luoan7248

Results 12 issues of luoan7248

Mostly the pictures in the fine-tuned corpus have a clear focus on something, such as a character, a puppy, etc. But the picture content is more mixed cases, such as...

How do I specify a graphics card for inference? Setting os.environ["CUDA_VISIBLE_DEVICES"] = '6' didn't work.

**Describe the bug** Looking in indexes: https://pypi.tuna.tsinghua.edu.cn/simple Collecting auto-gptq Using cached https://pypi.tuna.tsinghua.edu.cn/packages/90/e5/b22697903982284fe284568fb2663a2196694a8eee637f5cf4ccfe435a38/auto_gptq-0.7.1.tar.gz (126 kB) Preparing metadata (setup.py) ... error error: subprocess-exited-with-error × python setup.py egg_info did not run successfully. │...

bug

![image](https://github.com/user-attachments/assets/6bfb214a-a954-48ba-829c-0797e07005ad) 如上图,使用官方的案例进行推理,当提示词使用中文时,推理并没有报错,但是没有生成图片。 英文提示词推理结果: ![image](https://github.com/user-attachments/assets/5b902976-2cee-44fe-9eaa-4ab027415b5c) 使用中文推理,自定义的图片: ![image](https://github.com/user-attachments/assets/eda63361-8a45-4bed-bf8a-a85390fa17df) 请求作者大大帮忙一下

How to select seed? Randomly selected? Sincerely requesting the author's help!

直接使用lora微调的权重后推理,原本的大模型能力好像没有了,我不知道是不是微调的效果不好,还是权重没有合并的原因。

![image](https://github.com/user-attachments/assets/8f8723b4-7d93-4ee8-a556-4846bedc66a5) 请问哪一个是最佳的权重,如何通过查看loss值或者图查看查看训练的效果好不好呢?

Are there any special conventions for setting image size parameters for fine-tuning and reasoning. When lora is trained, the height setting of 1024 and the width setting of 560 report...

在设置微调和推理的图像尺寸参数时,是否有什么特殊的约定。 在训练 lora 时,高度设置为 1024、宽度设置为 560 会报错,推理也是这样设置的,也会报错。 但如果同时设置 1024,就不会出错。 具体信息如下 Epoch 0: 0%| | 0/250 [00:00