NeMo Framework已正式釋出, 歡迎到官網申請並下載!
NeMo Framework支援跨GPU和跨Node的LLM訓練, 訓練手法包含foundation model training, fine-tuning, instruction tuning (SFT, RLHF)和customization (prompt tuning, P-tuning, Adapter, AliBi, PEFT with IA3). NeMo Framework目前支援的模型類型包含GPT, Llama 2, Falcon, BLOOM等. 詳細介紹請參考原文.