.071\n",
"5 torch 16 0.207 0.196\n",
"6 onnx 32 0.158 0.134\n",
"7 torch 32 0.440 0.374\n",
"8 onnx 64 0.325 0.258\n",
"9 torch 64 0.919 0.719"
],
"text/html": [
"\n",
" | \n", " | backend | \n", "batch | \n", "encode_image | \n", "encode_text | \n", "
|---|---|---|---|---|
| 0 | \n", "onnx | \n", "2 | \n", "0.136 | \n", "0.021 | \n", "
| 1 | \n", "torch | \n", "2 | \n", "0.020 | \n", "0.035 | \n", "
| 2 | \n", "onnx | \n", "8 | \n", "0.054 | \n", "0.040 | \n", "
| 3 | \n", "torch | \n", "8 | \n", "0.081 | \n", "0.098 | \n", "
| 4 | \n", "onnx | \n", "16 | \n", "0.089 | \n", "0.071 | \n", "
| 5 | \n", "torch | \n", "16 | \n", "0.207 | \n", "0.196 | \n", "
| 6 | \n", "onnx | \n", "32 | \n", "0.158 | \n", "0.134 | \n", "
| 7 | \n", "torch | \n", "32 | \n", "0.440 | \n", "0.374 | \n", "
| 8 | \n", "onnx | \n", "64 | \n", "0.325 | \n", "0.258 | \n", "
| 9 | \n", "torch | \n", "64 | \n", "0.919 | \n", "0.719 | \n", "
| \n", " | ONNX | \n", "batch | \n", "encode_image | \n", "encode_text | \n", "total | \n", "
|---|---|---|---|---|---|
| 0 | \n", "ViT-B/32 | \n", "2 | \n", "0.136 | \n", "0.021 | \n", "0.157 | \n", "
| 1 | \n", "ViT-B/32 | \n", "8 | \n", "0.054 | \n", "0.040 | \n", "0.094 | \n", "
| 2 | \n", "ViT-B/32 | \n", "16 | \n", "0.089 | \n", "0.071 | \n", "0.160 | \n", "
| 3 | \n", "ViT-B/32 | \n", "32 | \n", "0.158 | \n", "0.134 | \n", "0.292 | \n", "
| 4 | \n", "ViT-B/32 | \n", "64 | \n", "0.325 | \n", "0.258 | \n", "0.583 | \n", "
"
]
},
{
"cell_type": "markdown",
"source": [
"## Restart colab session after installation\n",
"Reload the session if something doesn't work"
],
"metadata": {
"id": "fxPg_VvZuScV"
}
},
{
"cell_type": "code",
"execution_count": 1,
"metadata": {
"id": "al_QNjyFq6Jj"
},
"outputs": [],
"source": [
"%%capture\n",
"!pip install git+https://github.com/Lednik7/CLIP-ONNX.git@dev\n",
"!pip install git+https://github.com/openai/CLIP.git\n",
"!pip install onnxruntime-gpu"
]
},
{
"cell_type": "code",
"source": [
"%%capture\n",
"!wget -c -O CLIP.png https://github.com/openai/CLIP/blob/main/CLIP.png?raw=true"
],
"metadata": {
"id": "42eeJz9lTdJ6"
},
"execution_count": 1,
"outputs": []
},
{
"cell_type": "code",
"source": [
"!nvidia-smi"
],
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "XuauIZIBSEUX",
"outputId": "3e459c2c-8f31-4aff-c288-f2e6c4684e36"
},
"execution_count": 1,
"outputs": [
{
"output_type": "stream",
"name": "stdout",
"text": [
"Tue May 3 07:10:09 2022 \n",
"+-----------------------------------------------------------------------------+\n",
"| NVIDIA-SMI 460.32.03 Driver Version: 460.32.03 CUDA Version: 11.2 |\n",
"|-------------------------------+----------------------+----------------------+\n",
"| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |\n",
"| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |\n",
"| | | MIG M. |\n",
"|===============================+======================+======================|\n",
"| 0 Tesla T4 Off | 00000000:00:04.0 Off | 0 |\n",
"| N/A 38C P8 9W / 70W | 0MiB / 15109MiB | 0% Default |\n",
"| | | N/A |\n",
"+-------------------------------+----------------------+----------------------+\n",
" \n",
"+-----------------------------------------------------------------------------+\n",
"| Processes: |\n",
"| GPU GI CI PID Type Process name GPU Memory |\n",
"| ID ID Usage |\n",
"|=============================================================================|\n",
"| No running processes found |\n",
"+-----------------------------------------------------------------------------+\n"
]
}
]
},
{
"cell_type": "code",
"source": [
"import onnxruntime\n",
"print(onnxruntime.get_device())"
],
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "gqvxpdajRX5_",
"outputId": "48a89abb-a326-4563-f99a-40c7d25145af"
},
"execution_count": 2,
"outputs": [
{
"output_type": "stream",
"name": "stdout",
"text": [
"GPU\n"
]
}
]
},
{
"cell_type": "markdown",
"source": [
"## GPU inference mode\n",
"Select a runtime GPU to continue:\n",
"\n",
"Click Runtime -> Change Runtime Type -> switch \"Harware accelerator\" to be GPU. Save it, and you maybe connect to GPU"
],
"metadata": {
"id": "010k-ksVTjAu"
}
},
{
"cell_type": "markdown",
"source": [
"### Torch CLIP"
],
"metadata": {
"id": "KdTz0IJWVBqE"
}
},
{
"cell_type": "code",
"source": [
"import clip\n",
"from PIL import Image\n",
"import numpy as np\n",
"\n",
"# onnx cannot work with cuda\n",
"model, preprocess = clip.load(\"ViT-B/32\", device=\"cpu\", jit=False)\n",
"\n",
"# batch first\n",
"image = preprocess(Image.open(\"CLIP.png\")).unsqueeze(0) # [1, 3, 224, 224]\n",
"image_onnx = image.detach().cpu().numpy().astype(np.float32)\n",
"\n",
"# batch first\n",
"text = clip.tokenize([\"a diagram\", \"a dog\", \"a cat\"]) # [3, 77]\n",
"text_onnx = text.detach().cpu().numpy().astype(np.int32)"
],
"metadata": {
"id": "9ROPwKYurOhP"
},
"execution_count": 3,
"outputs": []
},
{
"cell_type": "markdown",
"source": [
"### CLIP-ONNX"
],
"metadata": {
"id": "Ao2MriaVVG6Y"
}
},
{
"cell_type": "code",
"source": [
"from clip_onnx import clip_onnx\n",
"\n",
"onnx_model = clip_onnx(model)\n",
"onnx_model.convert2onnx(image, text, verbose=True)\n",
"# ['TensorrtExecutionProvider', 'CUDAExecutionProvider', 'CPUExecutionProvider']\n",
"onnx_model.start_sessions(providers=[\"CPUExecutionProvider\"]) # GPU mode"
],
"metadata": {
"id": "nSeG9uAZrcph",
"colab": {
"base_uri": "https://localhost:8080/"
},
"outputId": "1186f909-6cfb-400b-c2d9-3dddc93d318b"
},
"execution_count": 4,
"outputs": [
{
"output_type": "stream",
"name": "stdout",
"text": [
"[CLIP ONNX] Start convert visual model\n",
"[CLIP ONNX] Start check visual model\n",
"[CLIP ONNX] Start convert textual model\n"
]
},
{
"output_type": "stream",
"name": "stderr",
"text": [
"/usr/local/lib/python3.7/dist-packages/torch/onnx/symbolic_opset9.py:2909: UserWarning: Exporting aten::index operator of advanced indexing in opset 12 is achieved by combination of multiple ONNX operators, including Reshape, Transpose, Concat, and Gather. If indices include negative values, the exported graph will produce incorrect results.\n",
" \"If indices include negative values, the exported graph will produce incorrect results.\")\n"
]
},
{
"output_type": "stream",
"name": "stdout",
"text": [
"[CLIP ONNX] Start check textual model\n",
"[CLIP ONNX] Models converts successfully\n"
]
}
]
},
{
"cell_type": "code",
"source": [
"onnx_model = clip_onnx(model)\n",
"onnx_model.load_onnx(\"/content/clip_visual.onnx\",\n",
" \"/content/clip_textual.onnx\",\n",
" model.logit_scale.exp())\n",
"onnx_model.start_sessions(providers=[\"CUDAExecutionProvider\"]) # GPU mode"
],
"metadata": {
"id": "PsDS7ty79zZf"
},
"execution_count": 5,
"outputs": []
},
{
"cell_type": "code",
"source": [
"onnx_model.visual_session.get_providers()"
],
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "aZsGJNrbNCYe",
"outputId": "05464d1a-7047-4efd-80fe-32870cf34afd"
},
"execution_count": 6,
"outputs": [
{
"output_type": "execute_result",
"data": {
"text/plain": [
"['CUDAExecutionProvider', 'CPUExecutionProvider']"
]
},
"metadata": {},
"execution_count": 6
}
]
},
{
"cell_type": "markdown",
"source": [
"## Benchmark"
],
"metadata": {
"id": "J5IcOG_6jAFz"
}
},
{
"cell_type": "code",
"source": [
"model, preprocess = clip.load(\"ViT-B/32\", device=\"cuda\", jit=False)"
],
"metadata": {
"id": "SJ_5_x7vLepK"
},
"execution_count": 7,
"outputs": []
},
{
"cell_type": "code",
"source": [
"model.eval()\n",
"for x in model.parameters():\n",
" x.requires_grad = False"
],
"metadata": {
"id": "OnOzZ3LMuubW"
},
"execution_count": 8,
"outputs": []
},
{
"cell_type": "code",
"source": [
"import numpy, random, torch"
],
"metadata": {
"id": "wDwqRRrTGKUS"
