Hugging Face's logo

OpenGVLab
/
InternVL2-26B

Image-Text-to-Text
Transformers
Safetensors
multilingual
internvl_chat
feature-extraction
internvl
custom_code
conversational
Model card Files Community
8
czczup commited on
Commit
8683961
·
verified ·
1 Parent(s): 8839270

Upload folder using huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +38 -37
README.md CHANGED
@@ -27,23 +27,23 @@ InternVL 2.0 is a multimodal large language model series, featuring models of va
27
  | :--------------------------: | :-------------: | :------------: | :----------------: | :-----------: |
28
  | Model Size | - | - | 25.5B | 25.5B |
29
  | | | | | |
30
- | DocVQA<sub>test</sub> | 87.2 | 86.5 | 90.9 | |
31
- | ChartQA<sub>test</sub> | 78.1 | 81.3 | 83.8 | |
32
- | InfoVQA<sub>test</sub> | - | 72.7 | 72.5 | |
33
- | TextVQA<sub>val</sub> | - | 73.5 | 80.6 | |
34
- | OCRBench | 678 | 754 | 724 | |
35
- | MME<sub>sum</sub> | 2070.2 | - | 2187.8 | |
36
- | RealWorldQA | 68.0 | 67.5 | 66.0 | |
37
- | AI2D<sub>test</sub> | 89.4 | 80.3 | 80.7 | |
38
- | MMMU<sub>val</sub> | 63.1 | 58.5 | 45.2 | |
39
- | MMBench-EN<sub>test</sub> | 81.0 | 73.9 | 82.2 | |
40
- | MMBench-CN<sub>test</sub> | 80.2 | 73.8 | 82.0 | |
41
- | CCBench<sub>dev</sub> | 57.3 | 28.4 | 69.8 | |
42
- | MMVet<sub>GPT-4-0613</sub> | - | - | 62.8 | |
43
- | MMVet<sub>GPT-4-Turbo</sub> | 67.5 | 64.0 | 55.4 | |
44
- | SEED-Image | - | - | 76.0 | |
45
- | HallBench<sub>avg</sub> | 43.9 | 45.6 | 49.3 | |
46
- | MathVista<sub>testmini</sub> | 58.1 | 57.7 | 53.5 | |
47
 
48
  - We simultaneously use InternVL and VLMEvalKit repositories for model evaluation. Specifically, the results reported for DocVQA, ChartQA, InfoVQA, TextVQA, MME, AI2D, MMBench, CCBench, MMVet, and SEED-Image were tested using the InternVL repository. MMMU, OCRBench, RealWorldQA, HallBench, and MathVista were evaluated using the VLMEvalKit.
49
 
@@ -263,26 +263,27 @@ InternVL 2.0 是一个多模态大语言模型系列，包含各种规模的模
263
 
264
  ## 性能测试
265
 
266
- | 评测数据集 | MiniCPM-Llama3-V-2_5 | InternVL-Chat-V1-5 | InternVL2-26B |
267
- | :--------------------------: | :------------------: | :----------------: | :-----------: |
268
- | 模型大小 | 8.5B | 25.5B | 25.5B |
269
- | | | | |
270
- | DocVQA<sub>test</sub> | 84.8 | 90.9 | |
271
- | ChartQA<sub>test</sub> | - | 83.8 | |
272
- | InfoVQA<sub>test</sub> | - | 72.5 | |
273
- | TextVQA<sub>val</sub> | 76.6 | 80.6 | |
274
- | OCRBench | 725 | 724 | |
275
- | MME<sub>sum</sub> | 2024.6 | 2187.8 | |
276
- | RealWorldQA | 63.5 | 66.0 | |
277
- | AI2D<sub>test</sub> | 78.4 | 80.7 | |
278
- | MMMU<sub>val</sub> | 45.8 | 45.2 | |
279
- | MMBench-EN<sub>test</sub> | 77.2 | 82.2 | |
280
- | MMBench-CN<sub>test</sub> | 74.2 | 82.0 | |
281
- | CCBench<sub>dev</sub> | 45.9 | 69.8 | |
282
- | MMVet<sub>GPT-4-0613</sub> | - | 62.8 | |
283
- | SEED-Image | 72.3 | 76.0 | |
284
- | HallBench<sub>avg</sub> | 42.4 | 49.3 | |
285
- | MathVista<sub>testmini</sub> | 54.3 | 53.5 | |
 
286
 
287
  - 我们同时使用 InternVL 和 VLMEvalKit 仓库进行模型评估。具体来说，DocVQA、ChartQA、InfoVQA、TextVQA、MME、AI2D、MMBench、CCBench、MMVet 和 SEED-Image 的结果是使用 InternVL 仓库测试的。MMMU、OCRBench、RealWorldQA、HallBench 和 MathVista 是使用 VLMEvalKit 进行评估的。
288
 
