Image-Text-to-Text
Transformers
Safetensors
llava_onevision2_5
text-generation
multimodal
video-understanding
streaming
llava-onevision
conversational
custom_code
Instructions to use lmms-lab-encoder/LLaVA-OneVision-2.5-4B-Instruct with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use lmms-lab-encoder/LLaVA-OneVision-2.5-4B-Instruct with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("image-text-to-text", model="lmms-lab-encoder/LLaVA-OneVision-2.5-4B-Instruct", trust_remote_code=True) messages = [ { "role": "user", "content": [ {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"}, {"type": "text", "text": "What animal is on the candy?"} ] }, ] pipe(text=messages)# pip install -U transformers accelerate # Load model directly from transformers import AutoModelForCausalLM model = AutoModelForCausalLM.from_pretrained("lmms-lab-encoder/LLaVA-OneVision-2.5-4B-Instruct", trust_remote_code=True, device_map="auto") - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use lmms-lab-encoder/LLaVA-OneVision-2.5-4B-Instruct with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "lmms-lab-encoder/LLaVA-OneVision-2.5-4B-Instruct" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "lmms-lab-encoder/LLaVA-OneVision-2.5-4B-Instruct", "messages": [ { "role": "user", "content": [ { "type": "text", "text": "Describe this image in one sentence." }, { "type": "image_url", "image_url": { "url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg" } } ] } ] }'Use Docker
docker model run hf.co/lmms-lab-encoder/LLaVA-OneVision-2.5-4B-Instruct
- SGLang
How to use lmms-lab-encoder/LLaVA-OneVision-2.5-4B-Instruct with SGLang:
Install from pip and serve model
# Install SGLang from pip: pip install sglang # Start the SGLang server: python3 -m sglang.launch_server \ --model-path "lmms-lab-encoder/LLaVA-OneVision-2.5-4B-Instruct" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "lmms-lab-encoder/LLaVA-OneVision-2.5-4B-Instruct", "messages": [ { "role": "user", "content": [ { "type": "text", "text": "Describe this image in one sentence." }, { "type": "image_url", "image_url": { "url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg" } } ] } ] }'Use Docker images
docker run --gpus all \ --shm-size 32g \ -p 30000:30000 \ -v ~/.cache/huggingface:/root/.cache/huggingface \ --env "HF_TOKEN=<secret>" \ --ipc=host \ lmsysorg/sglang:latest \ python3 -m sglang.launch_server \ --model-path "lmms-lab-encoder/LLaVA-OneVision-2.5-4B-Instruct" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "lmms-lab-encoder/LLaVA-OneVision-2.5-4B-Instruct", "messages": [ { "role": "user", "content": [ { "type": "text", "text": "Describe this image in one sentence." }, { "type": "image_url", "image_url": { "url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg" } } ] } ] }' - Docker Model Runner
How to use lmms-lab-encoder/LLaVA-OneVision-2.5-4B-Instruct with Docker Model Runner:
docker model run hf.co/lmms-lab-encoder/LLaVA-OneVision-2.5-4B-Instruct
Model card: render both benchmark tables at full, equal width
Browse files- README.md +50 -32
- README_CN.md +50 -32
README.md
CHANGED
|
@@ -36,41 +36,59 @@ LLaVA-OneVision-2.5-4B-Instruct is a 4B vision-language model for single-image,
|
|
| 36 |
