Support qwen3 deepep (#6120)

This commit is contained in:
lukec
2025-05-22 11:04:45 -07:00
committed by GitHub
parent d71f3f0a2a
commit fc0e3b9174
2 changed files with 125 additions and 8 deletions
+4 -1
View File
@@ -607,7 +607,10 @@ class Qwen2MoeModel(nn.Module):
)
else:
if hidden_states.shape[0] != 0:
hidden_states, _ = self.norm(hidden_states, residual)
if residual is None:
hidden_states = self.norm(hidden_states)
else:
hidden_states, _ = self.norm(hidden_states, residual)
return hidden_states