HuangTao

icon location https://huangtao36.github.io/

Results 16 comments of HuangTao

SomeQuestion

ooh, thanks, this is a great tip, I will try it. And I will try this model on the person images (contain the body), hoping to have a good effect.

Supplementary figure 3

Also looking forward to

Question about figure 2.b

I got the same result as you

RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cuda:1!

把模型和数据都扔到cuda: 0

how to get the 'txt_feat, txt_feat_mismatch, txt_feat_relevant'?

Thank you for your reply, I am still confused about this. In a batch, `np.roll() ` is just shuffle the caption of each picture, like `txt_feat_mismatch = torch.Tensor(np.roll(txt_feat_np, 1, axis=0))`,...

How to prepare 2023-02-12_oasst_prod.jsonl

Are there any open datasets where I can try to train it

关于如何做微调的一些疑问

我感觉它主要是测试的代码，不需要生成mask，就没加，训练应该还是需要的吧。官方能解答不

推理速度有点慢，有什么好方法加速吗

好奇chatGPT为啥可以这么快呢，有啥技术要点不

训练速度

> 写错了，是A100，这可真慢啊，我发现增加batchsize对于加速一点用没有

训练速度

> 对啊，总共就是18w数据跑2个epoch，batch_size等于1的时候时间也一样呢