Add speculator attention backend switch (#9981)

2025-09-08 06:44:36 +02:00
parent 3b99f23c44
commit 8c5930f08a
6 changed files with 130 additions and 54 deletions
--- a/python/sglang/srt/managers/schedule_batch.py
+++ b/python/sglang/srt/managers/schedule_batch.py
@@ -98,6 +98,7 @@ GLOBAL_SERVER_ARGS_KEYS = [
    "sampling_backend",
    "speculative_accept_threshold_single",
    "speculative_accept_threshold_acc",
+    "speculative_attention_backend",
    "torchao_config",
    "triton_attention_reduce_in_fp32",
    "num_reserved_decode_tokens",