minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

vlm: support piecewise cuda graph for Kimi-K2.5 (#20747)

Browse Source
This commit is contained in:
Yuhao Yang
2026-03-18 00:32:07 +08:00
committed by GitHub
parent 7f99319c56
commit 24a27d5320
1 changed files with 2 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/models/kimi_k25.py
+2
View File
@@ -716,6 +716,8 @@ class KimiK25ForConditionalGeneration(nn.Module):
), ),
) )
self.model = self.language_model.model
# Ensure that the dtype of the vision_tower and mm_projector matches that of the language_model. # Ensure that the dtype of the vision_tower and mm_projector matches that of the language_model.
# This solves the dtype mismatch issue when using device_map="auto" and torch_dtype. # This solves the dtype mismatch issue when using device_map="auto" and torch_dtype.
if hasattr(self.language_model, "dtype"): if hasattr(self.language_model, "dtype"):