num_key_value_heads
#1
by
penut85420
- opened
The num_key_value_heads for Sheared-LLaMA-2.7B is 20, while for Sheared-LLaMA-2.7B-ShareGPT it is 32, which makes the model unusable.
The num_key_value_heads for Sheared-LLaMA-2.7B is 20, while for Sheared-LLaMA-2.7B-ShareGPT it is 32, which makes the model unusable.