},
"execution_count": 9,
"outputs": []
},
{
"cell_type": "code",
"source": [
"def set_seed():\n",
" torch.manual_seed(12)\n",
" torch.cuda.manual_seed(12)\n",
" np.random.seed(12)\n",
" random.seed(12)\n",
"\n",
" torch.backends.cudnn.deterministic=True"
],
"metadata": {
"id": "9H17n_6gGJgT"
},
"execution_count": 10,
"outputs": []
},
{
"cell_type": "code",
"source": [
"import torch\n",
"import time\n",
"\n",
"n = 5\n",
"clip_results = {\"encode_image\": [],\n",
" \"encode_text\": []}\n",
"onnx_results = {\"encode_image\": [],\n",
" \"encode_text\": []}\n",
"for batch in [2, 8, 16, 32, 64]:\n",
" set_seed()\n",
" t_mean = []\n",
" for _ in range(n):\n",
" image_input = torch.randint(1, 255, (batch, 3, 224, 224))\n",
" image_input_onnx = image_input.detach().cpu().numpy().astype(np.float32)\n",
" t = time.time()\n",
" onnx_model.encode_image(image_input_onnx)\n",
" t_mean.append(time.time() - t)\n",
" print(\"onnx\", batch, \"encode_image\", round(sum(t_mean) / n, 3))\n",
" torch.cuda.empty_cache()\n",
" onnx_results[\"encode_image\"].append([batch, round(sum(t_mean) / n, 3)])\n",
"\n",
" set_seed()\n",
" with torch.inference_mode():\n",
" t_mean = []\n",
" for _ in range(n):\n",
" image_input = torch.randint(1, 255, (batch, 3, 224, 224)).cuda()\n",
" t = time.time()\n",
" model.encode_image(image_input)\n",
" t_mean.append(time.time() - t)\n",
" print(\"torch\", batch, \"encode_image\", round(sum(t_mean) / n, 3))\n",
" torch.cuda.empty_cache()\n",
" clip_results[\"encode_image\"].append([batch, round(sum(t_mean) / n, 3)])\n",
"\n",
" set_seed()\n",
" t_mean = []\n",
" for _ in range(n):\n",
" text_input = torch.randint(320, 49407, (batch, 77))\n",
" text_input_onnx = text_input.detach().cpu().numpy().astype(np.int32)\n",
" t = time.time()\n",
" onnx_model.encode_text(text_input_onnx)\n",
" t_mean.append(time.time() - t)\n",
" print(\"onnx\", batch, \"encode_text\", round(sum(t_mean) / n, 3))\n",
" torch.cuda.empty_cache()\n",
" onnx_results[\"encode_text\"].append([batch, round(sum(t_mean) / n, 3)])\n",
"\n",
" set_seed()\n",
" with torch.inference_mode():\n",
" t_mean = []\n",
" for _ in range(n):\n",
" text_input = torch.randint(320, 49407, (batch, 77)).cuda()\n",
" t = time.time()\n",
" model.encode_text(text_input)\n",
" t_mean.append(time.time() - t)\n",
" print(\"torch\", batch, \"encode_text\", round(sum(t_mean) / n, 3))\n",
" torch.cuda.empty_cache()\n",
" clip_results[\"encode_text\"].append([batch, round(sum(t_mean) / n, 3)])\n",
"\n",
" print(\"-\" * 78)"
],
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "4lFL6tzWjiWL",
"outputId": "c2b9f0e4-9b93-408b-96bf-3fdb3057e15b"
},
"execution_count": 11,
"outputs": [
{
"output_type": "stream",
"name": "stdout",
"text": [
"onnx 2 encode_image 0.155\n",
"torch 2 encode_image 0.017\n",
"onnx 2 encode_text 0.01\n",
"torch 2 encode_text 0.009\n",
"------------------------------------------------------------------------------\n",
"onnx 8 encode_image 0.032\n",
"torch 8 encode_image 0.008\n",
"onnx 8 encode_text 0.014\n",
"torch 8 encode_text 0.008\n",
"------------------------------------------------------------------------------\n",
"onnx 16 encode_image 0.037\n",
"torch 16 encode_image 0.009\n",
"onnx 16 encode_text 0.029\n",
"torch 16 encode_text 0.012\n",
"------------------------------------------------------------------------------\n",
"onnx 32 encode_image 0.076\n",
"torch 32 encode_image 0.008\n",
"onnx 32 encode_text 0.059\n",
"torch 32 encode_text 0.025\n",
"------------------------------------------------------------------------------\n",
"onnx 64 encode_image 0.169\n",
"torch 64 encode_image 0.009\n",
"onnx 64 encode_text 0.117\n",
"torch 64 encode_text 0.049\n",
"------------------------------------------------------------------------------\n"
]
}
]
},
{
"cell_type": "code",
"source": [
"import pandas as pd"
],
"metadata": {
"id": "P2YhbE9v_4ci"
},
"execution_count": 12,
"outputs": []
},
{
"cell_type": "code",
"source": [
"pd.DataFrame({\"backend\": [\"onnx\", \"torch\"] * 5,\n",
" \"batch\": [2, 2, 8, 8, 16, 16, 32, 32, 64, 64],\n",
" \"encode_image\": [j[1] for i in zip(onnx_results[\"encode_image\"],\n",
" clip_results[\"encode_image\"]) for j in i],\n",
" \"encode_text\": [j[1] for i in zip(onnx_results[\"encode_text\"],\n",
" clip_results[\"encode_text\"]) for j in i]})"
],
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/",
"height": 362
},
"id": "WfZfDk4PAlqm",
"outputId": "3375eac7-47b0-40ba-c2d6-c30fda2ab6d5"
},
"execution_count": 13,
"outputs": [
{
"output_type": "execute_result",
"data": {
"text/plain": [
" backend batch encode_image encode_text\n",
"0 onnx 2 0.155 0.010\n",
"1 torch 2 0.017 0.009\n",
"2 onnx 8 0.032 0.014\n",
"3 torch 8 0.008 0.008\n",
"4 onnx 16 0.037 0.029\n",
"5 torch 16 0.009 0.012\n",
"6 onnx 32 0.076 0.059\n",
"7 torch 32 0.008 0.025\n",
"8 onnx 64 0.169 0.117\n",
"9 torch 64 0.009 0.049"
],
"text/html": [
"\n",
" | \n", " | backend | \n", "batch | \n", "encode_image | \n", "encode_text | \n", "
|---|---|---|---|---|
| 0 | \n", "onnx | \n", "2 | \n", "0.155 | \n", "0.010 | \n", "
| 1 | \n", "torch | \n", "2 | \n", "0.017 | \n", "0.009 | \n", "
| 2 | \n", "onnx | \n", "8 | \n", "0.032 | \n", "0.014 | \n", "
| 3 | \n", "torch | \n", "8 | \n", "0.008 | \n", "0.008 | \n", "
| 4 | \n", "onnx | \n", "16 | \n", "0.037 | \n", "0.029 | \n", "
| 5 | \n", "torch | \n", "16 | \n", "0.009 | \n", "0.012 | \n", "
| 6 | \n", "onnx | \n", "32 | \n", "0.076 | \n", "0.059 | \n", "
| 7 | \n", "torch | \n", "32 | \n", "0.008 | \n", "0.025 | \n", "
| 8 | \n", "onnx | \n", "64 | \n", "0.169 | \n", "0.117 | \n", "
| 9 | \n", "torch | \n", "64 | \n", "0.009 | \n", "0.049 | \n", "
| \n", " | ONNX | \n", "batch | \n", "encode_image | \n", "encode_text | \n", "total | \n", "
|---|---|---|---|---|---|
| 0 | \n", "ViT-B/32 | \n", "2 | \n", "0.155 | \n", "0.010 | \n", "0.165 | \n", "
| 1 | \n", "ViT-B/32 | \n", "8 | \n", "0.032 | \n", "0.014 | \n", "0.046 | \n", "
| 2 | \n", "ViT-B/32 | \n", "16 | \n", "0.037 | \n", "0.029 | \n", "0.066 | \n", "
| 3 | \n", "ViT-B/32 | \n", "32 | \n", "0.076 | \n", "0.059 | \n", "0.135 | \n", "
| 4 | \n", "ViT-B/32 | \n", "64 | \n", "0.169 | \n", "0.117 | \n", "0.286 | \n", "
"
]
},
{
"cell_type": "markdown",
"source": [
"## Restart colab session after installation\n",
"Reload the session if something doesn't work"
],
"metadata": {
"id": "whlsBiJgR8le"
}
},
{
"cell_type": "code",
"source": [
"%%capture\n",
"!pip install git+https://github.com/Lednik7/CLIP-ONNX.git\n",
"!pip install git+https://github.com/openai/CLIP.git\n",
"!pip install onnxruntime-gpu"
],
"metadata": {
"id": "HnbpAkvuR73L"
},
"execution_count": 1,
"outputs": []
},
{
"cell_type": "code",
"source": [
"%%capture\n",
"!wget -c -O CLIP.png https://github.com/openai/CLIP/blob/main/CLIP.png?raw=true"
],
"metadata": {
"id": "tqy0zKM4R-7M"
},
"execution_count": 2,
"outputs": []
},
{
"cell_type": "code",
"source": [
"!nvidia-smi # CPU Provider"
],
"metadata": {
"id": "eKqETHL4YscZ",
"outputId": "7ff0bc18-fb40-4296-ab05-b079043e46a1",
"colab": {
"base_uri": "https://localhost:8080/"
}
},
"execution_count": 3,
"outputs": [
{
"output_type": "stream",
"name": "stdout",
"text": [
"NVIDIA-SMI has failed because it couldn't communicate with the NVIDIA driver. Make sure that the latest NVIDIA driver is installed and running.\n",
"\n"
]
}
]
},
{
"cell_type": "code",
"source": [
"import onnxruntime\n",
"\n",
"print(onnxruntime.get_device()) # priority device"
],
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "x8IN72OnSAIh",
"outputId": "81d14047-91fa-4a5c-a1e3-f5b550556591"
},
"execution_count": 4,
"outputs": [
{
