AssertionError: Padding_idx must be within num_embeddings
anyone else got this error?
happens in this line:
model = AutoModelForVision2Seq.from_pretrained(
"HuggingFaceM4/Idefics3-8B-Llama3",
torch_dtype=torch.bfloat16,
).to(DEVICE)
i have the updated transformers using the PR.
Hi @ZeevRispler , merve had this issue yesterday and managed to fix it by updating the cache on transformers. Can you try that?
Hey @andito Thanks fo the resopnse! can you maybe send a link? i tried deleiting the cache but it didn't help.
A link to what?
If it's the issue from merve, we talked about it on slack, sorry, no link :sweat_smile:
Let me check this again on a clean repo
I hate to say this but it works on my computer :( Are you following a tutorial or what are you doing exactly?
Same problem here...straight after downloading the model from the repo.
In my case all I did was running the line of code in the model card
model = AutoModelForVision2Seq.from_pretrained(
"HuggingFaceM4/Idefics3-8B-Llama3",
).to(DEVICE)