fix: Llama 4 BF16 load on Blackwell (#12308)

2025-10-28 18:59:01 -07:00
parent b48354c537
commit c143f416ce
1 changed files with 5 additions and 4 deletions
@@ -972,6 +972,7 @@ class ServerArgs:
                    "Use trtllm_mha as attention backend on sm100 for Llama4 model"
                )
            if is_sm100_supported() and self.moe_runner_backend == "auto":
+                if self.quantization in {"fp8", "modelopt_fp8"}:
                    self.moe_runner_backend = "flashinfer_trtllm"
                    logger.info(
                        "Use flashinfer_trtllm as MoE runner backend on SM100 for Llama4"