 
27
  | :--------------------------: | :-------------: | :------------: | :----------------: | :-----------: |
28
  | Model Size | - | - | 25.5B | 25.5B |
29
  | | | | | |
30
+ | DocVQA<sub>test</sub> | 87.2 | 86.5 | 90.9 | 92.9 |
31
+ | ChartQA<sub>test</sub> | 78.1 | 81.3 | 83.8 | 84.9 |
32
+ | InfoVQA<sub>test</sub> | - | 72.7 | 72.5 | 75.9 |
33
+ | TextVQA<sub>val</sub> | - | 73.5 | 80.6 | 82.3 |
34
+ | OCRBench | 678 | 754 | 724 | 825 |
35
+ | MME<sub>sum</sub> | 2070.2 | 2110.6 | 2187.8 | 2260.7 |
36
+ | RealWorldQA | 68.0 | 67.5 | 66.0 | 68.3 |
37
+ | AI2D<sub>test</sub> | 89.4 | 80.3 | 80.7 | 84.5 |
38
+ | MMMU<sub>val</sub> | 63.1 | 58.5 | 45.2 | 48.3 |
39
+ | MMBench-EN<sub>test</sub> | 81.0 | 73.9 | 82.2 | 83.4 |
40
+ | MMBench-CN<sub>test</sub> | 80.2 | 73.8 | 82.0 | 82.0 |
41
+ | CCBench<sub>dev</sub> | 57.3 | 28.4 | 69.8 | 73.5 |
42
+ | MMVet<sub>GPT-4-0613</sub> | - | - | 62.8 | 64.2 |
43
+ | MMVet<sub>GPT-4-Turbo</sub> | 67.5 | 64.0 | 55.4 | 62.1 |
44
+ | SEED-Image | - | - | 76.0 | 76.8 |
45
+ | HallBench<sub>avg</sub> | 43.9 | 45.6 | 49.3 | 50.7 |
46
+ | MathVista<sub>testmini</sub> | 58.1 | 57.7 | 53.5 | 59.4 |
47
 
48
  - We simultaneously use InternVL and VLMEvalKit repositories for model evaluation. Specifically, the results reported for DocVQA, ChartQA, InfoVQA, TextVQA, MME, AI2D, MMBench, CCBench, MMVet, and SEED-Image were tested using the InternVL repository. MMMU, OCRBench, RealWorldQA, HallBench, and MathVista were evaluated using the VLMEvalKit.
49
 
 
263
 
264
  ## 性能测试
265
 
266
+ | 评测数据集 | GPT-4T-20240409 | Gemini-1.5-Pro | InternVL-Chat-V1-5 | InternVL2-26B |
267
+ | :--------------------------: | :-------------: | :------------: | :----------------: | :-----------: |
268
+ | 模型大小 | - | - | 25.5B | 25.5B |
269
+ | | | | | |
270
+ | DocVQA<sub>test</sub> | 87.2 | 86.5 | 90.9 | 92.9 |
271
+ | ChartQA<sub>test</sub> | 78.1 | 81.3 | 83.8 | 84.9 |
272
+ | InfoVQA<sub>test</sub> | - | 72.7 | 72.5 | 75.9 |
273
+ | TextVQA<sub>val</sub> | - | 73.5 | 80.6 | 82.3 |
274
+ | OCRBench | 678 | 754 | 724 | 825 |
275
+ | MME<sub>sum</sub> | 2070.2 | 2110.6 | 2187.8 | 2260.7 |
276
+ | RealWorldQA | 68.0 | 67.5 | 66.0 | 68.3 |
277
+ | AI2D<sub>test</sub> | 89.4 | 80.3 | 80.7 | 84.5 |
278
+ | MMMU<sub>val</sub> | 63.1 | 58.5 | 45.2 | 48.3 |
279
+ | MMBench-EN<sub>test</sub> | 81.0 | 73.9 | 82.2 | 83.4 |
280
+ | MMBench-CN<sub>test</sub> | 80.2 | 73.8 | 82.0 | 82.0 |
281
+ | CCBench<sub>dev</sub> | 57.3 | 28.4 | 69.8 | 73.5 |
282
+ | MMVet<sub>GPT-4-0613</sub> | - | - | 62.8 | 64.2 |
283
+ | MMVet<sub>GPT-4-Turbo</sub> | 67.5 | 64.0 | 55.4 | 62.1 |
284
+ | SEED-Image | - | - | 76.0 | 76.8 |
285
+ | HallBench<sub>avg</sub> | 43.9 | 45.6 | 49.3 | 50.7 |
286
+ | MathVista<sub>testmini</sub> | 58.1 | 57.7 | 53.5 | 59.4 |
287
 
288
  - 我们同时使用 InternVL 和 VLMEvalKit 仓库进行模型评估。具体来说，DocVQA、ChartQA、InfoVQA、TextVQA、MME、AI2D、MMBench、CCBench、MMVet 和 SEED-Image 的结果是使用 InternVL 仓库测试的。MMMU、OCRBench、RealWorldQA、HallBench 和 MathVista 是使用 VLMEvalKit 进行评估的。
289