"output_type": "stream",
"name": "stdout",
"text": [
"CPU\n"
]
}
]
},
{
"cell_type": "markdown",
"source": [
"## CPU inference mode"
],
"metadata": {
"id": "U1Pr-YTtSEhs"
}
},
{
"cell_type": "code",
"source": [
"import warnings\n",
"\n",
"warnings.filterwarnings(\"ignore\", category=UserWarning)"
],
"metadata": {
"id": "gZTxanR26knr"
},
"execution_count": 5,
"outputs": []
},
{
"cell_type": "code",
"source": [
"import clip\n",
"from PIL import Image\n",
"import numpy as np\n",
"\n",
"# onnx cannot export with cuda\n",
"model, preprocess = clip.load(\"ViT-B/32\", device=\"cpu\", jit=False)\n",
"\n",
"# batch first\n",
"image = preprocess(Image.open(\"CLIP.png\")).unsqueeze(0).cpu() # [1, 3, 224, 224]\n",
"image_onnx = image.detach().cpu().numpy().astype(np.float32)\n",
"\n",
"# batch first\n",
"text = clip.tokenize([\"a diagram\", \"a dog\", \"a cat\"]).cpu() # [3, 77]\n",
"text_onnx = text.detach().cpu().numpy().astype(np.int32)"
],
"metadata": {
"id": "rPwc6A2SSGyl"
},
"execution_count": 6,
"outputs": []
},
{
"cell_type": "code",
"source": [
"from clip_onnx import clip_onnx, attention\n",
"# clip.model.ResidualAttentionBlock.attention = attention\n",
"\n",
"visual_path = \"clip_visual.onnx\"\n",
"textual_path = \"clip_textual.onnx\"\n",
"\n",
"onnx_model = clip_onnx(model, visual_path=visual_path, textual_path=textual_path)\n",
"onnx_model.convert2onnx(image, text, verbose=True)\n",
"# ['TensorrtExecutionProvider', 'CUDAExecutionProvider', 'CPUExecutionProvider']\n",
"onnx_model.start_sessions(providers=[\"CPUExecutionProvider\"]) # cpu mode"
],
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "oYM5FDSGSJBW",
"outputId": "816705b1-3829-4424-c7c4-5426cf21cc18"
},
"execution_count": 7,
"outputs": [
{
"output_type": "stream",
"name": "stdout",
"text": [
"[CLIP ONNX] Start convert visual model\n",
"[CLIP ONNX] Start check visual model\n",
"[CLIP ONNX] Start convert textual model\n",
"[CLIP ONNX] Start check textual model\n",
"[CLIP ONNX] Models converts successfully\n"
]
}
]
},
{
"cell_type": "code",
"source": [
"image_features = onnx_model.encode_image(image_onnx)\n",
"text_features = onnx_model.encode_text(text_onnx)\n",
"\n",
"logits_per_image, logits_per_text = onnx_model(image_onnx, text_onnx)\n",
"probs = logits_per_image.softmax(dim=-1).detach().cpu().numpy()\n",
"\n",
"print(\"Label probs:\", probs) # prints: [[0.9927937 0.00421067 0.00299571]]"
],
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "tYVuk72nSLw6",
"outputId": "41608059-3732-4ea7-c619-66f803af4185"
},
"execution_count": 8,
"outputs": [
{
"output_type": "stream",
"name": "stdout",
"text": [
"Label probs: [[0.9927937 0.00421067 0.00299571]]\n"
]
}
]
}
]
}
================================================
FILE: examples/ru_CLIP_tiny_onnx.ipynb
================================================
{
"nbformat": 4,
"nbformat_minor": 0,
"metadata": {
"colab": {
"name": "ru_CLIP_tiny_onnx.ipynb",
"provenance": [],
"collapsed_sections": [
"WWXCt_2NLhN_",
"PHb4CAoRL3qC",
"re2sSYAYO3D-",
"ithu4-z0PIm5",
"FWm0GAhWPzSW"
],
"machine_shape": "hm"
},
"kernelspec": {
"name": "python3",
"display_name": "Python 3"
},
"language_info": {
"name": "python"
},
"accelerator": "GPU",
"widgets": {
"application/vnd.jupyter.widget-state+json": {
"5319c7971f234d4bb615508f76475f9e": {
"model_module": "@jupyter-widgets/controls",
"model_name": "HBoxModel",
"model_module_version": "1.5.0",
"state": {
"_view_name": "HBoxView",
"_dom_classes": [],
"_model_name": "HBoxModel",
"_view_module": "@jupyter-widgets/controls",
"_model_module_version": "1.5.0",
"_view_count": null,
"_view_module_version": "1.5.0",
"box_style": "",
"layout": "IPY_MODEL_c43027a0735e459ca1f710e5a9c43177",
"_model_module": "@jupyter-widgets/controls",
"children": [
"IPY_MODEL_c00c959249db4f2a9b97adabd2684c3c",
"IPY_MODEL_d9e4edd05c1e40f991eb2c2f1fc9ebc1",
"IPY_MODEL_ab4928c0a86449a384e36d8c0bc25717"
]
}
},
"c43027a0735e459ca1f710e5a9c43177": {
"model_module": "@jupyter-widgets/base",
"model_name": "LayoutModel",
"model_module_version": "1.2.0",
"state": {
"_view_name": "LayoutView",
"grid_template_rows": null,
"right": null,
"justify_content": null,
"_view_module": "@jupyter-widgets/base",
"overflow": null,
"_model_module_version": "1.2.0",
"_view_count": null,
"flex_flow": null,
"width": null,
"min_width": null,
"border": null,
"align_items": null,
"bottom": null,
"_model_module": "@jupyter-widgets/base",
"top": null,
"grid_column": null,
"overflow_y": null,
"overflow_x": null,
"grid_auto_flow": null,
"grid_area": null,
"grid_template_columns": null,
"flex": null,
"_model_name": "LayoutModel",
"justify_items": null,
"grid_row": null,
"max_height": null,
"align_content": null,
"visibility": null,
"align_self": null,
"height": null,
"min_height": null,
"padding": null,
"grid_auto_rows": null,
"grid_gap": null,
"max_width": null,
"order": null,
"_view_module_version": "1.2.0",
"grid_template_areas": null,
"object_position": null,
"object_fit": null,
"grid_auto_columns": null,
"margin": null,
"display": null,
"left": null
}
},
"c00c959249db4f2a9b97adabd2684c3c": {
"model_module": "@jupyter-widgets/controls",
"model_name": "HTMLModel",
"model_module_version": "1.5.0",
"state": {
"_view_name": "HTMLView",
"style": "IPY_MODEL_3251223dac8f43c081701ff7f663cb35",
"_dom_classes": [],
"description": "",
"_model_name": "HTMLModel",
"placeholder": "",
"_view_module": "@jupyter-widgets/controls",
"_model_module_version": "1.5.0",
"value": "Downloading: 100%",
"_view_count": null,
"_view_module_version": "1.5.0",
"description_tooltip": null,
"_model_module": "@jupyter-widgets/controls",
"layout": "IPY_MODEL_d63d5559ce534b86969132d3ff8d875b"
}
},
"d9e4edd05c1e40f991eb2c2f1fc9ebc1": {
"model_module": "@jupyter-widgets/controls",
"model_name": "FloatProgressModel",
"model_module_version": "1.5.0",
"state": {
"_view_name": "ProgressView",
"style": "IPY_MODEL_59618f021fc4495e9c401a421d28d4a0",
"_dom_classes": [],
"description": "",
"_model_name": "FloatProgressModel",
"bar_style": "success",
"max": 381781,
"_view_module": "@jupyter-widgets/controls",
"_model_module_version": "1.5.0",
"value": 381781,
"_view_count": null,
"_view_module_version": "1.5.0",
"orientation": "horizontal",
"min": 0,
"description_tooltip": null,
"_model_module": "@jupyter-widgets/controls",
"layout": "IPY_MODEL_c56ba935682647dca4bdcc593fe0d2cc"
}
},
"ab4928c0a86449a384e36d8c0bc25717": {
"model_module": "@jupyter-widgets/controls",
"model_name": "HTMLModel",
"model_module_version": "1.5.0",
"state": {
"_view_name": "HTMLView",
"style": "IPY_MODEL_01808c7fec8447368d60a33b2d683851",
"_dom_classes": [],
"description": "",
"_model_name": "HTMLModel",
"placeholder": "",
"_view_module": "@jupyter-widgets/controls",
"_model_module_version": "1.5.0",
"value": " 373k/373k [00:00<00:00, 876kB/s]",
"_view_count": null,
"_view_module_version": "1.5.0",
"description_tooltip": null,
"_model_module": "@jupyter-widgets/controls",
"layout": "IPY_MODEL_f9466e0349c84633a0fb8ceeffa2a984"
}
},
"3251223dac8f43c081701ff7f663cb35": {
"model_module": "@jupyter-widgets/controls",
"model_name": "DescriptionStyleModel",
"model_module_version": "1.5.0",
"state": {
"_view_name": "StyleView",
"_model_name": "DescriptionStyleModel",
"description_width": "",
"_view_module": "@jupyter-widgets/base",
"_model_module_version": "1.5.0",
"_view_count": null,
"_view_module_version": "1.2.0",
"_model_module": "@jupyter-widgets/controls"
}
},
"d63d5559ce534b86969132d3ff8d875b": {
"model_module": "@jupyter-widgets/base",
"model_name": "LayoutModel",
"model_module_version": "1.2.0",
"state": {
"_view_name": "LayoutView",
"grid_template_rows": null,
"right": null,
"justify_content": null,
