add_trained_weights
#1
by
leejunhyeok - opened
No description provided.
leejunhyeok changed pull request status to
open
@leejunhyeok
It might be better to update the MotifAttention class docstring to mention Grouped Differential Attention instead of Differential Attention.
@leejunhyeok we might need to remove "_attn_implementation": "flash_attention_2" option as described in config.json of Motif-2-12.7-Base.
SungminLee changed pull request status to
merged