|
| 37 |
### Video
|
| 38 |
|
| 39 |
-
|
| 40 |
-
|
| 41 |
-
|
| 42 |
-
|
| 43 |
-
|
| 44 |
-
|
| 45 |
-
|
| 46 |
-
|
| 47 |
-
|
| 48 |
-
|
| 49 |
-
|
| 50 |
-
|
| 51 |
-
|
| 52 |
-
|
| 53 |
-
|
| 54 |
-
|
| 55 |
-
|
| 56 |
-
|
| 57 |
-
|
| 58 |
-
|
| 59 |
-
|
| 60 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 61 |
|
| 62 |
### Image
|
| 63 |
|
| 64 |
-
|
| 65 |
-
|
| 66 |
-
|
| 67 |
-
|
| 68 |
-
|
| 69 |
-
|
| 70 |
-
|
| 71 |
-
|
| 72 |
-
|
| 73 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 74 |
|
| 75 |
<sub>Compared with Qwen3.5-4B, thinking disabled. **Bold** is the best score in each row; `--` means not available. The superscript on each video score is the setting that produced it: frames × pixels per frame, or for codec, canvases (TC) × pixels per canvas. VideoMME is the mean over the short, medium and long splits. Each side reports its best score over the settings tried, from single runs. `*` On TOMATO the two sides use different frame settings. The image table shows selected benchmarks; full results, settings and protocol will be given in the technical report.</sub>
|
| 76 |
|
|
|
|
| 36 |
|
| 37 |
### Video
|
| 38 |
|
| 39 |
+
<table style="width:100%; display:table;">
|
| 40 |
+
<thead><tr>
|
| 41 |
+
<th align="left" style="width:28%">Benchmark</th>
|
| 42 |
+
<th align="center" style="width:24%">Qwen3.5-4B</th>
|
| 43 |
+
<th align="center" style="width:24%">OV2.5-4B (frames)</th>
|
| 44 |
+
<th align="center" style="width:24%">OV2.5-4B (codec)</th>
|
| 45 |
+
</tr></thead>
|
| 46 |
+
<tbody>
|
| 47 |
+
<tr><td colspan="4" align="left"><b>General video understanding</b></td></tr>
|
| 48 |
+
<tr><td align="left">VideoMME (no subtitles)</td><td align="center">71.7</td><td align="center">69.7</td><td align="center"><b>71.9</b></td></tr>
|
| 49 |
+
<tr><td align="left">VideoMME (subtitles)</td><td align="center"><b>75.9</b></td><td align="center">75.2</td><td align="center">74.7</td></tr>
|
| 50 |
+
<tr><td align="left">MotionBench</td><td align="center">60.1 <sup>128f×153,664</sup></td><td align="center">62.4 <sup>128f×153,664</sup></td><td align="center"><b>64.2</b> <sup>TC128×313,600</sup></td></tr>
|
| 51 |
+
<tr><td align="left">TOMATO *</td><td align="center">34.9 <sup>16f×153,664, fps=2</sup></td><td align="center">33.6 <sup>model default</sup></td><td align="center"><b>38.6</b> <sup>TC128×313,600</sup></td></tr>
|
| 52 |
+
<tr><td align="left">MVBench</td><td align="center">64.0 <sup>64f×153,664</sup></td><td align="center">72.0 <sup>64f×153,664</sup></td><td align="center"><b>75.5</b> <sup>TC64×313,600</sup></td></tr>
|
| 53 |
+
<tr><td align="left">NextQA MC test</td><td align="center">80.7 <sup>64f×153,664</sup></td><td align="center">86.1 <sup>64f×153,664</sup></td><td align="center"><b>86.4</b> <sup>TC64×313,600</sup></td></tr>
|
| 54 |
+
<tr><td align="left">TempCompass MC</td><td align="center">73.4 <sup>64f×153,664</sup></td><td align="center">75.8 <sup>64f×153,664</sup></td><td align="center"><b>79.7</b> <sup>TC64×313,600</sup></td></tr>
|
| 55 |
+
<tr><td align="left">Video-MME-v2</td><td align="center">15.5 <sup>64f×330,000</sup></td><td align="center">16.4 <sup>64f×330,000</sup></td><td align="center"><b>17.2</b> <sup>TC64×313,600</sup></td></tr>
|
| 56 |
+
<tr><td align="left">EgoSchema subset</td><td align="center">68.2 <sup>64f×153,664</sup></td><td align="center"><b>80.0</b> <sup>64f×153,664</sup></td><td align="center">--</td></tr>
|
| 57 |
+
<tr><td align="left">Video-Index 512f</td><td align="center">19.6 <sup>task frames, ≤262,144/f</sup></td><td align="center"><b>20.0</b> <sup>task frames, ≤262,144/f</sup></td><td align="center">19.8 <sup>TC≤256×313,600</sup></td></tr>
|
| 58 |
+
<tr><td colspan="4" align="left"><b>Long video</b></td></tr>
|
| 59 |
+
<tr><td align="left">LVBench</td><td align="center">56.5 <sup>2048f×120,000</sup></td><td align="center"><b>62.0</b> <sup>950f×200,000</sup></td><td align="center">61.6 <sup>TC1024×156,800</sup></td></tr>
|
| 60 |
+
<tr><td align="left">MLVU dev</td><td align="center">73.8 <sup>512f×72,900</sup></td><td align="center">75.3 <sup>512f×72,900</sup></td><td align="center"><b>77.1</b> <sup>TC512×313,600</sup></td></tr>
|
| 61 |
+
<tr><td align="left">LongVideoBench val</td><td align="center">62.7 <sup>700f×300,000</sup></td><td align="center">69.0 <sup>1280f×80,000</sup></td><td align="center"><b>71.1</b> <sup>TC768×235,200</sup></td></tr>
|
| 62 |
+
<tr><td align="left">VideoEval-Pro</td><td align="center">62.8 <sup>2048f×100,000</sup></td><td align="center">64.4 <sup>950f×200,000</sup></td><td align="center"><b>64.6</b> <sup>TC768×235,200</sup></td></tr>
|
| 63 |
+
<tr><td colspan="4" align="left"><b>Temporal grounding (mIoU)</b></td></tr>
|
| 64 |
+
<tr><td align="left">TimeLens ActivityNet</td><td align="center">53.8 <sup>2048f×409,600, fps=4</sup></td><td align="center">58.8 <sup>2048f×409,600, fps=4</sup></td><td align="center"><b>59.1</b> <sup>TC128×313,600</sup></td></tr>
|
| 65 |
+
<tr><td align="left">TimeLens Charades</td><td align="center">48.1 <sup>2048f×409,600, fps=4</sup></td><td align="center">60.6 <sup>2048f×409,600, fps=4</sup></td><td align="center"><b>61.3</b> <sup>TC128×313,600</sup></td></tr>
|
| 66 |
+
<tr><td align="left">TimeLens QVHighlights</td><td align="center">67.5 <sup>2048f×409,600, fps=4</sup></td><td align="center">68.4 <sup>2048f×409,600, fps=4</sup></td><td align="center"><b>70.3</b> <sup>TC128×313,600</sup></td></tr>
|
| 67 |
+
</tbody>
|
| 68 |
+
</table>
|
| 69 |
|
| 70 |
### Image
|
| 71 |
|
| 72 |
+
<table style="width:100%; display:table;">
|
| 73 |
+
<thead><tr>
|
| 74 |
+
<th align="left" style="width:24%">General perception and reading</th>
|
| 75 |
+
<th align="center" style="width:13%">Qwen3.5-4B</th>
|
| 76 |
+
<th align="center" style="width:13%">OV2.5-4B</th>
|
| 77 |
+
<th align="left" style="width:24%">Grounding and spatial</th>
|
| 78 |
+
<th align="center" style="width:13%">Qwen3.5-4B</th>
|
| 79 |
+
<th align="center" style="width:13%">OV2.5-4B</th>
|
| 80 |
+
</tr></thead>
|
| 81 |
+
<tbody>
|
| 82 |
+
<tr><td align="left">MMStar</td><td align="center"><b>66.9</b></td><td align="center">63.9</td><td align="left">RefCOCO val (Acc@0.5)</td><td align="center">89.1</td><td align="center"><b>92.6</b></td></tr>
|
| 83 |
+
<tr><td align="left">MMBench-EN dev</td><td align="center"><b>88.8</b></td><td align="center">88.2</td><td align="left">RefCOCOg val (Acc@0.5)</td><td align="center">87.7</td><td align="center"><b>88.9</b></td></tr>