"_view_module": "@jupyter-widgets/base",
"overflow": null,
"_model_module_version": "1.2.0",
"_view_count": null,
"flex_flow": null,
"width": null,
"min_width": null,
"border": null,
"align_items": null,
"bottom": null,
"_model_module": "@jupyter-widgets/base",
"top": null,
"grid_column": null,
"overflow_y": null,
"overflow_x": null,
"grid_auto_flow": null,
"grid_area": null,
"grid_template_columns": null,
"flex": null,
"_model_name": "LayoutModel",
"justify_items": null,
"grid_row": null,
"max_height": null,
"align_content": null,
"visibility": null,
"align_self": null,
"height": null,
"min_height": null,
"padding": null,
"grid_auto_rows": null,
"grid_gap": null,
"max_width": null,
"order": null,
"_view_module_version": "1.2.0",
"grid_template_areas": null,
"object_position": null,
"object_fit": null,
"grid_auto_columns": null,
"margin": null,
"display": null,
"left": null
}
},
"59618f021fc4495e9c401a421d28d4a0": {
"model_module": "@jupyter-widgets/controls",
"model_name": "ProgressStyleModel",
"model_module_version": "1.5.0",
"state": {
"_view_name": "StyleView",
"_model_name": "ProgressStyleModel",
"description_width": "",
"_view_module": "@jupyter-widgets/base",
"_model_module_version": "1.5.0",
"_view_count": null,
"_view_module_version": "1.2.0",
"bar_color": null,
"_model_module": "@jupyter-widgets/controls"
}
},
"c56ba935682647dca4bdcc593fe0d2cc": {
"model_module": "@jupyter-widgets/base",
"model_name": "LayoutModel",
"model_module_version": "1.2.0",
"state": {
"_view_name": "LayoutView",
"grid_template_rows": null,
"right": null,
"justify_content": null,
"_view_module": "@jupyter-widgets/base",
"overflow": null,
"_model_module_version": "1.2.0",
"_view_count": null,
"flex_flow": null,
"width": null,
"min_width": null,
"border": null,
"align_items": null,
"bottom": null,
"_model_module": "@jupyter-widgets/base",
"top": null,
"grid_column": null,
"overflow_y": null,
"overflow_x": null,
"grid_auto_flow": null,
"grid_area": null,
"grid_template_columns": null,
"flex": null,
"_model_name": "LayoutModel",
"justify_items": null,
"grid_row": null,
"max_height": null,
"align_content": null,
"visibility": null,
"align_self": null,
"height": null,
"min_height": null,
"padding": null,
"grid_auto_rows": null,
"grid_gap": null,
"max_width": null,
"order": null,
"_view_module_version": "1.2.0",
"grid_template_areas": null,
"object_position": null,
"object_fit": null,
"grid_auto_columns": null,
"margin": null,
"display": null,
"left": null
}
},
"01808c7fec8447368d60a33b2d683851": {
"model_module": "@jupyter-widgets/controls",
"model_name": "DescriptionStyleModel",
"model_module_version": "1.5.0",
"state": {
"_view_name": "StyleView",
"_model_name": "DescriptionStyleModel",
"description_width": "",
"_view_module": "@jupyter-widgets/base",
"_model_module_version": "1.5.0",
"_view_count": null,
"_view_module_version": "1.2.0",
"_model_module": "@jupyter-widgets/controls"
}
},
"f9466e0349c84633a0fb8ceeffa2a984": {
"model_module": "@jupyter-widgets/base",
"model_name": "LayoutModel",
"model_module_version": "1.2.0",
"state": {
"_view_name": "LayoutView",
"grid_template_rows": null,
"right": null,
"justify_content": null,
"_view_module": "@jupyter-widgets/base",
"overflow": null,
"_model_module_version": "1.2.0",
"_view_count": null,
"flex_flow": null,
"width": null,
"min_width": null,
"border": null,
"align_items": null,
"bottom": null,
"_model_module": "@jupyter-widgets/base",
"top": null,
"grid_column": null,
"overflow_y": null,
"overflow_x": null,
"grid_auto_flow": null,
"grid_area": null,
"grid_template_columns": null,
"flex": null,
"_model_name": "LayoutModel",
"justify_items": null,
"grid_row": null,
"max_height": null,
"align_content": null,
"visibility": null,
"align_self": null,
"height": null,
"min_height": null,
"padding": null,
"grid_auto_rows": null,
"grid_gap": null,
"max_width": null,
"order": null,
"_view_module_version": "1.2.0",
"grid_template_areas": null,
"object_position": null,
"object_fit": null,
"grid_auto_columns": null,
"margin": null,
"display": null,
"left": null
}
},
"10ee9777b41e42129e2c9cc9327ad88f": {
"model_module": "@jupyter-widgets/controls",
"model_name": "HBoxModel",
"model_module_version": "1.5.0",
"state": {
"_view_name": "HBoxView",
"_dom_classes": [],
"_model_name": "HBoxModel",
"_view_module": "@jupyter-widgets/controls",
"_model_module_version": "1.5.0",
"_view_count": null,
"_view_module_version": "1.5.0",
"box_style": "",
"layout": "IPY_MODEL_cb6a647757244da3941602127ec38ccb",
"_model_module": "@jupyter-widgets/controls",
"children": [
"IPY_MODEL_a64a223312144f2f9736729b63ab1ce5",
"IPY_MODEL_7e7bce13eeed41179e4e15fc7afc89d5",
"IPY_MODEL_7bacd13c23cf415fa5d58e9243c4a785"
]
}
},
"cb6a647757244da3941602127ec38ccb": {
"model_module": "@jupyter-widgets/base",
"model_name": "LayoutModel",
"model_module_version": "1.2.0",
"state": {
"_view_name": "LayoutView",
"grid_template_rows": null,
"right": null,
"justify_content": null,
"_view_module": "@jupyter-widgets/base",
"overflow": null,
"_model_module_version": "1.2.0",
"_view_count": null,
"flex_flow": null,
"width": null,
"min_width": null,
"border": null,
"align_items": null,
"bottom": null,
"_model_module": "@jupyter-widgets/base",
"top": null,
"grid_column": null,
"overflow_y": null,
"overflow_x": null,
"grid_auto_flow": null,
"grid_area": null,
"grid_template_columns": null,
"flex": null,
"_model_name": "LayoutModel",
"justify_items": null,
"grid_row": null,
"max_height": null,
"align_content": null,
"visibility": null,
"align_self": null,
"height": null,
"min_height": null,
"padding": null,
"grid_auto_rows": null,
"grid_gap": null,
"max_width": null,
"order": null,
"_view_module_version": "1.2.0",
"grid_template_areas": null,
"object_position": null,
"object_fit": null,
"grid_auto_columns": null,
"margin": null,
"display": null,
"left": null
}
},
"a64a223312144f2f9736729b63ab1ce5": {
"model_module": "@jupyter-widgets/controls",
"model_name": "HTMLModel",
"model_module_version": "1.5.0",
"state": {
"_view_name": "HTMLView",
"style": "IPY_MODEL_9fe6e1167e5d45fbad2adab3d59e017d",
"_dom_classes": [],
"description": "",
"_model_name": "HTMLModel",
"placeholder": "",
"_view_module": "@jupyter-widgets/controls",
"_model_module_version": "1.5.0",
"value": "Downloading: 100%",
"_view_count": null,
"_view_module_version": "1.5.0",
"description_tooltip": null,
"_model_module": "@jupyter-widgets/controls",
"layout": "IPY_MODEL_655c507d8fcf423f8bd6746201f569ae"
}
},
"7e7bce13eeed41179e4e15fc7afc89d5": {
"model_module": "@jupyter-widgets/controls",
"model_name": "FloatProgressModel",
"model_module_version": "1.5.0",
"state": {
"_view_name": "ProgressView",
"style": "IPY_MODEL_8c4812afaaec4d65bf84a1e77840d356",
"_dom_classes": [],
"description": "",
"_model_name": "FloatProgressModel",
"bar_style": "success",
"max": 112,
"_view_module": "@jupyter-widgets/controls",
"_model_module_version": "1.5.0",
"value": 112,
"_view_count": null,
"_view_module_version": "1.5.0",
"orientation": "horizontal",
"min": 0,
"description_tooltip": null,
"_model_module": "@jupyter-widgets/controls",
"layout": "IPY_MODEL_64dfa71e3dff4236908e0592e4f90250"
}
},
"7bacd13c23cf415fa5d58e9243c4a785": {
"model_module": "@jupyter-widgets/controls",
"model_name": "HTMLModel",
"model_module_version": "1.5.0",
"state": {
"_view_name": "HTMLView",
"style": "IPY_MODEL_ec8d98c1edb148d3ae1c518b61e8155b",
"_dom_classes": [],
"description": "",
"_model_name": "HTMLModel",
"placeholder": "",
"_view_module": "@jupyter-widgets/controls",
"_model_module_version": "1.5.0",
"value": " 112/112 [00:00<00:00, 3.41kB/s]",
"_view_count": null,
"_view_module_version": "1.5.0",
