Spaces:
Running
'list' object has no attribute 'data_ptr'
This is for TWO loRa that i've trained on HF spaces rather than via notebooks. Trying to convery pytorch bin files to checkpoints
Is your issue still there ?
Could you share something reproducible ?
It's mostly likely something is not sending the weights properly to the save function, but it's hard to know if it's the script, the oriignal torch weights or something else that's failing.
It's a lora trained with the HF lora dreambooth so sadly i can't
I can't get it to convert with the converter space either cause it's a bin file XD
The converter works by iteratively looking through bin files and converting them in the repo.
Could you point me to steps to reproduce maybe ?
I'm gonna see if it'll do ti again later :#
Yea i just add the repo and HF token and this is what happens
Happens on both the 2.1 and 1.5 versions i have
Yea i just add the repo and HF token and this is what happens
@Duskfallcrew you might want to refresh your token to prevent misuse (if it's a write token or if you have private repos you don't want to share), if you haven't already. Go to https://hf.co/settings/tokens, click manage on the token that's clearly visible in this image, then click invalidate and refresh.
@Narsil I recommend using password input for the API key. As a bonus, if someone comes back to the space later they can have the browser save the token.
AH oops lol, yea i have like 10 tokens in range, so yea i'll do that LOL
Hi @Duskfallcrew ,
The error is that the convertion script is trying to look for data_ptr in order to save the tensor within the model weights. However, it seems here something was saved as a raw list instead of a tensor.
It seems the error comes from the script trying to convert scheduler.bin. Currently there's no drop-in replacement for that since safetensors expects tensors, not arbitrary python data. I'll fix it so it skips those faulty files for now and keep converting the actual useful core model.
Ok i'll test it soon!
I made the update, however I can test right now lots of requests coming in (most likely coming from the API itself which autoconverts models for its own use)
Ok conversion went sucessfully : https://huggingface.co/Duskfallcrew/the-crystal-exarch-15/discussions/2
Oh ok sweet i'll put it in for testing on stream right now!