llm-foundry

85403c08 - Bumping flash attention version to 2.6.3 and adding option for softcap in attention and lm_head logits. (#1374)