diff --git a/vllm/model_executor/layers/fused_moe/experts/batched_deep_gemm_moe.py b/vllm/model_executor/layers/fused_moe/experts/batched_deep_gemm_moe.py index 275f80c68f..90972fc7f0 100644 --- a/vllm/model_executor/layers/fused_moe/experts/batched_deep_gemm_moe.py +++ b/vllm/model_executor/layers/fused_moe/experts/batched_deep_gemm_moe.py @@ -210,7 +210,7 @@ def persistent_masked_m_silu_mul_quant( DeepGemmQuantScaleFMT.UE8M0, ] - device_capability = current_platform.get_device_capability(device_id=y.device.index) + device_capability = current_platform.get_device_capability() assert device_capability is not None cuda_arch = device_capability.to_int()