Allow FlashInfer A2A backends for TRTLLM FP8 MoE Modular (#46661)
Signed-off-by: Thien Tran <gau.nernst@yahoo.com.sg>
Thien Tran committed
Jul 9, 2026 at 05:58 UTC
0d2f4e7c9c88d27fdfbb46dc4e2458d615102cb5
1 file changed
+2
vllm/model_executor/layers/fused_moe/experts/trtllm_fp8_moe.py
+2
@@ -136,6 +136,8 @@ class TrtLlmFp8ExpertsModular(TrtLlmFp8ExpertsBase, mk.FusedMoEExpertsModular):
136
not moe_parallel_config.use_all2all_kernels
137
or moe_parallel_config.use_ag_rs_all2all_kernels
138
or moe_parallel_config.use_deepep_v2_kernels
139
+ or moe_parallel_config.use_fi_nvl_one_sided_kernels
140
+ or moe_parallel_config.use_fi_nvl_two_sided_kernels
141
) and not moe_parallel_config.enable_eplb
142
143
@staticmethod