|
| 84 |
+
<tr><td align="left">RealWorldQA</td><td align="center"><b>74.4</b></td><td align="center">73.9</td><td align="left">RefSpatial-Bench</td><td align="center">29.1</td><td align="center"><b>62.1</b></td></tr>
|
| 85 |
+
<tr><td align="left">AI2D</td><td align="center"><b>82.4</b></td><td align="center">82.2</td><td align="left">CrossPoint</td><td align="center">33.7</td><td align="center"><b>80.5</b></td></tr>
|
| 86 |
+
<tr><td align="left">BLINK</td><td align="center">63.6</td><td align="center"><b>67.1</b></td><td align="left">CV-Bench 2D</td><td align="center">80.7</td><td align="center"><b>83.3</b></td></tr>
|
| 87 |
+
<tr><td align="left">V*-Bench</td><td align="center">81.7</td><td align="center"><b>89.5</b></td><td align="left">CV-Bench 3D</td><td align="center">92.0</td><td align="center"><b>94.5</b></td></tr>
|
| 88 |
+
<tr><td align="left">OCRBench</td><td align="center"><b>85.2</b></td><td align="center">79.8</td><td align="left">EmbSpatial</td><td align="center">74.7</td><td align="center"><b>83.0</b></td></tr>
|
| 89 |
+
<tr><td align="left">DocVQA val</td><td align="center"><b>95.2</b></td><td align="center">92.9</td><td align="left">MMSI-Bench</td><td align="center">31.4</td><td align="center"><b>33.3</b></td></tr>
|
| 90 |
+
</tbody>
|
| 91 |
+
</table>
|
| 92 |
|
| 93 |
<sub>Compared with Qwen3.5-4B, thinking disabled. **Bold** is the best score in each row; `--` means not available. The superscript on each video score is the setting that produced it: frames × pixels per frame, or for codec, canvases (TC) × pixels per canvas. VideoMME is the mean over the short, medium and long splits. Each side reports its best score over the settings tried, from single runs. `*` On TOMATO the two sides use different frame settings. The image table shows selected benchmarks; full results, settings and protocol will be given in the technical report.</sub>
|
| 94 |
|
README_CN.md
CHANGED
|
@@ -36,41 +36,59 @@ LLaVA-OneVision-2.5-4B-Instruct 是一个 4B 视觉语言模型,支持单图
|
|
| 36 |
|
| 37 |
### 视频
|
| 38 |
|
| 39 |
-
|
| 40 |
-
|
| 41 |
-
|
| 42 |
-
|
| 43 |
-
|
| 44 |
-
|
| 45 |
-
|
| 46 |
-
|
| 47 |
-
|
| 48 |
-
|
| 49 |
-
|
| 50 |
-
|
| 51 |
-
|
| 52 |
-
|
| 53 |
-
|
| 54 |
-
|
| 55 |
-
|
| 56 |
-
|
| 57 |
-
|
| 58 |
-
|
| 59 |
-
|
| 60 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 61 |
|
| 62 |
### 图像
|
| 63 |
|
| 64 |
-
|
| 65 |
-
|
| 66 |
-
|
| 67 |
-
|
| 68 |
-
|
| 69 |
-
|
| 70 |
-
|
| 71 |
-
|
| 72 |
-
|
| 73 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 74 |
|
| 75 |
<sub>对比对象为关闭 thinking 的 Qwen3.5-4B。**加粗**为该行最高分,`--` 表示暂无结果。视频分数的上标是取得该分数时的设置:帧数 × 每帧像素;codec 为画布数(TC)× 每画布像素。VideoMME 为 short、medium、long 三个子集的平均。两边各取已测设置中的最高分,均为单次运行。`*` TOMATO 两边帧数设置不同。图像表为代表性 bench 的节选,完整结果、设置与评测口径见后续技术报告。</sub>
|
| 76 |
|
|
|
|
| 36 |
|
| 37 |
### 视频
|
| 38 |
|
| 39 |
+
<table style="width:100%; display:table;">
|
| 40 |
+
<thead><tr>
|
| 41 |
+
<th align="left" style="width:28%">Benchmark</th>
|
| 42 |
+
<th align="center" style="width:24%">Qwen3.5-4B</th>
|
| 43 |
+
<th align="center" style="width:24%">OV2.5-4B(抽帧)</th>
|
| 44 |
+
<th align="center" style="width:24%">OV2.5-4B(codec)</th>
|
| 45 |
+
