search

CambioML / pykoi-rlhf-finetuned-transformers

pykoi: Active learning in one unified interface
https://www.cambioml.com
Apache License 2.0
407 stars 43 forks source link

support sft training on d2l #100

Closed llauraa23 closed 9 months ago

llauraa23 commented 9 months ago

out of memory on 24G single gpu during training

CambioML commented 9 months ago

@llauraa23 Take a look at https://github.com/huggingface/trl/issues/1083#issuecomment-1861227236 and https://huggingface.co/docs/trl/sft_trainer#advanced-usage

llauraa23 commented 9 months ago

@llauraa23 Take a look at huggingface/trl#1083 (comment) and https://huggingface.co/docs/trl/sft_trainer#advanced-usage

Thanks! I will have a look over the weekend. Have to work on the dpo and post writing.

CambioML commented 9 months ago

close this PR because it is also in https://github.com/CambioML/pykoi/pull/101. The future work is to use https://github.com/CambioML/pykoi/pull/101 into two PR for SFT and DPO.