EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

Merge branch 'v0.5.4_dev_qwennext' into 'v0.5.4_dev'

Browse Source 
适配qwen3-next

See merge request OpenDAS/sglang!4
This commit is contained in:
maxiao1
2025-11-03 05:18:16 +00:00
parent 0fbecc4364 477fddf28d
commit 47e4d92348
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
python/sglang/srt/models/qwen3_next.py
View File

@@ -396,7 +396,7 @@ class Qwen3GatedDeltaNet(nn.Module):
def _forward_input_proj(self, hidden_states: torch.Tensor): def _forward_input_proj(self, hidden_states: torch.Tensor):
DUAL_STREAM_TOKEN_THRESHOLD = 1024 if not _is_npu else 0 DUAL_STREAM_TOKEN_THRESHOLD = 1024 if not _is_npu else 0
seq_len, _ = hidden_states.shape seq_len, _ = hidden_states.shape
if seq_len < DUAL_STREAM_TOKEN_THRESHOLD: if seq_len < DUAL_STREAM_TOKEN_THRESHOLD and self.alt_stream is not None:
current_stream = torch.cuda.current_stream() current_stream = torch.cuda.current_stream()
self.alt_stream.wait_stream(current_stream) self.alt_stream.wait_stream(current_stream)
projected_states_qkvz, _ = self.in_proj_qkvz(hidden_states) projected_states_qkvz, _ = self.in_proj_qkvz(hidden_states)