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

fix(xpu): read enable_deterministic_inference from the config bag (#36149)

Browse Source
This commit is contained in:
Shangming Cai
2026-08-24 13:56:41 +08:00
committed by GitHub
parent ec334b17e2
commit 852b04b358
1 changed files with 2 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/layers/attention/xpu_backend.py
+2 -1
View File
@@ -16,6 +16,7 @@ from sglang.srt.mem_cache.memory_pool import KVWriteLoc
from sglang.srt.mem_cache.swa_memory_pool import SWAKVPool from sglang.srt.mem_cache.swa_memory_pool import SWAKVPool
from sglang.srt.model_executor.forward_batch_info import ForwardBatch, ForwardMode from sglang.srt.model_executor.forward_batch_info import ForwardBatch, ForwardMode
from sglang.srt.runtime_context import ( from sglang.srt.runtime_context import (
get_exec,
get_schedule, get_schedule,
get_spec, get_spec,
) )
@@ -106,7 +107,7 @@ class XPUAttentionBackend(AttentionBackend):
# when deterministic inference is enabled to keep attention reduction # when deterministic inference is enabled to keep attention reduction
# order fixed. This mirrors the flash-attention (fa3) backend. # order fixed. This mirrors the flash-attention (fa3) backend.
self.num_splits = ( self.num_splits = (
1 if model_runner.server_args.enable_deterministic_inference else 0 1 if get_exec().deterministic.enable_deterministic_inference else 0
) )
self.is_encoder_decoder = model_runner.model_config.is_encoder_decoder self.is_encoder_decoder = model_runner.model_config.is_encoder_decoder