"description_tooltip": null,
"_model_module": "@jupyter-widgets/controls",
"layout": "IPY_MODEL_a8212828565d4c9884d44fb45dc51ee5"
}
},
"9fe6e1167e5d45fbad2adab3d59e017d": {
"model_module": "@jupyter-widgets/controls",
"model_name": "DescriptionStyleModel",
"model_module_version": "1.5.0",
"state": {
"_view_name": "StyleView",
"_model_name": "DescriptionStyleModel",
"description_width": "",
"_view_module": "@jupyter-widgets/base",
"_model_module_version": "1.5.0",
"_view_count": null,
"_view_module_version": "1.2.0",
"_model_module": "@jupyter-widgets/controls"
}
},
"655c507d8fcf423f8bd6746201f569ae": {
"model_module": "@jupyter-widgets/base",
"model_name": "LayoutModel",
"model_module_version": "1.2.0",
"state": {
"_view_name": "LayoutView",
"grid_template_rows": null,
"right": null,
"justify_content": null,
"_view_module": "@jupyter-widgets/base",
"overflow": null,
"_model_module_version": "1.2.0",
"_view_count": null,
"flex_flow": null,
"width": null,
"min_width": null,
"border": null,
"align_items": null,
"bottom": null,
"_model_module": "@jupyter-widgets/base",
"top": null,
"grid_column": null,
"overflow_y": null,
"overflow_x": null,
"grid_auto_flow": null,
"grid_area": null,
"grid_template_columns": null,
"flex": null,
"_model_name": "LayoutModel",
"justify_items": null,
"grid_row": null,
"max_height": null,
"align_content": null,
"visibility": null,
"align_self": null,
"height": null,
"min_height": null,
"padding": null,
"grid_auto_rows": null,
"grid_gap": null,
"max_width": null,
"order": null,
"_view_module_version": "1.2.0",
"grid_template_areas": null,
"object_position": null,
"object_fit": null,
"grid_auto_columns": null,
"margin": null,
"display": null,
"left": null
}
},
"8c4812afaaec4d65bf84a1e77840d356": {
"model_module": "@jupyter-widgets/controls",
"model_name": "ProgressStyleModel",
"model_module_version": "1.5.0",
"state": {
"_view_name": "StyleView",
"_model_name": "ProgressStyleModel",
"description_width": "",
"_view_module": "@jupyter-widgets/base",
"_model_module_version": "1.5.0",
"_view_count": null,
"_view_module_version": "1.2.0",
"bar_color": null,
"_model_module": "@jupyter-widgets/controls"
}
},
"64dfa71e3dff4236908e0592e4f90250": {
"model_module": "@jupyter-widgets/base",
"model_name": "LayoutModel",
"model_module_version": "1.2.0",
"state": {
"_view_name": "LayoutView",
"grid_template_rows": null,
"right": null,
"justify_content": null,
"_view_module": "@jupyter-widgets/base",
"overflow": null,
"_model_module_version": "1.2.0",
"_view_count": null,
"flex_flow": null,
"width": null,
"min_width": null,
"border": null,
"align_items": null,
"bottom": null,
"_model_module": "@jupyter-widgets/base",
"top": null,
"grid_column": null,
"overflow_y": null,
"overflow_x": null,
"grid_auto_flow": null,
"grid_area": null,
"grid_template_columns": null,
"flex": null,
"_model_name": "LayoutModel",
"justify_items": null,
"grid_row": null,
"max_height": null,
"align_content": null,
"visibility": null,
"align_self": null,
"height": null,
"min_height": null,
"padding": null,
"grid_auto_rows": null,
"grid_gap": null,
"max_width": null,
"order": null,
"_view_module_version": "1.2.0",
"grid_template_areas": null,
"object_position": null,
"object_fit": null,
"grid_auto_columns": null,
"margin": null,
"display": null,
"left": null
}
},
"ec8d98c1edb148d3ae1c518b61e8155b": {
"model_module": "@jupyter-widgets/controls",
"model_name": "DescriptionStyleModel",
"model_module_version": "1.5.0",
"state": {
"_view_name": "StyleView",
"_model_name": "DescriptionStyleModel",
"description_width": "",
"_view_module": "@jupyter-widgets/base",
"_model_module_version": "1.5.0",
"_view_count": null,
"_view_module_version": "1.2.0",
"_model_module": "@jupyter-widgets/controls"
}
},
"a8212828565d4c9884d44fb45dc51ee5": {
"model_module": "@jupyter-widgets/base",
"model_name": "LayoutModel",
"model_module_version": "1.2.0",
"state": {
"_view_name": "LayoutView",
"grid_template_rows": null,
"right": null,
"justify_content": null,
"_view_module": "@jupyter-widgets/base",
"overflow": null,
"_model_module_version": "1.2.0",
"_view_count": null,
"flex_flow": null,
"width": null,
"min_width": null,
"border": null,
"align_items": null,
"bottom": null,
"_model_module": "@jupyter-widgets/base",
"top": null,
"grid_column": null,
"overflow_y": null,
"overflow_x": null,
"grid_auto_flow": null,
"grid_area": null,
"grid_template_columns": null,
"flex": null,
"_model_name": "LayoutModel",
"justify_items": null,
"grid_row": null,
"max_height": null,
"align_content": null,
"visibility": null,
"align_self": null,
"height": null,
"min_height": null,
"padding": null,
"grid_auto_rows": null,
"grid_gap": null,
"max_width": null,
"order": null,
"_view_module_version": "1.2.0",
"grid_template_areas": null,
"object_position": null,
"object_fit": null,
"grid_auto_columns": null,
"margin": null,
"display": null,
"left": null
}
},
"9a2d4d7da3024cc0828b1a6dafd0dd16": {
"model_module": "@jupyter-widgets/controls",
"model_name": "HBoxModel",
"model_module_version": "1.5.0",
"state": {
"_view_name": "HBoxView",
"_dom_classes": [],
"_model_name": "HBoxModel",
"_view_module": "@jupyter-widgets/controls",
"_model_module_version": "1.5.0",
"_view_count": null,
"_view_module_version": "1.5.0",
"box_style": "",
"layout": "IPY_MODEL_fe4daa4d7d024187aa2f622dbf3577a8",
"_model_module": "@jupyter-widgets/controls",
"children": [
"IPY_MODEL_8380bf9a899645e8aef576e640b41ea2",
"IPY_MODEL_37c593d2f442497483cd0026498bab05",
"IPY_MODEL_3cc7b132c94f427ba44858e4c4ce3019"
]
}
},
"fe4daa4d7d024187aa2f622dbf3577a8": {
"model_module": "@jupyter-widgets/base",
"model_name": "LayoutModel",
"model_module_version": "1.2.0",
"state": {
"_view_name": "LayoutView",
"grid_template_rows": null,
"right": null,
"justify_content": null,
"_view_module": "@jupyter-widgets/base",
"overflow": null,
"_model_module_version": "1.2.0",
"_view_count": null,
"flex_flow": null,
"width": null,
"min_width": null,
"border": null,
"align_items": null,
"bottom": null,
"_model_module": "@jupyter-widgets/base",
"top": null,
"grid_column": null,
"overflow_y": null,
"overflow_x": null,
"grid_auto_flow": null,
"grid_area": null,
"grid_template_columns": null,
"flex": null,
"_model_name": "LayoutModel",
"justify_items": null,
"grid_row": null,
"max_height": null,
"align_content": null,
"visibility": null,
"align_self": null,
"height": null,
"min_height": null,
"padding": null,
"grid_auto_rows": null,
"grid_gap": null,
"max_width": null,
"order": null,
"_view_module_version": "1.2.0",
"grid_template_areas": null,
"object_position": null,
"object_fit": null,
"grid_auto_columns": null,
"margin": null,
"display": null,
"left": null
}
},
"8380bf9a899645e8aef576e640b41ea2": {
"model_module": "@jupyter-widgets/controls",
"model_name": "HTMLModel",
"model_module_version": "1.5.0",
"state": {
"_view_name": "HTMLView",
"style": "IPY_MODEL_d2aed3c0f95b4677bd6e949a4ed0403e",
"_dom_classes": [],
"description": "",
"_model_name": "HTMLModel",
"placeholder": "",
"_view_module": "@jupyter-widgets/controls",
"_model_module_version": "1.5.0",
"value": "Downloading: 100%",
"_view_count": null,
"_view_module_version": "1.5.0",
"description_tooltip": null,
"_model_module": "@jupyter-widgets/controls",
"layout": "IPY_MODEL_bbc52e0e0b2f4758bd7d6cf44b4670ae"
}
},
"37c593d2f442497483cd0026498bab05": {
"model_module": "@jupyter-widgets/controls",
"model_name": "FloatProgressModel",
"model_module_version": "1.5.0",
"state": {
"_view_name": "ProgressView",
"style": "IPY_MODEL_8513d262d8764d99aa5d3f2f178b875e",
"_dom_classes": [],
"description": "",
"_model_name": "FloatProgressModel",
"bar_style": "success",
"max": 239,
