Sylvain Gugger comments

Results 633 comments of


                                            Sylvain Gugger

NER Training on custom Data

Please use the [forums](https://discuss.huggingface.co/) to discuss questions like this as we keep the issues for bugs and feature requests only.

[Feature Request] Support Megatron-Deepspeed

cc @pacman100

Gradient accumulation trick and Activation Checkpointing feature

cc @sanchit-gandhi

Using `auto_map` in `tokenizer_config.json` gives `TypeError: argument of type 'NoneType' is not iterable`

Thanks for flagging! The PR linked above should fix this.

Performance Regression from commit 7dcd870

cc @gante and @ArthurZucker

RecursionError: maximum recursion depth exceeded while calling a Python object

It's going to be hard to help you without any reproducer.

fix AutoTP in deepspeed could not work for bloom

Actually, just checked the modeling file and this function is only used in this class, so it would be cleaner to just make it a method. Could you update your...

fix AutoTP in deepspeed could not work for bloom

Ok so let's keep it as a function in that module. I'd still prefer a real method (that directly returns the result of the function) to setting a function attribute...

Contradictory information in documentation about the ability to push qunatized models to hub

cc @younesbelkada

Calling parallelize() on T5ForConditionalGeneration for ByT5 results in device_map error

Note that the `parallelize` API is going to be deprecated soon. You should load your model like this to use Accelerate instead: ```python model = T5ForConditionalGeneration.from_pretrained("google/byt5-xl", device_map="balanced") ```