A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 15, 07:15 UTC
[TPU] Skip hanging tests (#19115)
Signed-off-by: Siyuan Liu <lsiyuan@google.com>
Siyuan Liu committed
Jun 4, 2025 at 01:43 UTC
8e972d9c44cc8a6b1d0a3596c41604c56a492977
2 files changed
+9
-6
.buildkite/scripts/hardware_ci/run-tpu-v1-test.sh
+1
-1
@@ -150,7 +150,7 @@ run_and_track_test 9 "test_multimodal.py" \
150
run_and_track_test 10 "test_pallas.py" \
151
"python3 -m pytest -s -v /workspace/vllm/tests/v1/tpu/test_pallas.py"
152
run_and_track_test 11 "test_struct_output_generate.py" \
153
- "python3 -m pytest -s -v /workspace/vllm/tests/v1/entrypoints/llm/test_struct_output_generate.py"
153
+ "python3 -m pytest -s -v /workspace/vllm/tests/v1/entrypoints/llm/test_struct_output_generate.py -k 'not test_structured_output_with_reasoning_matrices'"
154
run_and_track_test 12 "test_moe_pallas.py" \
155
"python3 -m pytest -s -v /workspace/vllm/tests/tpu/test_moe_pallas.py"
156
run_and_track_test 13 "test_lora.py" \
tests/v1/tpu/test_spmd_model_weight_loading.py
+8
-5
@@ -45,11 +45,14 @@ def _get_spmd_mesh():
45
return MESH
46
47
48
-@pytest.mark.parametrize("model", [
49
- "Qwen/Qwen2-1.5B-Instruct",
50
- "meta-llama/Llama-3.1-8B-Instruct",
51
- "meta-llama/Llama-3.1-70B-Instruct",
52
-])
48
+@pytest.mark.parametrize(
49
+ "model",
50
+ [
51
+ "Qwen/Qwen2-1.5B-Instruct",
52
+ # Skip large models due to CI runner disk space limitations
53
+ # "meta-llama/Llama-3.1-8B-Instruct",
54
+ # "meta-llama/Llama-3.1-70B-Instruct",
55
+ ])
56
def test_tpu_model_loader(model):
57
# Skip the 70B test if there are less than 8 chips
58
# TODO: Query using torch xla API, the query API is not working