Revert "make radix cache deterministic" (#11728)

This commit is contained in:
Baizhou Zhang
2025-10-16 16:36:15 -05:00
committed by GitHub
parent c7962868c1
commit b0d1d717e1
5 changed files with 17 additions and 81 deletions

View File

@@ -3441,16 +3441,3 @@ def cached_triton_kernel(key_fn=None):
return CachedKernel(fn, key_fn)
return decorator
DEFAULT_DETERMINISTIC_INFERENCE_BACKEND_SIZE = 4096
DEFAULT_DETERMINISTIC_INFERENCE_BACKEND_SIZE_CONFIG = {
"flashinfer": (
"SGLANG_FLASHINFER_PREFILL_SPLIT_TILE_SIZE",
DEFAULT_DETERMINISTIC_INFERENCE_BACKEND_SIZE,
),
"triton": (
"SGLANG_TRITON_PREFILL_TRUNCATION_ALIGN_SIZE",
DEFAULT_DETERMINISTIC_INFERENCE_BACKEND_SIZE,
),
}