ttthhh

api调用失败

1

你好。目前api是不能用了么，返回的是2004的错误码

why cuda.amp was removed from version 3.35.0

2

### System Info transformers==3.35.0 and transformers==3.34.0 ### Who can help? _No response_ ### Information - [ ] The official example scripts - [ ] My own modified scripts ### Tasks...

Failed to install pytorch3d when using Google Colab

4

 Any solutions?

How many samples are in the dataset?

2

In the paper, it seems 2000 for perception and 800 for cognition. But only 2374 samples are in the .tsv file.

**Describe the bug** What the bug is, and how to reproduce, better with screenshots(描述bug以及复现过程，最好有截图) 当设置 model.generation_config.num_return_sequences = 2 （默认是1，只要设置大于1的数，inference还是只返回一个结果） 查看代码后，发现是templete.get_generate_ids中做了限制，默认只取第一个sequence的结果  **Your hardware and system info** Write your system info...

我看文章里说在训练之前，是用mclip来过滤数据的，clip score低于0.26的都过滤掉。这个clip score就是计算的图像emb和文本emb的余弦相似度吗。我测试了一下，发现无论图文多匹配，能超过0.26的都不多。