"_view_module": "@jupyter-widgets/controls",
"_model_module_version": "1.5.0",
"value": 239,
"_view_count": null,
"_view_module_version": "1.5.0",
"orientation": "horizontal",
"min": 0,
"description_tooltip": null,
"_model_module": "@jupyter-widgets/controls",
"layout": "IPY_MODEL_57984bbd46a84a7bb2b7629e6b2f9ef9"
}
},
"3cc7b132c94f427ba44858e4c4ce3019": {
"model_module": "@jupyter-widgets/controls",
"model_name": "HTMLModel",
"model_module_version": "1.5.0",
"state": {
"_view_name": "HTMLView",
"style": "IPY_MODEL_b886e2b6bbcd46cf806ff3a0b3cb8d33",
"_dom_classes": [],
"description": "",
"_model_name": "HTMLModel",
"placeholder": "",
"_view_module": "@jupyter-widgets/controls",
"_model_module_version": "1.5.0",
"value": " 239/239 [00:00<00:00, 5.49kB/s]",
"_view_count": null,
"_view_module_version": "1.5.0",
"description_tooltip": null,
"_model_module": "@jupyter-widgets/controls",
"layout": "IPY_MODEL_21ac91113f7e4548b416a32b1b3f66a9"
}
},
"d2aed3c0f95b4677bd6e949a4ed0403e": {
"model_module": "@jupyter-widgets/controls",
"model_name": "DescriptionStyleModel",
"model_module_version": "1.5.0",
"state": {
"_view_name": "StyleView",
"_model_name": "DescriptionStyleModel",
"description_width": "",
"_view_module": "@jupyter-widgets/base",
"_model_module_version": "1.5.0",
"_view_count": null,
"_view_module_version": "1.2.0",
"_model_module": "@jupyter-widgets/controls"
}
},
"bbc52e0e0b2f4758bd7d6cf44b4670ae": {
"model_module": "@jupyter-widgets/base",
"model_name": "LayoutModel",
"model_module_version": "1.2.0",
"state": {
"_view_name": "LayoutView",
"grid_template_rows": null,
"right": null,
"justify_content": null,
"_view_module": "@jupyter-widgets/base",
"overflow": null,
"_model_module_version": "1.2.0",
"_view_count": null,
"flex_flow": null,
"width": null,
"min_width": null,
"border": null,
"align_items": null,
"bottom": null,
"_model_module": "@jupyter-widgets/base",
"top": null,
"grid_column": null,
"overflow_y": null,
"overflow_x": null,
"grid_auto_flow": null,
"grid_area": null,
"grid_template_columns": null,
"flex": null,
"_model_name": "LayoutModel",
"justify_items": null,
"grid_row": null,
"max_height": null,
"align_content": null,
"visibility": null,
"align_self": null,
"height": null,
"min_height": null,
"padding": null,
"grid_auto_rows": null,
"grid_gap": null,
"max_width": null,
"order": null,
"_view_module_version": "1.2.0",
"grid_template_areas": null,
"object_position": null,
"object_fit": null,
"grid_auto_columns": null,
"margin": null,
"display": null,
"left": null
}
},
"8513d262d8764d99aa5d3f2f178b875e": {
"model_module": "@jupyter-widgets/controls",
"model_name": "ProgressStyleModel",
"model_module_version": "1.5.0",
"state": {
"_view_name": "StyleView",
"_model_name": "ProgressStyleModel",
"description_width": "",
"_view_module": "@jupyter-widgets/base",
"_model_module_version": "1.5.0",
"_view_count": null,
"_view_module_version": "1.2.0",
"bar_color": null,
"_model_module": "@jupyter-widgets/controls"
}
},
"57984bbd46a84a7bb2b7629e6b2f9ef9": {
"model_module": "@jupyter-widgets/base",
"model_name": "LayoutModel",
"model_module_version": "1.2.0",
"state": {
"_view_name": "LayoutView",
"grid_template_rows": null,
"right": null,
"justify_content": null,
"_view_module": "@jupyter-widgets/base",
"overflow": null,
"_model_module_version": "1.2.0",
"_view_count": null,
"flex_flow": null,
"width": null,
"min_width": null,
"border": null,
"align_items": null,
"bottom": null,
"_model_module": "@jupyter-widgets/base",
"top": null,
"grid_column": null,
"overflow_y": null,
"overflow_x": null,
"grid_auto_flow": null,
"grid_area": null,
"grid_template_columns": null,
"flex": null,
"_model_name": "LayoutModel",
"justify_items": null,
"grid_row": null,
"max_height": null,
"align_content": null,
"visibility": null,
"align_self": null,
"height": null,
"min_height": null,
"padding": null,
"grid_auto_rows": null,
"grid_gap": null,
"max_width": null,
"order": null,
"_view_module_version": "1.2.0",
"grid_template_areas": null,
"object_position": null,
"object_fit": null,
"grid_auto_columns": null,
"margin": null,
"display": null,
"left": null
}
},
"b886e2b6bbcd46cf806ff3a0b3cb8d33": {
"model_module": "@jupyter-widgets/controls",
"model_name": "DescriptionStyleModel",
"model_module_version": "1.5.0",
"state": {
"_view_name": "StyleView",
"_model_name": "DescriptionStyleModel",
"description_width": "",
"_view_module": "@jupyter-widgets/base",
"_model_module_version": "1.5.0",
"_view_count": null,
"_view_module_version": "1.2.0",
"_model_module": "@jupyter-widgets/controls"
}
},
"21ac91113f7e4548b416a32b1b3f66a9": {
"model_module": "@jupyter-widgets/base",
"model_name": "LayoutModel",
"model_module_version": "1.2.0",
"state": {
"_view_name": "LayoutView",
"grid_template_rows": null,
"right": null,
"justify_content": null,
"_view_module": "@jupyter-widgets/base",
"overflow": null,
"_model_module_version": "1.2.0",
"_view_count": null,
"flex_flow": null,
"width": null,
"min_width": null,
"border": null,
"align_items": null,
"bottom": null,
"_model_module": "@jupyter-widgets/base",
"top": null,
"grid_column": null,
"overflow_y": null,
"overflow_x": null,
"grid_auto_flow": null,
"grid_area": null,
"grid_template_columns": null,
"flex": null,
"_model_name": "LayoutModel",
"justify_items": null,
"grid_row": null,
"max_height": null,
"align_content": null,
"visibility": null,
"align_self": null,
"height": null,
"min_height": null,
"padding": null,
"grid_auto_rows": null,
"grid_gap": null,
"max_width": null,
"order": null,
"_view_module_version": "1.2.0",
"grid_template_areas": null,
"object_position": null,
"object_fit": null,
"grid_auto_columns": null,
"margin": null,
"display": null,
"left": null
}
},
"f8958c6de2394fecab9f95388a365431": {
"model_module": "@jupyter-widgets/controls",
"model_name": "HBoxModel",
"model_module_version": "1.5.0",
"state": {
"_view_name": "HBoxView",
"_dom_classes": [],
"_model_name": "HBoxModel",
"_view_module": "@jupyter-widgets/controls",
"_model_module_version": "1.5.0",
"_view_count": null,
"_view_module_version": "1.5.0",
"box_style": "",
"layout": "IPY_MODEL_11a8a4b2d39d4ea8904c0f1b2f6dd906",
"_model_module": "@jupyter-widgets/controls",
"children": [
"IPY_MODEL_9657501af7514a60b30fcd60a223980c",
"IPY_MODEL_61274b2bac5e4835a8bd33dc201bc155",
"IPY_MODEL_973300b095554b10ac290244772e0a6f"
]
}
},
"11a8a4b2d39d4ea8904c0f1b2f6dd906": {
"model_module": "@jupyter-widgets/base",
"model_name": "LayoutModel",
"model_module_version": "1.2.0",
"state": {
"_view_name": "LayoutView",
"grid_template_rows": null,
"right": null,
"justify_content": null,
"_view_module": "@jupyter-widgets/base",
"overflow": null,
"_model_module_version": "1.2.0",
"_view_count": null,
"flex_flow": null,
"width": null,
"min_width": null,
"border": null,
"align_items": null,
"bottom": null,
"_model_module": "@jupyter-widgets/base",
"top": null,
"grid_column": null,
"overflow_y": null,
"overflow_x": null,
"grid_auto_flow": null,
"grid_area": null,
"grid_template_columns": null,
"flex": null,
"_model_name": "LayoutModel",
"justify_items": null,
"grid_row": null,
"max_height": null,
"align_content": null,
"visibility": null,
"align_self": null,
"height": null,
"min_height": null,
"padding": null,
"grid_auto_rows": null,
"grid_gap": null,
"max_width": null,
"order": null,
"_view_module_version": "1.2.0",
"grid_template_areas": null,
"object_position": null,
"object_fit": null,
"grid_auto_columns": null,
"margin": null,
"display": null,
"left": null
}
},
"9657501af7514a60b30fcd60a223980c": {
"model_module": "@jupyter-widgets/controls",
"model_name": "HTMLModel",
"model_module_version": "1.5.0",
"state": {
"_view_name": "HTMLView",
