1 result found Sort:

CoinCheung
gdGPT CoinCheung
8
90
apache-2.0
1
Train llm (bloom, llama, baichuan2-7b, chatglm3-6b) with deepspeed pipeline mode. Faster than zero/zero++/fsdp.
llm nlp bloom llama2 pytorch pipeline deepspeed chatglm3-6b baichuan2-7b mixtral-8x7b full-finetune flash-attention model-parallization
Created 2023-06-24
27 commits to master branch, last one 9 months ago