From 8c0f029ac7718af4400d7542bb99417cb901d220 Mon Sep 17 00:00:00 2001 From: liuzhenwei Date: Thu, 6 Aug 2026 14:33:33 +0800 Subject: [PATCH] [XPU] update warning of XPU Graph (#50236) Signed-off-by: zhenwei-intel Co-authored-by: Kunshang Ji --- .buildkite/intel_jobs/test-intel.yaml | 29 +++++++++++++++++++++++++++ vllm/platforms/xpu.py | 9 ++------- 2 files changed, 31 insertions(+), 7 deletions(-) diff --git a/.buildkite/intel_jobs/test-intel.yaml b/.buildkite/intel_jobs/test-intel.yaml index c278b10bc4..ed1872952b 100644 --- a/.buildkite/intel_jobs/test-intel.yaml +++ b/.buildkite/intel_jobs/test-intel.yaml @@ -196,3 +196,32 @@ steps: bash .buildkite/scripts/hardware_ci/run-intel-test.sh 'cd tests && pytest -v -s quantization/test_compressed_tensors.py::test_compressed_tensors_fp8' + - label: "XPU Graph" + depends_on: + - image-build-xpu + timeout_in_minutes: 60 + device: intel_gpu + agent_tags: + label: production + gpu: 1+ + mem: 32+ + no_plugin: true + env: + REGISTRY: "public.ecr.aws/q9t5s3a7" + REPO: "vllm-ci-test-repo" + VLLM_TEST_DEVICE: "xpu" + source_file_dependencies: + - vllm/compilation/ + - vllm/platforms/xpu.py + - vllm/v1/worker/xpu_model_runner.py + - vllm/v1/worker/xpu_worker.py + - .buildkite/intel_jobs/test-intel.yaml + commands: + - >- + bash .buildkite/scripts/hardware_ci/run-intel-test.sh + 'export VLLM_XPU_ENABLE_XPU_GRAPH=1 && + python3 examples/basic/offline_inference/generate.py --model Qwen/Qwen3-0.6B && + python3 examples/basic/offline_inference/generate.py --model Qwen/Qwen3-0.6B --kv-cache-dtype fp8 && + python3 examples/basic/offline_inference/generate.py --model INCModel/Qwen3-30B-A3B-Instruct-2507-MXFP4-CT-AutoRound --max-model-len 4096' + # Temporarily disabled (multi-GPU/MXFP8 not yet supported with XPU Graph): + # python3 examples/basic/offline_inference/generate.py --model INCModel/Qwen3-30B-A3B-Instruct-2507-MXFP8-CT-AutoRound -tp 2 --max-model-len 4096 diff --git a/vllm/platforms/xpu.py b/vllm/platforms/xpu.py index 1104f29189..af19377548 100644 --- a/vllm/platforms/xpu.py +++ b/vllm/platforms/xpu.py @@ -301,13 +301,8 @@ class XPUPlatform(Platform): ) else: logger.warning_once( - "XPU Graph support is experimental and has known limitations: " - "(1) only single-GPU execution is supported; " - "(2) FLASH_ATTN supports PIECEWISE mode only; use TRITON_ATTN " - "for FULL mode; " - "(3) XPU Graph may increase device memory usage, " - "potentially causing OOM errors or leaving less memory " - "for the KV cache and reducing performance." + "XPU Graph support is experimental and currently only supports " + "single-GPU execution." ) # Disable fusion passes not yet supported on XPU.