[V1][Spec Decode] Optimize Rejection Sampler with Triton Kernels (#14930)

Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>
2025-03-18 14:31:54 -07:00
parent 3a1e648158
commit 99abb8b650
8 changed files with 875 additions and 408 deletions
--- a/vllm/envs.py
+++ b/vllm/envs.py
@@ -35,7 +35,6 @@ if TYPE_CHECKING:
    VLLM_TRACE_FUNCTION: int = 0
    VLLM_ATTENTION_BACKEND: Optional[str] = None
    VLLM_USE_FLASHINFER_SAMPLER: Optional[bool] = None
-    VLLM_USE_FLASHINFER_REJECTION_SAMPLER: bool = False
    VLLM_FLASHINFER_FORCE_TENSOR_CORES: bool = False
    VLLM_PP_LAYER_PARTITION: Optional[str] = None
    VLLM_CPU_KVCACHE_SPACE: int = 0