diff --git a/.buildkite/intel_jobs/test-intel.yaml b/.buildkite/intel_jobs/test-intel.yaml index 3ebae7284c..898c248a48 100644 --- a/.buildkite/intel_jobs/test-intel.yaml +++ b/.buildkite/intel_jobs/test-intel.yaml @@ -36,6 +36,7 @@ steps: python3 examples/basic/offline_inference/generate.py --model facebook/opt-125m --block-size 64 --enforce-eager --attention-backend=TRITON_ATTN && python3 examples/basic/offline_inference/generate.py --model facebook/opt-125m --block-size 64 --enforce-eager --quantization fp8 && python3 examples/basic/offline_inference/generate.py --model facebook/opt-125m --block-size 64 --enforce-eager --kv-cache-dtype fp8 && + python3 examples/basic/offline_inference/generate.py --model nvidia/Llama-3.1-8B-Instruct-FP8 --block-size 64 --enforce-eager --quantization modelopt --kv-cache-dtype fp8 --attention-backend TRITON_ATTN --max-model-len 4096 && python3 examples/basic/offline_inference/generate.py --model superjob/Qwen3-4B-Instruct-2507-GPTQ-Int4 --block-size 64 --enforce-eager --max-model-len 8192 && python3 examples/basic/offline_inference/generate.py --model ibm-research/PowerMoE-3b --block-size 64 --enforce-eager -tp 2 && python3 examples/basic/offline_inference/generate.py --model ibm-research/PowerMoE-3b --block-size 64 --enforce-eager -tp 2 --enable-expert-parallel' diff --git a/vllm/model_executor/kernels/linear/scaled_mm/xpu.py b/vllm/model_executor/kernels/linear/scaled_mm/xpu.py index 6d75a420e7..7e44fb943f 100644 --- a/vllm/model_executor/kernels/linear/scaled_mm/xpu.py +++ b/vllm/model_executor/kernels/linear/scaled_mm/xpu.py @@ -46,7 +46,18 @@ class XPUFP8ScaledMMLinearKernel(FP8ScaledMMLinearKernel): self.layer_param_names = layer_param_names def process_weights_after_loading(self, layer: torch.nn.Module) -> None: - replace_parameter(layer, "weight", layer.weight.data.t()) + # fp8_gemm_w8a16 expects weight in [in, out] layout. + # Transpose if weight is still in [out, in] layout. + # For square matrices, use contiguity as tie-breaker: + # checkpoint weights are contiguous, .t() views are not. + weight = layer.weight + out_features, in_features = self.config.weight_shape + + if weight.shape == (out_features, in_features) and ( + in_features != out_features or weight.is_contiguous() + ): + replace_parameter(layer, "weight", weight.data.t()) + # else: already in [in, out] layout — no-op def apply_weights( self,