radu/LLamaRecipes

spogulis no https://github.com/facebookresearch/llama-recipes.git

Autors	SHA1 Ziņojums	Datums
Matthias Reso	9def4fbafd Remove micro_batch_training parameter and replace with gradient_accumulation_steps	2 gadi atpakaļ
Brian Vaughan	3faf005226 fix a bug in the config for use_fast_kernels	2 gadi atpakaļ
lchu	feaa344af3 resolve conflicts	2 gadi atpakaļ
Hamid Shojanazeri	44ef280d31 adding flash attention and xformer memory efficient through PT SDPA	2 gadi atpakaļ
lchu	895dfcea30 add nightly check for using low_cpu_fsdp mode	2 gadi atpakaļ
chauhang	4767f09ecd Initial commit	2 gadi atpakaļ