Lap-Hang Ho comments

Results 7 comments of


                                            Lap-Hang Ho

[Whisper] Word level and character level timestamps

NB: word level timestamps were added to openai/whisper last week. Tried it out, it seems to work. https://github.com/openai/whisper/commit/500d0fe9668fae5fe2af2b6a3c4950f8a29aa145

toxigen task measures toxicity classification rather than whether generations are toxic?

Thanks for the response, I'll keen an eye on the big-refactor getting merged into main!

toxigen task measures toxicity classification rather than whether generations are toxic?

Hi @danihinjos, no I didn't end up implementing this unfortunately. I did try the real toxicity task, but found the perspective api too slow. As an aside, I did gain...

need to be able to set return data size for each request in batch transform

Had the same issue for realtime endpoints, passing in TS_MAX_RESPONSE_SIZE in the env variable solved this as well.

need to be able to set return data size for each request in batch transform

@levatas-weston yes to the PyTorchModel, see the base class docs below. https://sagemaker.readthedocs.io/en/stable/api/inference/model.html#sagemaker.model.Model

How to QLoRA training with ZeRO-3 on two or more GPUs?

I'm getting this issue as well (trying qlora with ZeRO-3 and 4 gpus, same error message), @Di-Zayn were you able to solve it?

How to QLoRA training with ZeRO-3 on two or more GPUs?

I was keen on sharding the model across gpus in order to be able to allow for larger models. As an aside, the latest FSDP and qlora examples are working...