Cheng-Ping Hsieh comments

Repositories
Issues
Comments

Results 3 comments of


                                            Cheng-Ping Hsieh

Davidm/gpt sft dataset fix

Hi @Davood-M, if your entire input length (template + placeholder strings) > sequence length, then we'll need to do the truncation. We truncate only the `truncation_field` part in your placeholder...

Davidm/gpt sft dataset fix

Hi @aditya-malte, in your setup what values are your `truncation_field` and `prompt_template`?

A bug when run run_chunkllama_100k with flash decoding

I think the error is because we should install `flash_attn >=2.5.3,