</tr></thead>
|
| 46 |
+
<tbody>
|
| 47 |
+
<tr><td colspan="4" align="left"><b>通用视频理解</b></td></tr>
|
| 48 |
+
<tr><td align="left">VideoMME(无字幕)</td><td align="center">71.7</td><td align="center">69.7</td><td align="center"><b>71.9</b></td></tr>
|
| 49 |
+
<tr><td align="left">VideoMME(有字幕)</td><td align="center"><b>75.9</b></td><td align="center">75.2</td><td align="center">74.7</td></tr>
|
| 50 |
+
<tr><td align="left">MotionBench</td><td align="center">60.1 <sup>128f×153,664</sup></td><td align="center">62.4 <sup>128f×153,664</sup></td><td align="center"><b>64.2</b> <sup>TC128×313,600</sup></td></tr>
|
| 51 |
+
<tr><td align="left">TOMATO *</td><td align="center">34.9 <sup>16f×153,664, fps=2</sup></td><td align="center">33.6 <sup>model default</sup></td><td align="center"><b>38.6</b> <sup>TC128×313,600</sup></td></tr>
|
| 52 |
+
<tr><td align="left">MVBench</td><td align="center">64.0 <sup>64f×153,664</sup></td><td align="center">72.0 <sup>64f×153,664</sup></td><td align="center"><b>75.5</b> <sup>TC64×313,600</sup></td></tr>
|
| 53 |
+
<tr><td align="left">NextQA MC test</td><td align="center">80.7 <sup>64f×153,664</sup></td><td align="center">86.1 <sup>64f×153,664</sup></td><td align="center"><b>86.4</b> <sup>TC64×313,600</sup></td></tr>
|
| 54 |
+
<tr><td align="left">TempCompass MC</td><td align="center">73.4 <sup>64f×153,664</sup></td><td align="center">75.8 <sup>64f×153,664</sup></td><td align="center"><b>79.7</b> <sup>TC64×313,600</sup></td></tr>
|
| 55 |
+
<tr><td align="left">Video-MME-v2</td><td align="center">15.5 <sup>64f×330,000</sup></td><td align="center">16.4 <sup>64f×330,000</sup></td><td align="center"><b>17.2</b> <sup>TC64×313,600</sup></td></tr>
|
| 56 |
+
<tr><td align="left">EgoSchema subset</td><td align="center">68.2 <sup>64f×153,664</sup></td><td align="center"><b>80.0</b> <sup>64f×153,664</sup></td><td align="center">--</td></tr>
|
| 57 |
+
<tr><td align="left">Video-Index 512f</td><td align="center">19.6 <sup>task frames, ≤262,144/f</sup></td><td align="center"><b>20.0</b> <sup>task frames, ≤262,144/f</sup></td><td align="center">19.8 <sup>TC≤256×313,600</sup></td></tr>
|
| 58 |
+
<tr><td colspan="4" align="left"><b>长视频</b></td></tr>
|
| 59 |
+
<tr><td align="left">LVBench</td><td align="center">56.5 <sup>2048f×120,000</sup></td><td align="center"><b>62.0</b> <sup>950f×200,000</sup></td><td align="center">61.6 <sup>TC1024×156,800</sup></td></tr>
|
| 60 |
+
<tr><td align="left">MLVU dev</td><td align="center">73.8 <sup>512f×72,900</sup></td><td align="center">75.3 <sup>512f×72,900</sup></td><td align="center"><b>77.1</b> <sup>TC512×313,600</sup></td></tr>
|
| 61 |
+
<tr><td align="left">LongVideoBench val</td><td align="center">62.7 <sup>700f×300,000</sup></td><td align="center">69.0 <sup>1280f×80,000</sup></td><td align="center"><b>71.1</b> <sup>TC768×235,200</sup></td></tr>
|
| 62 |
+
<tr><td align="left">VideoEval-Pro</td><td align="center">62.8 <sup>2048f×100,000</sup></td><td align="center">64.4 <sup>950f×200,000</sup></td><td align="center"><b>64.6</b> <sup>TC768×235,200</sup></td></tr>
|
| 63 |
+
<tr><td colspan="4" align="left"><b>时间定位(mIoU)</b></td></tr>
|
| 64 |
+
