Showing 1 of 1 repositories tagged #model-parallization, ranked by stars
Train llm (bloom, llama, baichuan2-7b, chatglm3-6b) with deepspeed pipeline mode. Faster than zero/zero++/fsdp.
