docs/LLaVA_OneVision_Tutorials.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# (Frustratingly Easy) LLaVA OneVision Tutorial\n",
    "\n",
    "We know that it's always beneficial to have a unified interface for different tasks. So we are trying to unify the interface for image, text, image-text interleaved, and video input. And in this tutorial, we aim to provide the most straightforward way to use our model. \n",
    "\n",
    "We use our 0.5B version as an example. This could be running on a GPU with 4GB memory. And with the following examples, you could see it's surprisingly have promising performance on understanding the image, interleaved image-text, and video. Tiny but mighty!\n",
    "\n",
    "The same code could be used for 7B model as well.\n",
    "\n",
    "## Inference Guidance\n",
    "\n",
    "First please install our repo with code and environments: pip install git+https://github.com/LLaVA-VL/LLaVA-NeXT.git\n",
    "\n",
    "Here is a quick inference code using [lmms-lab/qwen2-0.5b-si](https://huggingface.co/lmms-lab/llava-onevision-qwen2-0.5b-si) as an example. You will need to install `flash-attn` to use this code snippet. If you don't want to install it, you can set `attn_implementation=None` when load_pretrained_model\n",
    "\n",
    "### Image Input\n",
    "Tackling the single image input with LLaVA OneVision is pretty straightforward."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from llava.model.builder import load_pretrained_model\n",
    "from llava.mm_utils import get_model_name_from_path, process_images, tokenizer_image_token\n",
    "from llava.constants import IMAGE_TOKEN_INDEX, DEFAULT_IMAGE_TOKEN, DEFAULT_IM_START_TOKEN, DEFAULT_IM_END_TOKEN, IGNORE_INDEX\n",
    "from llava.conversation import conv_templates, SeparatorStyle\n",
    "\n",
    "from PIL import Image\n",
    "import requests\n",
    "import copy\n",
    "import torch\n",
    "\n",
    "import sys\n",
    "import warnings\n",
    "\n",
    "warnings.filterwarnings(\"ignore\")\n",
    "pretrained = \"lmms-lab/llava-onevision-qwen2-0.5b-si\"\n",
    "model_name = \"llava_qwen\"\n",
    "device = \"cuda\"\n",
    "device_map = \"auto\"\n",
    "llava_model_args = {\n",
    "    \"multimodal\": True,\n",
    "    \"attn_implementation\": \"sdpa\",\n",
    "}\n",
    "tokenizer, model, image_processor, max_length = load_pretrained_model(pretrained, None, model_name, device_map=device_map, **llava_model_args)  # Add any other thing you want to pass in llava_model_args\n",
    "\n",
    "model.eval()\n",
    "\n",
    "url = \"https://github.com/haotian-liu/LLaVA/blob/1a91fc274d7c35a9b50b3cb29c4247ae5837ce39/images/llava_v1_5_radar.jpg?raw=true\"\n",
    "image = Image.open(requests.get(url, stream=True).raw)\n",
    "image_tensor = process_images([image], image_processor, model.config)\n",
    "image_tensor = [_image.to(dtype=torch.float16, device=device) for _image in image_tensor]\n",
    "\n",
    "conv_template = \"qwen_1_5\"  # Make sure you use correct chat template for different models\n",
    "question = DEFAULT_IMAGE_TOKEN + \"\\nWhat is shown in this image?\"\n",
    "conv = copy.deepcopy(conv_templates[conv_template])\n",
    "conv.append_message(conv.roles[0], question)\n",
    "conv.append_message(conv.roles[1], None)\n",
    "prompt_question = conv.get_prompt()\n",
    "\n",
    "input_ids = tokenizer_image_token(prompt_question, tokenizer, IMAGE_TOKEN_INDEX, return_tensors=\"pt\").unsqueeze(0).to(device)\n",
    "image_sizes = [image.size]\n",
    "\n",
    "\n",
    "cont = model.generate(\n",
    "    input_ids,\n",
    "    images=image_tensor,\n",
    "    image_sizes=image_sizes,\n",
    "    do_sample=False,\n",
    "    temperature=0,\n",
    "    max_new_tokens=4096,\n",
    ")\n",
    "text_outputs = tokenizer.batch_decode(cont, skip_special_tokens=True)\n",
