• Released Vicuna v1.5 based on Llama 2 with 4K and 16K context lengths. Download weights
• Released Chatbot Arena Conversations, a dataset containing 33k conversations with human preferences. Download it here.
• Serving
  • Add a multi-model worker that can host multiple models on a single GPU and share base weights for PEFT models. #1866 #1905
  • AWQ 4-bit quantization support. #2103
  • Support model models (Llama 2, Claude 2, ChatGLM 2, StarChat, Baichuan-13B, InternLM, airoboros, PEFT adapters).
  • Better support for AMD GPUs, Intel XPUs. #1954 #2052
• Training
  • Support rope scaling. #2013
  • Support flash attention 2. #2059
  • Support xformer. #1970