A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 9, 18:15 UTC
[BugFix] Fix invalid flashinfer_fused_moe_blockscale_fp8 op registration (#32855)
Signed-off-by: Fadi Arafeh <fadi.arafeh@arm.com>
Fadi Arafeh committed
Jan 22, 2026 at 22:27 UTC
fc56f4a0714bd1727ea100a9d75d18ce1a138684
1 file changed
+1
-1
vllm/model_executor/layers/fused_moe/flashinfer_trtllm_moe.py
+1
-1
@@ -131,7 +131,7 @@ def flashinfer_fused_moe_blockscale_fp8(
131
expert_offset: int,
132
local_num_experts: int,
133
block_shape: list[int],
134
- routing_method_type: int = RoutingMethodType.DeepSeekV3,
134
+ routing_method_type: int = int(RoutingMethodType.DeepSeekV3),
135
routed_scaling: float | None = 1.0,
136
) -> torch.Tensor:
137
from vllm.utils.flashinfer import flashinfer_trtllm_fp8_block_scale_moe