    "print(text_outputs)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "You could use the following code to make it streaming in terminal, this would be pretty useful when creating a chatbot."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from threading import Thread\n",
    "from transformers import TextIteratorStreamer\n",
    "import json\n",
    "\n",
    "url = \"https://github.com/haotian-liu/LLaVA/blob/1a91fc274d7c35a9b50b3cb29c4247ae5837ce39/images/llava_v1_5_radar.jpg?raw=true\"\n",
    "image = Image.open(requests.get(url, stream=True).raw)\n",
    "image_tensor = process_images([image], image_processor, model.config)\n",
    "image_tensor = [_image.to(dtype=torch.float16, device=device) for _image in image_tensor]\n",
    "\n",
    "conv_template = \"qwen_1_5\"\n",
    "question = DEFAULT_IMAGE_TOKEN + \"\\nWhat is shown in this image?\"\n",
    "conv = copy.deepcopy(conv_templates[conv_template])\n",
    "conv.append_message(conv.roles[0], question)\n",
    "conv.append_message(conv.roles[1], None)\n",
    "prompt_question = conv.get_prompt()\n",
    "\n",
    "input_ids = tokenizer_image_token(prompt_question, tokenizer, IMAGE_TOKEN_INDEX, return_tensors=\"pt\").unsqueeze(0).to(device)\n",
    "image_sizes = [image.size]\n",
    "\n",
    "max_context_length = getattr(model.config, \"max_position_embeddings\", 2048)\n",
    "num_image_tokens = question.count(DEFAULT_IMAGE_TOKEN) * model.get_vision_tower().num_patches\n",
    "\n",
    "streamer = TextIteratorStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True, timeout=15)\n",
    "\n",
    "max_new_tokens = min(4096, max_context_length - input_ids.shape[-1] - num_image_tokens)\n",
    "\n",
    "if max_new_tokens < 1:\n",
    "    print(\n",
    "        json.dumps(\n",
    "            {\n",
    "                \"text\": question + \"Exceeds max token length. Please start a new conversation, thanks.\",\n",
    "                \"error_code\": 0,\n",
    "            }\n",
    "        )\n",
    "    )\n",
    "else:\n",
    "    gen_kwargs = {\n",
    "        \"do_sample\": False,\n",
    "        \"temperature\": 0,\n",
    "        \"max_new_tokens\": max_new_tokens,\n",
    "        \"images\": image_tensor,\n",
    "        \"image_sizes\": image_sizes,\n",
    "    }\n",
    "\n",
    "    thread = Thread(\n",
    "        target=model.generate,\n",
    "        kwargs=dict(\n",
    "            inputs=input_ids,\n",
    "            streamer=streamer,\n",
    "            **gen_kwargs,\n",
    "        ),\n",
    "    )\n",
    "    thread.start()\n",
    "\n",
    "    generated_text = \"\"\n",
    "    for new_text in streamer:\n",
    "        generated_text += new_text\n",
    "        print(generated_text, flush=True)\n",
    "        # print(json.dumps({\"text\": generated_text, \"error_code\": 0}), flush=True)\n",
    "\n",
    "    print(\"Final output:\", generated_text)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Image-Text Interleaved Input\n",
    "\n",
    "Now switching to our onevision model for more complex tasks. You should start to use `llava-onevision-qwen2-0.5b-ov` for image-text interleaved input and video input.\n",
    "\n",
    "Processing image-text interleaved input is a bit more complicated. But following the code below should work."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Load model\n",
    "pretrained = \"lmms-lab/llava-onevision-qwen2-0.5b-ov\"\n",
    "model_name = \"llava_qwen\"\n",
    "device = \"cuda\"\n",
    "device_map = \"auto\"\n",
    "llava_model_args = {\n",
    "        \"multimodal\": True,\n",
    "    }\n",
    "overwrite_config = {}\n",
    "overwrite_config[\"image_aspect_ratio\"] = \"pad\"\n",
    "llava_model_args[\"overwrite_config\"] = overwrite_config\n",
    "tokenizer, model, image_processor, max_length = load_pretrained_model(pretrained, None, model_name, device_map=device_map, **llava_model_args)\n",
    "\n",
