Disable enforce_eager for V1 TPU sampler and structured output tests (vllm-project#17016)

Signed-off-by: mgoin <mgoin64@gmail.com>

Author: mgoin

.buildkite/scripts/hardware_ci/run-tpu-v1-test.sh

@@ -19,6 +19,7 @@ docker run --privileged --net host --shm-size=16G -it \
     vllm-tpu /bin/bash -c "python3 -m pip install git+https://github.com/thuml/depyf.git \
     && python3 -m pip install pytest pytest-asyncio tpu-info \
     && python3 -m pip install lm_eval[api]==0.4.4 \
+    && export VLLM_XLA_CACHE_PATH= \
     && export VLLM_USE_V1=1 \
     && export VLLM_XLA_CHECK_RECOMPILATION=1 \
     && echo HARDWARE \

tests/v1/entrypoints/llm/test_struct_output_generate.py

@@ -13,6 +13,7 @@
 
 from vllm.entrypoints.llm import LLM
 from vllm.outputs import RequestOutput
+from vllm.platforms import current_platform
 from vllm.sampling_params import GuidedDecodingParams, SamplingParams
 
 PARAMS_MODELS_BACKENDS_TOKENIZER_MODE = [
@@ -63,10 +64,13 @@ def test_structured_output(
 ):
     monkeypatch.setenv("VLLM_USE_V1", "1")
 
+    # 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=True,
+              enforce_eager=enforce_eager,
               max_model_len=1024,
               guided_decoding_backend=guided_decoding_backend,
               tokenizer_mode=tokenizer_mode)

tests/v1/tpu/test_sampler.py

@@ -23,7 +23,7 @@ def test_sampler_different(model_name: str):
     different results.
     """
     llm = LLM(model_name,
-              enforce_eager=True,
+              enforce_eager=False,
               max_num_seqs=1,
               max_model_len=512,
               max_num_batched_tokens=512)