From adae0c1f439263906b5da7b4d9a06b8cdc38719e Mon Sep 17 00:00:00 2001 From: Xiang Si Date: Fri, 3 Oct 2025 06:38:42 -0700 Subject: [PATCH] [CI/Build] do not enforce precompilation on tpu ci tests (#25992) Signed-off-by: Xiang Si --- tests/v1/entrypoints/llm/test_struct_output_generate.py | 5 +---- 1 file changed, 1 insertion(+), 4 deletions(-) diff --git a/tests/v1/entrypoints/llm/test_struct_output_generate.py b/tests/v1/entrypoints/llm/test_struct_output_generate.py index 5017c83025ba..83493e25b7a6 100644 --- a/tests/v1/entrypoints/llm/test_struct_output_generate.py +++ b/tests/v1/entrypoints/llm/test_struct_output_generate.py @@ -123,13 +123,10 @@ def test_structured_output( if current_platform.is_tpu() and speculative_config: pytest.skip("TPU does not support speculative decoding") - # Don't use eager execution on TPUs because we want to test for no - # recompilation at runtime - enforce_eager = bool(not current_platform.is_tpu()) # Use a single LLM instance for several scenarios to # speed up the test suite. llm = LLM(model=model_name, - enforce_eager=enforce_eager, + enforce_eager=True, max_model_len=1024, structured_outputs_config=dict(backend=backend, disable_any_whitespace=backend