[DeepEP] Add SGLANG_DEEPEP_BF16_DISPATCH env var in Normal mode (#13787)

This commit is contained in:
Xiaoyu Zhang
2025-11-23 17:32:43 +08:00
committed by GitHub
parent ac5505b04c
commit b964ce61d6

View File

@@ -377,6 +377,7 @@ class _DeepEPDispatcherImplNormal(_DeepEPDispatcherImplBase):
if (
deep_gemm_wrapper.ENABLE_JIT_DEEPGEMM
and not get_moe_runner_backend().is_cutlass()
and not get_bool_env_var("SGLANG_DEEPEP_BF16_DISPATCH")
):
# TODO hard code 128 block quant,use fp8 communication
hidden_states = sglang_per_token_group_quant_fp8(