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, 12:15 UTC
[easy] Print number of needed GPUs in skip message (#17594)
Signed-off-by: rzou <zou3519@gmail.com>
Richard Zou committed
May 2, 2025 at 18:27 UTC
b90b0852e9011b48c3e8ff9009aa57e9428350d1
1 file changed
+2
-1
tests/compile/test_basic_correctness.py
+2
-1
@@ -103,7 +103,8 @@ def test_compile_correctness(
103
method = test_setting.method
104
fullgraph = test_setting.fullgraph
105
if cuda_device_count_stateless() != pp_size * tp_size:
106
- pytest.skip("Not correct CUDA devices for the test.")
106
+ pytest.skip(f"Need exactly {pp_size}*{tp_size} CUDA gpus but got "
107
+ f"{cuda_device_count_stateless()}")
108
109
with monkeypatch.context() as m:
110
m.setenv("VLLM_ATTENTION_BACKEND", attn_backend)