diff --git a/.buildkite/intel_jobs/misc_intel.yaml b/.buildkite/intel_jobs/misc_intel.yaml index 20a9f5bd17..cdbbdc850b 100644 --- a/.buildkite/intel_jobs/misc_intel.yaml +++ b/.buildkite/intel_jobs/misc_intel.yaml @@ -63,7 +63,8 @@ steps: pytest -v -s v1/test_oracle.py && pytest -v -s v1/test_request.py && pytest -v -s v1/test_outputs.py && - pytest -v -s v1/sample/test_topk_topp_sampler.py' + pytest -v -s v1/sample/test_topk_topp_sampler.py && + pytest -v -s v1/sample/test_logprobs.py' - label: XPU CPU Offload timeout_in_minutes: 60 diff --git a/tests/v1/sample/test_logprobs.py b/tests/v1/sample/test_logprobs.py index 963e7423f7..49352683de 100644 --- a/tests/v1/sample/test_logprobs.py +++ b/tests/v1/sample/test_logprobs.py @@ -42,11 +42,12 @@ SAMPLE_PROMPT = BatchLogprobsComposition.SAMPLE_PROMPT # # Force LLM instances into an identical, deterministic execution # mode so the test isolates spec-decode correctness only: -ROCM_DETERMINISM_KWARGS: dict = ( - dict(max_num_seqs=1, attention_backend="TRITON_ATTN") - if current_platform.is_rocm() - else {} -) +if current_platform.is_rocm(): + GPU_DETERMINISM_KWARGS: dict = dict(max_num_seqs=1, attention_backend="TRITON_ATTN") +elif current_platform.is_xpu(): + GPU_DETERMINISM_KWARGS = dict(max_num_seqs=1, attention_backend="FLASH_ATTN") +else: + GPU_DETERMINISM_KWARGS = {} @pytest.fixture( @@ -1127,7 +1128,7 @@ def test_spec_decode_logprobs( enable_chunked_prefill=True, max_num_batched_tokens=32, enable_prefix_caching=False, - **ROCM_DETERMINISM_KWARGS, + **GPU_DETERMINISM_KWARGS, ) # Run base LLM.