    "model.eval()\n",
    "\n",
    "# Load two images\n",
    "url1 = \"https://github.com/haotian-liu/LLaVA/blob/1a91fc274d7c35a9b50b3cb29c4247ae5837ce39/images/llava_v1_5_radar.jpg?raw=true\"\n",
    "url2 = \"https://raw.githubusercontent.com/haotian-liu/LLaVA/main/images/llava_logo.png\"\n",
    "\n",
    "image1 = Image.open(requests.get(url1, stream=True).raw)\n",
    "image2 = Image.open(requests.get(url2, stream=True).raw)\n",
    "\n",
    "images = [image1, image2]\n",
    "image_tensors = process_images(images, image_processor, model.config)\n",
    "image_tensors = [_image.to(dtype=torch.float16, device=device) for _image in image_tensors]\n",
    "\n",
    "# Prepare interleaved text-image input\n",
    "conv_template = \"qwen_1_5\"\n",
    "question = f\"{DEFAULT_IMAGE_TOKEN} This is the first image. Can you describe what you see?\\n\\nNow, let's look at another image: {DEFAULT_IMAGE_TOKEN}\\nWhat's the difference between these two images?\"\n",
    "\n",
    "conv = copy.deepcopy(conv_templates[conv_template])\n",
    "conv.append_message(conv.roles[0], question)\n",
    "conv.append_message(conv.roles[1], None)\n",
    "prompt_question = conv.get_prompt()\n",
    "\n",
    "input_ids = tokenizer_image_token(prompt_question, tokenizer, IMAGE_TOKEN_INDEX, return_tensors=\"pt\").unsqueeze(0).to(device)\n",
    "image_sizes = [image.size for image in images]\n",
    "\n",
    "# Generate response\n",
    "cont = model.generate(\n",
    "    input_ids,\n",
    "    images=image_tensors,\n",
    "    image_sizes=image_sizes,\n",
    "    do_sample=False,\n",
    "    temperature=0,\n",
    "    max_new_tokens=4096,\n",
    ")\n",
    "text_outputs = tokenizer.batch_decode(cont, skip_special_tokens=True)\n",
    "print(text_outputs[0])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Video Input\n",
    "\n",
    "Now let's try video input. It's the same as image input, but you need to pass in a list of video frames. And remember to set the `<image>` token only once in the prompt, e.g. \"<image>\\nWhat is shown in this video?\", not \"<image>\\n<image>\\n<image>\\nWhat is shown in this video?\". Since we trained on this format, it's important to keep the format consistent."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/tiger/miniconda3/envs/public_llava/lib/python3.10/site-packages/tqdm/auto.py:21: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html\n",
      "  from .autonotebook import tqdm as notebook_tqdm\n",
      "/home/tiger/miniconda3/envs/public_llava/lib/python3.10/site-packages/huggingface_hub/file_download.py:1150: FutureWarning: `resume_download` is deprecated and will be removed in version 1.0.0. Downloads always resume when possible. If you want to force a new download, use `force_download=True`.\n",
      "  warnings.warn(\n",
      "Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Loaded LLaVA model: lmms-lab/llava-onevision-qwen2-7b-ov\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.\n",
      "You are using a model of type llava to instantiate a model of type llava_qwen. This is not supported for all configurations of models and can yield errors.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Loading vision tower: google/siglip-so400m-patch14-384\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Loading checkpoint shards: 100%|██████████| 4/4 [00:08<00:00,  2.07s/it]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model Class: LlavaQwenForCausalLM\n",
      "(16, 1024, 576, 3)\n",
      "The video features a person standing on a stage, dressed in a black shirt and dark pants. A large hand appears from the background, reaching towards the person's pocket. The text 'Source: Joshua AG' is displayed at the top left corner of the frames, and 'EVAN CARMICHAEL' is shown in the top right corner. The text 'Anyone know what this pocket is for?' appears as the hand continues to reach into the pocket. The person then looks down at their pocket, and the text 'I've always wondered that' appears. The hand finally pulls out a small white device labeled 'iPod Nano'. The person holds up the iPod Nano, and the text 'is the new iPod Nano' appears. The video concludes with a close-up of the person holding the iPod Nano, showing it from different angles.\n"
