Arthur comments

Results 795 comments of


                                            Arthur

[`BERT`] Add support for sdpa

Of course!

[`BERT`] Add support for sdpa

Hey! Sure I was off for a bit but will have a look

[`BERT`] Add support for sdpa

Oh wow

torch.arange use should not use dtype=float for integer ranges, conflicts w/ DS `zero.Init()`

@gante from my personal test, changing the inv_freq to float32 can increase performances on MMLU of about 20points. There are a few things to test: - `inv_freq` is a buffer,...

torch.arange use should not use dtype=float for integer ranges, conflicts w/ DS `zero.Init()`

super nice 🤗

torch.arange use should not use dtype=float for integer ranges, conflicts w/ DS `zero.Init()`

Also linked to #29285

torch.arange use should not use dtype=float for integer ranges, conflicts w/ DS `zero.Init()`

I think this was fixed! For llama at least and gemma 🤗

[`TokenizationUtils`] add support for `split_special_tokens`

Wip, not urgent! Can already be done but it's not save / loaded

[WIP] VMamba implementation

I don't have bandwidth yet so nice if you want to do ti!

Huggingface Agents Error 422: {'error': 'Input validation error: `max_new_tokens` must be <= 192

Hey! Thanks, you are probably right. Would you like to open a PR to change this and make it more friendly?