"style": "IPY_MODEL_c3f5f56bb14d44b6a5775a77f6763b94",
"_dom_classes": [],
"description": "",
"_model_name": "HTMLModel",
"placeholder": "",
"_view_module": "@jupyter-widgets/controls",
"_model_module_version": "1.5.0",
"value": "Downloading: 100%",
"_view_count": null,
"_view_module_version": "1.5.0",
"description_tooltip": null,
"_model_module": "@jupyter-widgets/controls",
"layout": "IPY_MODEL_fbf430940c8a49949953155b57d07766"
}
},
"61274b2bac5e4835a8bd33dc201bc155": {
"model_module": "@jupyter-widgets/controls",
"model_name": "FloatProgressModel",
"model_module_version": "1.5.0",
"state": {
"_view_name": "ProgressView",
"style": "IPY_MODEL_ab05641bcb9c49aab977110fab503a78",
"_dom_classes": [],
"description": "",
"_model_name": "FloatProgressModel",
"bar_style": "success",
"max": 175,
"_view_module": "@jupyter-widgets/controls",
"_model_module_version": "1.5.0",
"value": 175,
"_view_count": null,
"_view_module_version": "1.5.0",
"orientation": "horizontal",
"min": 0,
"description_tooltip": null,
"_model_module": "@jupyter-widgets/controls",
"layout": "IPY_MODEL_4f11a71d7df943e48ac9ea3bab5c6771"
}
},
"973300b095554b10ac290244772e0a6f": {
"model_module": "@jupyter-widgets/controls",
"model_name": "HTMLModel",
"model_module_version": "1.5.0",
"state": {
"_view_name": "HTMLView",
"style": "IPY_MODEL_b6004e09152045e18503cf75e32d4fa6",
"_dom_classes": [],
"description": "",
"_model_name": "HTMLModel",
"placeholder": "",
"_view_module": "@jupyter-widgets/controls",
"_model_module_version": "1.5.0",
"value": " 175/175 [00:00<00:00, 5.41kB/s]",
"_view_count": null,
"_view_module_version": "1.5.0",
"description_tooltip": null,
"_model_module": "@jupyter-widgets/controls",
"layout": "IPY_MODEL_590fb707d26948b5b9c8bb3b896f29e1"
}
},
"c3f5f56bb14d44b6a5775a77f6763b94": {
"model_module": "@jupyter-widgets/controls",
"model_name": "DescriptionStyleModel",
"model_module_version": "1.5.0",
"state": {
"_view_name": "StyleView",
"_model_name": "DescriptionStyleModel",
"description_width": "",
"_view_module": "@jupyter-widgets/base",
"_model_module_version": "1.5.0",
"_view_count": null,
"_view_module_version": "1.2.0",
"_model_module": "@jupyter-widgets/controls"
}
},
"fbf430940c8a49949953155b57d07766": {
"model_module": "@jupyter-widgets/base",
"model_name": "LayoutModel",
"model_module_version": "1.2.0",
"state": {
"_view_name": "LayoutView",
"grid_template_rows": null,
"right": null,
"justify_content": null,
"_view_module": "@jupyter-widgets/base",
"overflow": null,
"_model_module_version": "1.2.0",
"_view_count": null,
"flex_flow": null,
"width": null,
"min_width": null,
"border": null,
"align_items": null,
"bottom": null,
"_model_module": "@jupyter-widgets/base",
"top": null,
"grid_column": null,
"overflow_y": null,
"overflow_x": null,
"grid_auto_flow": null,
"grid_area": null,
"grid_template_columns": null,
"flex": null,
"_model_name": "LayoutModel",
"justify_items": null,
"grid_row": null,
"max_height": null,
"align_content": null,
"visibility": null,
"align_self": null,
"height": null,
"min_height": null,
"padding": null,
"grid_auto_rows": null,
"grid_gap": null,
"max_width": null,
"order": null,
"_view_module_version": "1.2.0",
"grid_template_areas": null,
"object_position": null,
"object_fit": null,
"grid_auto_columns": null,
"margin": null,
"display": null,
"left": null
}
},
"ab05641bcb9c49aab977110fab503a78": {
"model_module": "@jupyter-widgets/controls",
"model_name": "ProgressStyleModel",
"model_module_version": "1.5.0",
"state": {
"_view_name": "StyleView",
"_model_name": "ProgressStyleModel",
"description_width": "",
"_view_module": "@jupyter-widgets/base",
"_model_module_version": "1.5.0",
"_view_count": null,
"_view_module_version": "1.2.0",
"bar_color": null,
"_model_module": "@jupyter-widgets/controls"
}
},
"4f11a71d7df943e48ac9ea3bab5c6771": {
"model_module": "@jupyter-widgets/base",
"model_name": "LayoutModel",
"model_module_version": "1.2.0",
"state": {
"_view_name": "LayoutView",
"grid_template_rows": null,
"right": null,
"justify_content": null,
"_view_module": "@jupyter-widgets/base",
"overflow": null,
"_model_module_version": "1.2.0",
"_view_count": null,
"flex_flow": null,
"width": null,
"min_width": null,
"border": null,
"align_items": null,
"bottom": null,
"_model_module": "@jupyter-widgets/base",
"top": null,
"grid_column": null,
"overflow_y": null,
"overflow_x": null,
"grid_auto_flow": null,
"grid_area": null,
"grid_template_columns": null,
"flex": null,
"_model_name": "LayoutModel",
"justify_items": null,
"grid_row": null,
"max_height": null,
"align_content": null,
"visibility": null,
"align_self": null,
"height": null,
"min_height": null,
"padding": null,
"grid_auto_rows": null,
"grid_gap": null,
"max_width": null,
"order": null,
"_view_module_version": "1.2.0",
"grid_template_areas": null,
"object_position": null,
"object_fit": null,
"grid_auto_columns": null,
"margin": null,
"display": null,
"left": null
}
},
"b6004e09152045e18503cf75e32d4fa6": {
"model_module": "@jupyter-widgets/controls",
"model_name": "DescriptionStyleModel",
"model_module_version": "1.5.0",
"state": {
"_view_name": "StyleView",
"_model_name": "DescriptionStyleModel",
"description_width": "",
"_view_module": "@jupyter-widgets/base",
"_model_module_version": "1.5.0",
"_view_count": null,
"_view_module_version": "1.2.0",
"_model_module": "@jupyter-widgets/controls"
}
},
"590fb707d26948b5b9c8bb3b896f29e1": {
"model_module": "@jupyter-widgets/base",
"model_name": "LayoutModel",
"model_module_version": "1.2.0",
"state": {
"_view_name": "LayoutView",
"grid_template_rows": null,
"right": null,
"justify_content": null,
"_view_module": "@jupyter-widgets/base",
"overflow": null,
"_model_module_version": "1.2.0",
"_view_count": null,
"flex_flow": null,
"width": null,
"min_width": null,
"border": null,
"align_items": null,
"bottom": null,
"_model_module": "@jupyter-widgets/base",
"top": null,
"grid_column": null,
"overflow_y": null,
"overflow_x": null,
"grid_auto_flow": null,
"grid_area": null,
"grid_template_columns": null,
"flex": null,
"_model_name": "LayoutModel",
"justify_items": null,
"grid_row": null,
"max_height": null,
"align_content": null,
"visibility": null,
"align_self": null,
"height": null,
"min_height": null,
"padding": null,
"grid_auto_rows": null,
"grid_gap": null,
"max_width": null,
"order": null,
"_view_module_version": "1.2.0",
"grid_template_areas": null,
"object_position": null,
"object_fit": null,
"grid_auto_columns": null,
"margin": null,
"display": null,
"left": null
}
}
}
}
},
"cells": [
{
"cell_type": "markdown",
"source": [
"# [](https://colab.research.google.com/github/cene555/ru-clip-tiny/blob/main/notebooks/ru_CLIP_tiny_onnx.ipynb)"
],
"metadata": {
"id": "JsWuTduwaagq"
}
},
{
"cell_type": "markdown",
"source": [
"## Select a runtime GPU to continue:\n",
"\n",
"Click Runtime -> Change Runtime Type -> switch \"Harware accelerator\" to be GPU. Save it, and you maybe connect to GPU"
],
"metadata": {
"id": "VCCzmQdKJPkv"
}
},
{
"cell_type": "code",
"source": [
"#@title Allowed Resources\n",
"import multiprocessing\n",
"import torch\n",
"from psutil import virtual_memory\n",
"\n",
"ram_gb = round(virtual_memory().total / 1024**3, 1)\n",
"\n",
"print('CPU:', multiprocessing.cpu_count())\n",
"print('RAM GB:', ram_gb)\n",
"print(\"PyTorch version:\", torch.__version__)\n",
"print(\"CUDA version:\", torch.version.cuda)\n",
"print(\"cuDNN version:\", torch.backends.cudnn.version())\n",
"device = torch.device(\"cuda:0\" if torch.cuda.is_available() else \"cpu\")\n",
"print(\"device:\", device.type)\n",
"\n",
"!nvidia-smi"
],
"metadata": {
"cellView": "form",
"id": "6xdy_cPJEYXV",
