Umar Butler
Umar Butler
@debonte I'll be building a new PC within the next week or two so it makes more sense for me to use the older version, however, if the issue appears...
> Currently 12.3 is supported. This version can already support Flash Attention 2. You can give this version a try. There is no obvious difference between 12.4 and 12.3. The...
@charliermarsh Any movement on this? I agree with @robvdl that this is a bug. Having to do `__all__` goes against DRY, is unintuitive, and is cumbersome. @robvdl's fix unfortunately doesn't...
> Ruff doesn't support multifile analysis yet but we're working towards bringing this capability to Ruff. But it will take a while and we then also have to change this...
Personally, now that I have my workaround have started adding to all of my project's ruff configs, I'm not as annoyed by this, however, for a new user, there is...
+1 I'm getting lots and lots of these errors.
In their model README, they don't load their model with `torch_dtype = torch.bfloat16`. Presumably, that would cause it to be loaded in full bfloat16? Their paper doesn't mention that the...
I just checked and it does seem like the `pipeline` would be loading ModernBERT in full bfloat16, which is inconsistent with the previous example 😆 Now, its possible they're advising...
Hey @BBC-Esq, Sorry, it's taken me a while to get to this. Proposals 2 and 3 are a no-go because there are many good reasons why someone would want to...