[NPU]support model MiniCPM3-4B for npu (#16866)

This commit is contained in:
McZyWu
2026-01-24 08:25:12 +08:00
committed by GitHub
parent 4c7136bb36
commit 8a5ed2434f
5 changed files with 125 additions and 45 deletions

View File

@@ -756,8 +756,8 @@ class Phi3LongRoPEScaledRotaryEmbedding(nn.Module):
key: torch.Tensor,
offsets: Optional[torch.Tensor] = None,
) -> Tuple[torch.Tensor, torch.Tensor]:
query = query.view(*query.shape[:-1], -1, self.head_size)
key = key.view(*key.shape[:-1], -1, self.head_size)
query = query.unflatten(1, (-1, self.head_size))
key = key.unflatten(1, (-1, self.head_size))
k = self.original_max_position_embeddings
long_prompt_offset = (