InternVL
InternVL copied to clipboard
[CVPR 2024 Oral] InternVL Family: A Pioneering Open-Source Alternative to GPT-4o. 接近GPT-4o表现的开源多模态对话模型
### Motivation For the evaluation, could you please add the MMIU benchmark(https://mmiu-bench.github.io/) in https://internvl.readthedocs.io/en/latest/internvl2.0/evaluation.html? ### Related resources _No response_ ### Additional context _No response_
### Checklist - [x] 1. I have searched related issues but cannot get the expected help. - [ ] 2. The bug has not been fixed in the latest version....
### Checklist - [X] 1. I have searched related issues but cannot get the expected help. - [ ] 2. The bug has not been fixed in the latest version....
### Motivation 选择模型那里是灰色的,不能选择,是什么原因呢?麻烦请教一下怎么弄。模型用的是InternVL2-Llama3-76B。路径如下图: ### Related resources 请问下图中是不是路径不对?/root/InternVL/InternVL2-Llama3-76B 里面是 OpenGVLab/InternVL2-Llama3-76B 模型文件和其他文件。 ### Additional context  
### Checklist - [X] 1. I have searched related issues but cannot get the expected help. - [X] 2. The bug has not been fixed in the latest version. -...
### Checklist - [x] 1. I have searched related issues but cannot get the expected help. - [x] 2. The bug has not been fixed in the latest version. -...
### Checklist - [X] 1. I have searched related issues but cannot get the expected help. - [X] 2. The bug has not been fixed in the latest version. -...
InternVL-1.1 中的结果  InternVL-1.2 中的结果  可以看到InternVL-1.2 相比于InternVL-1.1 在SQA,TextVQA等好几个benchmark 上都要低,这是为什么?理论上1.2 版本应该比1.1 版本要高。
### Checklist - [X] 1. I have searched related issues but cannot get the expected help. - [ ] 2. The bug has not been fixed in the latest version....
At the moment the internvl_chat/internvl/train/internvl_chat_finetune.py and internvl_chat/internvl/train/internvl_chat_pretrain.py are almost the same. I am wondering if the ITM loss and Contrastive loss would be used. Am I miss something? Is generative...