Starry
Starry
我把lr_gen和lr_dis设成1e-5,batch_size设置成1024,在epoch 30的时候gen的accuracy达到最大0.884,比论文中的accuracy高一些
I have the same question
同样遇到这种问题,在obsidian设置了附件默认存放在指定文件夹和附件目录,链接![[Pasted image 20230517171742.png]]可以正常上传,但是上传后文章中的链接还是原来的样子,没有变。图片链接的形式使用的是wiki链接
同求能上传并替换文件夹下所有笔记的功能!!!
I found that vision tower has no gradient in llava's code, but they are set to have gradient during training. I don't know which priority is higher, and I haven't...
## 转换后评估结果: ```bash [2025-04-11 15:35:53,838] [ INFO] - ----------------------------- [2025-04-11 15:35:53,838] [ INFO] - Evaluation Precision: 0.00000 | Recall: 0.00000 | F1: 0.00000 ``` ## 转换onnx前评估代码: ```python # Copyright (c)...
Hi! @DrownFish19 这是因为macOS的原因吗,我目前还没在其他平台测试过这段代码
哈喽 @DrownFish19 ,我们换了一个平台测试了这个代码结果依然不行,模型输出数据的格式是正确的,但是不知道为何评估结果是0,请问有遇到过类似的问题吗?
same question
@YushunXiang Hi! Thank you for your reply! Could I use the config file of the VLMEvalKit when I use the lmdeploy or vllm? How to use the config in these...