Closed gaotianyu1350 closed 6 days ago
Great work! There is a typo in the README
enable_duo_attention_eval( model, attn_heads, num_recent_tokens=64, num_sink_tokens=256, )
num_recent_tokens -> sink_size, num_sink_tokens -> recent_size
num_recent_tokens
num_sink_tokens
Fixed. Thank you!
Great work! There is a typo in the README
num_recent_tokens
-> sink_size,num_sink_tokens
-> recent_size