修改了由于minicpm的GQA结构带来的模型转换bug

This commit is contained in:
刘丹 2024-07-02 15:43:07 +08:00
parent 023b8807e1
commit 47d821a08c

View file

@ -1605,7 +1605,7 @@ class MiniCPMModel(Model):
def _reverse_hf_permute(self, weights: Tensor, n_head: int, n_kv_head: int | None = None) -> Tensor:
if n_kv_head is not None and n_head != n_kv_head:
n_head //= n_kv_head
n_head = n_kv_head
return (
weights.reshape(n_head, 2, weights.shape[0] // n_head // 2, *weights.shape[1:])