"colab": {
"base_uri": "https://localhost:8080/"
},
"outputId": "9b5c5751-3377-4623-fd90-f59c21118c80"
},
"execution_count": null,
"outputs": [
{
"output_type": "stream",
"name": "stdout",
"text": [
"CPU: 2\n",
"RAM GB: 12.7\n",
"PyTorch version: 1.10.0+cu111\n",
"CUDA version: 11.1\n",
"cuDNN version: 8005\n",
"device: cuda\n",
"Tue Feb 1 17:26:24 2022 \n",
"+-----------------------------------------------------------------------------+\n",
"| NVIDIA-SMI 495.46 Driver Version: 460.32.03 CUDA Version: 11.2 |\n",
"|-------------------------------+----------------------+----------------------+\n",
"| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |\n",
"| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |\n",
"| | | MIG M. |\n",
"|===============================+======================+======================|\n",
"| 0 Tesla T4 Off | 00000000:00:04.0 Off | 0 |\n",
"| N/A 61C P8 11W / 70W | 3MiB / 15109MiB | 0% Default |\n",
"| | | N/A |\n",
"+-------------------------------+----------------------+----------------------+\n",
" \n",
"+-----------------------------------------------------------------------------+\n",
"| Processes: |\n",
"| GPU GI CI PID Type Process name GPU Memory |\n",
"| ID ID Usage |\n",
"|=============================================================================|\n",
"| No running processes found |\n",
"+-----------------------------------------------------------------------------+\n"
]
}
]
},
{
"cell_type": "markdown",
"source": [
"## Restart colab session after installation\n",
"Reload session if something doesn't work (may need multiple times)"
],
"metadata": {
"id": "hmNP7iJBj6XZ"
}
},
{
"cell_type": "markdown",
"source": [
"## Install requirements"
],
"metadata": {
"id": "WWXCt_2NLhN_"
}
},
{
"cell_type": "code",
"source": [
"%%capture\n",
"!gdown -O ru-clip-tiny.pkl https://drive.google.com/uc?id=1-3g3J90pZmHo9jbBzsEmr7ei5zm3VXOL\n",
"\n",
"!pip install git+https://github.com/cene555/ru-clip-tiny.git\n",
"!pip install git+https://github.com/Lednik7/CLIP-ONNX.git\n",
"!pip install onnxruntime-gpu\n",
"\n",
"!wget -c -O CLIP.png https://github.com/openai/CLIP/blob/main/CLIP.png?raw=true"
],
"metadata": {
"id": "FWEEtd7Vryaf"
},
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"source": [
"import onnxruntime\n",
"\n",
"# priority device (if available)\n",
"print(onnxruntime.get_device())"
],
"metadata": {
"id": "bUFx02Dhjap4",
"colab": {
"base_uri": "https://localhost:8080/"
},
"outputId": "f595c387-da47-47e5-f96a-2d84adf3286b"
},
"execution_count": null,
"outputs": [
{
"output_type": "stream",
"name": "stdout",
"text": [
"GPU\n"
]
}
]
},
{
"cell_type": "markdown",
"source": [
"## Import libraries"
],
"metadata": {
"id": "PHb4CAoRL3qC"
}
},
{
"cell_type": "code",
"source": [
"import torch\n",
"from rucliptiny import RuCLIPtiny\n",
"from rucliptiny.utils import get_transform\n",
"from rucliptiny.tokenizer import Tokenizer"
],
"metadata": {
"id": "cznZ7ozDL5-M"
},
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"source": [
"import warnings\n",
"\n",
"warnings.filterwarnings(\"ignore\", category=UserWarning)"
],
"metadata": {
"id": "57COx0BKCmFA"
},
"execution_count": null,
"outputs": []
},
{
"cell_type": "markdown",
"source": [
"## Load model"
],
"metadata": {
"id": "ithu4-z0PIm5"
}
},
{
"cell_type": "code",
"source": [
"#@title speed_test function\n",
"\n",
"import time\n",
"\n",
"def speed_test(func, data_gen, n=5, empty_cache=True, is_text=False,\n",
" first_run=True):\n",
" if empty_cache: torch.cuda.empty_cache()\n",
" if first_run:\n",
" if is_text:\n",
" input_data1, input_data2 = data_gen()\n",
" func(input_data1, input_data2)\n",
" else:\n",
" input_data = data_gen()\n",
" func(input_data)\n",
" torch.cuda.empty_cache()\n",
" \n",
" values = []\n",
" for _ in range(n):\n",
" if is_text:\n",
" input_data1, input_data2 = data_gen()\n",
" else:\n",
" input_data = data_gen()\n",
" if is_text:\n",
" t = time.time()\n",
" func(input_data1, input_data2)\n",
" else:\n",
" t = time.time()\n",
" func(input_data)\n",
" values.append(time.time() - t)\n",
" if empty_cache: torch.cuda.empty_cache()\n",
" return sum(values) / n"
],
"metadata": {
"id": "GqKM04tP4Vv3",
"cellView": "form"
},
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"source": [
"torch.manual_seed(1)\n",
"device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')"
],
"metadata": {
"id": "SSOHYDRQGif-"
},
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"source": [
"model = RuCLIPtiny()\n",
"model.load_state_dict(torch.load('ru-clip-tiny.pkl',\n",
" map_location=device))\n",
"model = model.to(device).eval()\n",
"for x in model.parameters(): x.requires_grad = False\n",
"torch.cuda.empty_cache()"
],
"metadata": {
"id": "OpFAZfq-_nJe"
},
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"source": [
"transforms = get_transform()\n",
"tokenizer = Tokenizer()"
],
"metadata": {
"id": "KEZj2WrwkzZz",
"colab": {
"base_uri": "https://localhost:8080/",
"height": 145,
"referenced_widgets": [
"5319c7971f234d4bb615508f76475f9e",
"c43027a0735e459ca1f710e5a9c43177",
"c00c959249db4f2a9b97adabd2684c3c",
"d9e4edd05c1e40f991eb2c2f1fc9ebc1",
"ab4928c0a86449a384e36d8c0bc25717",
"3251223dac8f43c081701ff7f663cb35",
"d63d5559ce534b86969132d3ff8d875b",
"59618f021fc4495e9c401a421d28d4a0",
"c56ba935682647dca4bdcc593fe0d2cc",
"01808c7fec8447368d60a33b2d683851",
"f9466e0349c84633a0fb8ceeffa2a984",
"10ee9777b41e42129e2c9cc9327ad88f",
"cb6a647757244da3941602127ec38ccb",
"a64a223312144f2f9736729b63ab1ce5",
"7e7bce13eeed41179e4e15fc7afc89d5",
"7bacd13c23cf415fa5d58e9243c4a785",
"9fe6e1167e5d45fbad2adab3d59e017d",
"655c507d8fcf423f8bd6746201f569ae",
"8c4812afaaec4d65bf84a1e77840d356",
"64dfa71e3dff4236908e0592e4f90250",
"ec8d98c1edb148d3ae1c518b61e8155b",
"a8212828565d4c9884d44fb45dc51ee5",
"9a2d4d7da3024cc0828b1a6dafd0dd16",
"fe4daa4d7d024187aa2f622dbf3577a8",
"8380bf9a899645e8aef576e640b41ea2",
"37c593d2f442497483cd0026498bab05",
"3cc7b132c94f427ba44858e4c4ce3019",
"d2aed3c0f95b4677bd6e949a4ed0403e",
"bbc52e0e0b2f4758bd7d6cf44b4670ae",
"8513d262d8764d99aa5d3f2f178b875e",
"57984bbd46a84a7bb2b7629e6b2f9ef9",
"b886e2b6bbcd46cf806ff3a0b3cb8d33",
"21ac91113f7e4548b416a32b1b3f66a9",
"f8958c6de2394fecab9f95388a365431",
"11a8a4b2d39d4ea8904c0f1b2f6dd906",
"9657501af7514a60b30fcd60a223980c",
"61274b2bac5e4835a8bd33dc201bc155",
"973300b095554b10ac290244772e0a6f",
"c3f5f56bb14d44b6a5775a77f6763b94",
"fbf430940c8a49949953155b57d07766",
"ab05641bcb9c49aab977110fab503a78",
"4f11a71d7df943e48ac9ea3bab5c6771",
"b6004e09152045e18503cf75e32d4fa6",
"590fb707d26948b5b9c8bb3b896f29e1"
]
},
"outputId": "466854ba-7fa2-4154-ada2-391626146c95"
},
"execution_count": null,
"outputs": [
{
"output_type": "display_data",
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "5319c7971f234d4bb615508f76475f9e",
"version_minor": 0,
"version_major": 2
},
"text/plain": [
"Downloading: 0%| | 0.00/373k [00:00\n",
" | \n", " | backend | \n", "batch | \n", "encode_image | \n", "encode_text | \n", "
|---|---|---|---|---|
| 0 | \n", "onnx | \n", "2 | \n", "0.011 | \n", "0.001 | \n", "
| 1 | \n", "torch | \n", "2 | \n", "0.018 | \n", "0.003 | \n", "
| 2 | \n", "onnx | \n", "8 | \n", "0.035 | \n", "0.002 | \n", "
| 3 | \n", "torch | \n", "8 | \n", "0.010 | \n", "0.003 | \n", "
| 4 | \n", "onnx | \n", "16 | \n", "0.070 | \n", "0.004 | \n", "
| 5 | \n", "torch | \n", "16 | \n", "0.010 | \n", "0.003 | \n", "
| 6 | \n", "onnx | \n", "32 | \n", "0.145 | \n", "0.007 | \n", "
| 7 | \n", "torch | \n", "32 | \n", "0.012 | \n", "0.004 | \n", "
| 8 | \n", "onnx | \n", "64 | \n", "0.294 | \n", "0.014 | \n", "
| 9 | \n", "torch | \n", "64 | \n", "0.013 | \n", "0.005 | \n", "