Geary.Z
Geary.Z
> In case it might be helpful, I encounted this problem as found the solution is quite simple🥲 . > > In fact the plots/curves don't disappear. Some new items...
Same bug raised by using `gen(choices=[...])` with image input, as mentioned in #222 .
Same issue. Nothing happens when calling Synctex From Cursor in VS Code.
我们的实现方法加入了单独的image token标记图片位置，这在原模型中是没有的。您需要在模型的`config.py`文件中添加`"image_token_index":92544`
我们统一用``作为image token，请检查下输入文本中image token的总数量与输入图片的总数量是否一致
> 方便展示一下internXC2模型的config.py文件吗 可以参考我们放出的checkpoint: https://huggingface.co/TideDra/internlm-xcomposer2-vl-7b-DPO/blob/main/config.json
Thanks for your review. Let me know if there is anything I can do.
已修复，感谢您的反馈。
> 作者您好，除了在benchmark上测试之外，或许有推理代码可供参考吗？即对于预训练的ckpt或微调之后的ckpt，进行简单的单样本推理。或者我应该参考哪个仓库/模型的代码？ 很棒的工作，感谢~ 您可以使用`src/vlrlhf/eval.utils.py`中提供的相关接口: ```python from vlrlhf.eval.utils import load_model_and_processor model,processor,generation_kwargs = load_model_and_processor(YourModelPath,None) image_path = 'a.jpg' prompt = 'Describe this image' prompt = processor.format_multimodal_prompt(prompt,image_path) inputs = processor(texts=[prompt], images_path=[image_path], check_format=False) inputs.pop('label',None) outputs...
您好，未来我们会考虑添加新的模型。您也可以参考`docs/CustomizedModel.md`添加自定义模型，欢迎您的PR!