     ]
    }
   ],
   "source": [
    "from operator import attrgetter\n",
    "from llava.model.builder import load_pretrained_model\n",
    "from llava.mm_utils import get_model_name_from_path, process_images, tokenizer_image_token\n",
    "from llava.constants import IMAGE_TOKEN_INDEX, DEFAULT_IMAGE_TOKEN, DEFAULT_IM_START_TOKEN, DEFAULT_IM_END_TOKEN, IGNORE_INDEX\n",
    "from llava.conversation import conv_templates, SeparatorStyle\n",
    "\n",
    "import torch\n",
    "import cv2\n",
    "import numpy as np\n",
    "from PIL import Image\n",
    "import requests\n",
    "import copy\n",
    "import warnings\n",
    "from decord import VideoReader, cpu\n",
    "\n",
    "warnings.filterwarnings(\"ignore\")\n",
    "# Load the OneVision model\n",
    "pretrained = \"lmms-lab/llava-onevision-qwen2-7b-ov\"\n",
    "model_name = \"llava_qwen\"\n",
    "device = \"cuda\"\n",
    "device_map = \"auto\"\n",
    "llava_model_args = {\n",
    "    \"multimodal\": True,\n",
    "}\n",
    "tokenizer, model, image_processor, max_length = load_pretrained_model(pretrained, None, model_name, device_map=device_map, attn_implementation=\"sdpa\", **llava_model_args)\n",
    "\n",
    "model.eval()\n",
    "\n",
    "\n",
    "# Function to extract frames from video\n",
    "def load_video(video_path, max_frames_num):\n",
    "    if type(video_path) == str:\n",
    "        vr = VideoReader(video_path, ctx=cpu(0))\n",
    "    else:\n",
    "        vr = VideoReader(video_path[0], ctx=cpu(0))\n",
    "    total_frame_num = len(vr)\n",
    "    uniform_sampled_frames = np.linspace(0, total_frame_num - 1, max_frames_num, dtype=int)\n",
    "    frame_idx = uniform_sampled_frames.tolist()\n",
    "    spare_frames = vr.get_batch(frame_idx).asnumpy()\n",
    "    return spare_frames  # (frames, height, width, channels)\n",
    "\n",
    "\n",
    "# Load and process video\n",
    "video_path = \"jobs.mp4\"\n",
    "video_frames = load_video(video_path, 16)\n",
    "print(video_frames.shape) # (16, 1024, 576, 3)\n",
    "image_tensors = []\n",
    "frames = image_processor.preprocess(video_frames, return_tensors=\"pt\")[\"pixel_values\"].half().cuda()\n",
    "image_tensors.append(frames)\n",
    "\n",
    "# Prepare conversation input\n",
    "conv_template = \"qwen_1_5\"\n",
    "question = f\"{DEFAULT_IMAGE_TOKEN}\\nDescribe what's happening in this video.\"\n",
    "\n",
    "conv = copy.deepcopy(conv_templates[conv_template])\n",
    "conv.append_message(conv.roles[0], question)\n",
    "conv.append_message(conv.roles[1], None)\n",
    "prompt_question = conv.get_prompt()\n",
    "\n",
    "input_ids = tokenizer_image_token(prompt_question, tokenizer, IMAGE_TOKEN_INDEX, return_tensors=\"pt\").unsqueeze(0).to(device)\n",
    "image_sizes = [frame.size for frame in video_frames]\n",
    "\n",
    "# Generate response\n",
    "cont = model.generate(\n",
    "    input_ids,\n",
    "    images=image_tensors,\n",
    "    image_sizes=image_sizes,\n",
    "    do_sample=False,\n",
    "    temperature=0,\n",
    "    max_new_tokens=4096,\n",
    "    modalities=[\"video\"],\n",
    ")\n",
    "text_outputs = tokenizer.batch_decode(cont, skip_special_tokens=True)\n",
    "print(text_outputs[0])"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3.9.2 64-bit",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.10.14"
  },
  "vscode": {
   "interpreter": {
    "hash": "31f2aee4e71d21fbe5cf8b01ff0e069b9275f58929596ceb00d14d90e3e16cd6"
   }
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}