InternVL

InternVL copied to clipboard

bobo0810

### Motivation Hi Team, Thanks for the great effort and open-sourcing the MLLM and report. It was a great read to understand. One key question I wanted to ask was...

amitbcp

### Motivation Deepspeed provide out of box tensor parallelism. However, when I modify config, for example, internvl_chat/zero_stage3_config.json adding "model_parallelism" parameters to fine 26B model: "model_parallel": { "enabled": true, "dp_world_size": 6,...

MrPanch

[Bug] VL2.5-8B有bug

3

### Checklist - [ ] 1. I have searched related issues but cannot get the expected help. - [ ] 2. The bug has not been fixed in the latest...

whysirier

微调internvl-chat loral，loss从2.6594下降到0.0001，但是模型输出结果为roproproproproproproproproproproproproproproproprop ...

1

在使用自己准备的训练数据微调模型，训练数据包括5w张商品图片和对应的prompt文本数据，迭代了2396个epoch，但是用微调好的模型推理时模型输出为：roproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproproprop 训练数据中并没有roprop...这种字符，图中是我的loss曲线，求助各位大佬这是什么情况？  也猜想过是否发生了过拟合，但是如果是过拟合那么输出也不应该这样的。。。

limin-al

### Motivation 作者您好，请问近期有计划支持GRPO训练吗，期待~ ### Related resources _No response_ ### Additional context _No response_

Wangman1

### Motivation I would like to suggest expanding the evaluation of visual reasoning to the **HumanEval-V** benchmark. This benchmark provides a more challenging set of tasks by introducing **complex diagrams**...

zfj1998

[Docs] 路径问题

3

### 📚 The doc issue 您好！在对internVL2.5-4b模型进行微调时，我注意到在下载hugging face上面官方给的权重后，还需要在pretained里面下载哪些问题呢，以及这些文件放置的位置，我下载了internvl和internvl_chat，在运行的时候会报错：[INFO|tokenization_auto.py:606] 2025-02-21 18:50:59,802 >> Could not locate the tokenizer configuration file, will try to use the model config instead. ### Suggest a potential alternative/fix...

Celina-love-sweet

[Docs]

1

### 📚 The doc issue 如何解决 ### Suggest a potential alternative/fix 使用internvl2-4b进行coco字幕测评时，发生如下警告。You are using a model of type internvl_chat to instantiate a model of type internvl. This is not supported...

liluoqaq

In the InternVL2.5-MPO paper, the author mentioned that cot is not as effective as direct answer for MLLM. I wonder why cot is so bad for MLLM compared to LLM?...

EchoDreamer