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

fix: specify dtype with begin_forward aka plan (#2404)

Browse Source
This commit is contained in:
Yineng Zhang
2024-12-08 20:07:30 +08:00
committed by GitHub
parent a2486eb58f
commit 6128f7cff5
1 changed files with 2 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
python/sglang/srt/layers/attention/flashinfer_backend.py
View File

@@ -678,6 +678,7 @@ class FlashInferIndicesUpdaterPrefill:
self.num_qo_heads, self.num_qo_heads,
self.num_kv_heads, self.num_kv_heads,
self.head_dim, self.head_dim,
q_data_type=self.q_data_type,
) )
# cached part # cached part
@@ -691,6 +692,7 @@ class FlashInferIndicesUpdaterPrefill:
self.num_kv_heads, self.num_kv_heads,
self.head_dim, self.head_dim,
1, 1,
q_data_type=self.q_data_type,
) )