Jakob Salomonsson comments

Results 6 comments of


                                            Jakob Salomonsson

ValueError: Query/Key/Value should either all have the same dtype,

[This](https://github.com/huggingface/accelerate/issues/2182) and [this](https://github.com/huggingface/diffusers/issues/6812) are likely related issues.

[BUG] Invalidate trace cache @ step 1: expected module 25, but got module 323,how to resolve it ?

Hi @tohtana, I get a similar error when using DeepSpeed via Hugging Face Accelerate to train SDXL. It happens during evaluation after the first epoch where the training simply freezes:...

ValueError: Query/Key/Value should either all have the same dtype training LORA

Do you happen to have a time frame for when you can look into that @sayakpaul?

Conflicting task names when using prefect-gcp and prefect-aws simultaneously

I get this error despite specifying unique names for each task using prefect `2.10.9`. I run everything locally and don't have `prefect-gcp` nor `prefect-aws` installed.

"AttributeError: 'NoneType' object has no attribute 'append'" using 4 or more threads. Fewer threads works fine.

I think there was a mistake in my code that caused the issue and not with Arctic. It's solved now. My apologies. Closing.

"AttributeError: 'NoneType' object has no attribute 'append'" using 4 or more threads. Fewer threads works fine.

After feeding a larger chunk of data (roughly 10x more than before) the problem as appeared again. Reopening in the hope that a solution can be found.