Closed HenryHZY closed 2 months ago
As mentioned in video_chat2/models/blip2/modeling_llama_mem.py:
from flash_attn.flash_attn_interface import flash_attn_varlen_qkvpacked_func # flash_attnv2
Awesome!
As mentioned in video_chat2/models/blip2/modeling_llama_mem.py: