From 531d811c61ee785713d050f0d89c5a5ab9537855 Mon Sep 17 00:00:00 2001
From: "github-actions[bot]"
 <41898282+github-actions[bot]@users.noreply.github.com>
Date: Wed, 28 Oct 2026 17:41:48 +0000
Subject: [PATCH] chore: move vllm-xpu-bench to
 intel/gpu-ai-skills@6e7bce819e2e

https://github.com/intel/gpu-ai-skills/compare/0b4fafd09c5e...6e7bce819e2e
---
 skills.yaml                        |  2 +-
 skills/vllm-xpu-bench/.source.json |  2 +-
 skills/vllm-xpu-bench/SKILL.md     | 12 +++++++-----
 3 files changed, 9 insertions(+), 7 deletions(-)

diff --git a/skills.yaml b/skills.yaml
index e4838dc..6ee2fa8 100644
--- a/skills.yaml
+++ b/skills.yaml
@@ -291,7 +291,7 @@
 - name: vllm-xpu-bench
   maintainer: "xaleryb"
   external-repo: "https://github.com/intel/gpu-ai-skills"
-  external-commit: "0b4fafd09c5eb4cc5daf532d915ef5984a919775"
+  external-commit: "6e7bce819e2e44a4aac692b0a49253f39c0f6957"
   external-path: "plugins/intel-gpu-ai-skills/skills/vllm-xpu-bench"
   external-license: "Apache-2.0"
   intel-products: "Intel GPU"
diff --git a/skills/vllm-xpu-bench/.source.json b/skills/vllm-xpu-bench/.source.json
index f94e3ab..234074d 100644
--- a/skills/vllm-xpu-bench/.source.json
+++ b/skills/vllm-xpu-bench/.source.json
@@ -1,6 +1,6 @@
 {
   "repo": "https://github.com/intel/gpu-ai-skills",
   "path": "plugins/intel-gpu-ai-skills/skills/vllm-xpu-bench",
-  "commit": "0b4fafd09c5eb4cc5daf532d915ef5984a919775",
+  "commit": "6e7bce819e2e44a4aac692b0a49253f39c0f6957",
   "license": "Apache-2.0"
 }
diff --git a/skills/vllm-xpu-bench/SKILL.md b/skills/vllm-xpu-bench/SKILL.md
index 68d46ae..e94703a 100644
--- a/skills/vllm-xpu-bench/SKILL.md
+++ b/skills/vllm-xpu-bench/SKILL.md
@@ -7,9 +7,13 @@ description: Benchmark a **running vLLM-XPU OpenAI-compatible server** on an Int
 
 `vllm bench` is the same CLI on Intel as on CUDA. The XPU-specific
 levers are the serve-side flags from **vllm-xpu-run**
-(`--enforce-eager`, `--max-model-len`, `--gpu-memory-utilization`,
-`--block-size=64`). Pure measurement; for fixes see profiling
-skills.
+(`--max-model-len`, `--gpu-memory-utilization`, `--block-size=64`).
+Pure measurement; for fixes see profiling skills.
+
+Leave XPU graphs on when benchmarking v0.31.0+ images: adding
+`--enforce-eager` roughly halves decode throughput (measured TPOT
+64 ms vs 36 ms), so an eager baseline understates what the deployment
+will actually do.
 
 ## Preflight — find the container and verify the server
 
@@ -78,7 +82,6 @@ docker run -d --name <container-name> \
     vllm/vllm-openai-xpu:latest \
     <model-id> \
         --dtype bfloat16 \
-        --enforce-eager \
         --block-size=64 \
         --max-model-len 4096 \
         --gpu-memory-utilization 0.85 \
@@ -226,7 +229,6 @@ docker run --rm \
     bench throughput \
         --model Qwen/Qwen2.5-1.5B-Instruct \
         --dtype bfloat16 \
-        --enforce-eager \
         --max-model-len 4096 \
         --gpu-memory-utilization 0.85 \
         --dataset-name random \