<tr><td align="left">TimeLens ActivityNet</td><td align="center">53.8 <sup>2048f×409,600, fps=4</sup></td><td align="center">58.8 <sup>2048f×409,600, fps=4</sup></td><td align="center"><b>59.1</b> <sup>TC128×313,600</sup></td></tr>
|
| 65 |
+
<tr><td align="left">TimeLens Charades</td><td align="center">48.1 <sup>2048f×409,600, fps=4</sup></td><td align="center">60.6 <sup>2048f×409,600, fps=4</sup></td><td align="center"><b>61.3</b> <sup>TC128×313,600</sup></td></tr>
|
| 66 |
+
<tr><td align="left">TimeLens QVHighlights</td><td align="center">67.5 <sup>2048f×409,600, fps=4</sup></td><td align="center">68.4 <sup>2048f×409,600, fps=4</sup></td><td align="center"><b>70.3</b> <sup>TC128×313,600</sup></td></tr>
|
| 67 |
+
</tbody>
|
| 68 |
+
</table>
|
| 69 |
|
| 70 |
### 图像
|
| 71 |
|
| 72 |
+
<table style="width:100%; display:table;">
|
| 73 |
+
<thead><tr>
|
| 74 |
+
<th align="left" style="width:24%">通用感知与阅读</th>
|
| 75 |
+
<th align="center" style="width:13%">Qwen3.5-4B</th>
|
| 76 |
+
<th align="center" style="width:13%">OV2.5-4B</th>
|
| 77 |
+
<th align="left" style="width:24%">定位与空间</th>
|
| 78 |
+
<th align="center" style="width:13%">Qwen3.5-4B</th>
|
| 79 |
+
<th align="center" style="width:13%">OV2.5-4B</th>
|
| 80 |
+
</tr></thead>
|
| 81 |
+
<tbody>
|
| 82 |
+
<tr><td align="left">MMStar</td><td align="center"><b>66.9</b></td><td align="center">63.9</td><td align="left">RefCOCO val (Acc@0.5)</td><td align="center">89.1</td><td align="center"><b>92.6</b></td></tr>
|
| 83 |
+
<tr><td align="left">MMBench-EN dev</td><td align="center"><b>88.8</b></td><td align="center">88.2</td><td align="left">RefCOCOg val (Acc@0.5)</td><td align="center">87.7</td><td align="center"><b>88.9</b></td></tr>
|
| 84 |
+
<tr><td align="left">RealWorldQA</td><td align="center"><b>74.4</b></td><td align="center">73.9</td><td align="left">RefSpatial-Bench</td><td align="center">29.1</td><td align="center"><b>62.1</b></td></tr>
|
| 85 |
+
<tr><td align="left">AI2D</td><td align="center"><b>82.4</b></td><td align="center">82.2</td><td align="left">CrossPoint</td><td align="center">33.7</td><td align="center"><b>80.5</b></td></tr>
|
| 86 |
+
<tr><td align="left">BLINK</td><td align="center">63.6</td><td align="center"><b>67.1</b></td><td align="left">CV-Bench 2D</td><td align="center">80.7</td><td align="center"><b>83.3</b></td></tr>
|
| 87 |
+
<tr><td align="left">V*-Bench</td><td align="center">81.7</td><td align="center"><b>89.5</b></td><td align="left">CV-Bench 3D</td><td align="center">92.0</td><td align="center"><b>94.5</b></td></tr>
|
| 88 |
+
<tr><td align="left">OCRBench</td><td align="center"><b>85.2</b></td><td align="center">79.8</td><td align="left">EmbSpatial</td><td align="center">74.7</td><td align="center"><b>83.0</b></td></tr>
|
| 89 |
+
<tr><td align="left">DocVQA val</td><td align="center"><b>95.2</b></td><td align="center">92.9</td><td align="left">MMSI-Bench</td><td align="center">31.4</td><td align="center"><b>33.3</b></td></tr>
|
| 90 |
+
</tbody>
|
| 91 |
+
</table>
|
| 92 |
|
| 93 |
<sub>对比对象为关闭 thinking 的 Qwen3.5-4B。**加粗**为该行最高分,`--` 表示暂无结果。视频分数的上标是取得该分数时的设置:帧数 × 每帧像素;codec 为画布数(TC)× 每画布像素。VideoMME 为 short、medium、long 三个子集的平均。两边各取已测设置中的最高分,均为单次运行。`*` TOMATO 两边帧数设置不同。图像表为代表性 bench 的节选,完整结果、设置与评测口径见后续技术报告。</sub>
|
| 94 |
|