apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2026-04-23 17:11:21 +08:00
Code Issues Actions 6 Packages Projects Releases Wiki Activity

[Optimization][DeepSeekV3.2]Reducing slot_mapping compute frequency from twice per layer to a single pre-processing step. (#7367)

Browse Source
This commit is contained in:
ShaneGZhu
2026-04-16 19:54:12 +08:00
committed by GitHub
parent d2d633b05c
commit 2d8338f9e4
10 changed files with 73 additions and 146 deletions
Download Patch File Download Diff File Expand all files Collapse all files
fastdeploy/model_executor/forward_meta.py
+2 -1
View File
@@ -160,7 +160,8 @@ class ForwardMeta:
# for mla & dsa
position_ids: Optional[paddle.Tensor] = None
mask_encoder_batch: Optional[paddle.Tensor] = None
# for kvcache slot
slot_mapping: Optional[paddle.Tensor] = None
real_bsz: int = 0