B. Shen
B. Shen
Strongly recommend this feature. The server that I'm using suffers from limited space of /home (which is the default model path). I want to move the model to another disk,...
Sorry for the late response. Would you please randomly init a Llama MHA model with `hidden_size=1536`, `num_attention_heads=32` (-> `head_dim=48`)? I tried to set `num_attention_heads=24` (-> `head_dim=64`) and it works fine....
It seems that run_qa_prune.py is not updated after the argument names are changed. Such as https://github.com/princeton-nlp/CoFiPruning/blob/793e3e1291827e2714b5de6d5c0b6b04bc1863e4/run_qa_prune.py#L139 `str_pruning_method` should be `pruning_type` https://github.com/princeton-nlp/CoFiPruning/blob/793e3e1291827e2714b5de6d5c0b6b04bc1863e4/run_glue_prune.py#L226