[XPU] bump up auto-round-lib to 0.16.0 (#60976)

Signed-off-by: Zhenzhong1 <zhenzhong.xu@intel.com>
Co-authored-by: Kunshang Ji <kunshang.ji@intel.com>

requirements/xpu.txt

@@ -19,5 +19,5 @@ torchaudio
 torchvision
 torchcodec >= 0.14 # Required for the torchcodec video decoding backend
 
-auto_round_lib==0.15.0
+auto_round_lib==0.16.0
 vllm_xpu_kernels==0.1.15.4

tests/quantization/test_auto_round.py

@@ -146,21 +146,6 @@ MODEL_RUNNER_KWARGS: dict[str, dict[str, Any]] = {
     },
 }
 
-# These checkpoints fail during XPU graph capture, so run them eagerly there.
-XPU_EAGER_ONLY_MODELS = {
-    "OPEA/Qwen2.5-0.5B-Instruct-int4-sym-inc",
-    "Intel/Qwen2-0.5B-Instruct-int4-sym-AutoRound",
-    "Intel/Qwen3-8B-w2g64-for-ut",
-    "INCModel/Qwen3-30B-A3B-12L-W4A16-test",
-}
-
-
-def _runner_kwargs(model: str) -> dict[str, Any]:
-    kwargs = dict(MODEL_RUNNER_KWARGS.get(model, {}))
-    if current_platform.is_xpu() and model in XPU_EAGER_ONLY_MODELS:
-        kwargs["enforce_eager"] = True
-    return kwargs
-
 
 @pytest.mark.skipif(
     not (
@@ -172,7 +157,7 @@ def _runner_kwargs(model: str) -> dict[str, Any]:
 )
 @pytest.mark.parametrize("model", MODELS + QWEN3_AUTOROUND_MODELS)
 def test_auto_round_model(vllm_runner, model):
-    with vllm_runner(model, **_runner_kwargs(model)) as llm:
+    with vllm_runner(model, **MODEL_RUNNER_KWARGS.get(model, {})) as llm:
         output = llm.generate_greedy(["The capital of France is"], max_tokens=8)
 
     assert output