Cunxiao Du comments

Results 13 comments of


                                            Cunxiao Du

Error while training with translation_multi_simple_epoch using custom datasets

this bug usually caused by you use an old version preprocess.py to process the data then use the 0.10 version to train

fold not loading after installation

got the same error...

StopIteration error?

facing the same issue

what are the train/val/test split of JDDC in your experiment

> Hi, I cannot find the JDDC corpus, can you provide this corpus? THX! @NonvolatileMemory just email the authors

what are the train/val/test split of JDDC in your experiment

> Okay, yesterday I have asked the author. they give you the data split?

Fused Linear and Cross-Entropy Loss `torch.nn.functional.linear_cross_entropy`

I have a question, do we need to compute the forward pass result? I think we can directly fuse all the things with a backward kernel during training, and thus,...

Fused Linear and Cross-Entropy Loss `torch.nn.functional.linear_cross_entropy`

Great! > Yeah, both Malek's version and mine (both linked above), directly compute gradients during the forward pass loop over N and discard anything else. During the backward pass, nothing...

Loss is so high when using SIGMOID_CROSS_ENTROPY_LOSS or EUCLIDEAN_LOSS

I think maybe it's your video data layers problem，can you show your xx.lst and data layer?

OLMoThreadError: generator thread data thread 0 failed

same issue here

ring attention实现原理

大佬能解析下为什么backward只需要修正lse就可以了嘛