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 7, 10:15 UTC
[ROCm][CI] Fix Assertion Logic For `test_gpt_oss` (#35806)
Signed-off-by: Micah Williamson <micah.williamson@amd.com>
Micah Williamson committed
Mar 2, 2026 at 23:08 UTC
8b9e8b74541ea21a9555f8726daf10034c907278
1 file changed
+5
-5
tests/models/quantization/test_gpt_oss.py
+5
-5
@@ -12,8 +12,8 @@ Config:
12
Run: pytest tests/models/quantization/test_gpt_oss.py
13
"""
14
15
-import importlib
15
import importlib.metadata
16
+import importlib.util
17
from dataclasses import dataclass
18
19
import huggingface_hub
@@ -104,7 +104,7 @@ def test_gpt_oss_attention_quantization(
104
)
105
106
rtol = 0.02
107
- assert (
108
- measured_accuracy - rtol < expected_accuracy
109
- and measured_accuracy + rtol > expected_accuracy
110
- ), f"Expected: {expected_accuracy} | Measured: {measured_accuracy}"
107
+ assert measured_accuracy >= expected_accuracy - rtol, (
108
+ f"Accuracy {measured_accuracy:.4f} is below threshold "
109
+ f"{expected_accuracy - rtol:.4f} (expected >= {expected_accuracy} - {rtol})"
110
+ )