minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

Lower AutoRound quantization MMLU threshold (#31748)

Browse Source
This commit is contained in:
Xiaoyu Zhang
2026-07-20 13:40:48 +08:00
committed by GitHub
parent 1843384c7a
commit 829e9ce9d5
1 changed files with 7 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files
test/registered/quant/test_autoround_quantization.py
+7 -3
View File
@@ -26,6 +26,10 @@ from sglang.test.test_utils import (
register_cuda_ci(est_time=120, stage="extra-a", runner_config="1-gpu-large") register_cuda_ci(est_time=120, stage="extra-a", runner_config="1-gpu-large")
MMLU_NUM_EXAMPLES = 32
MMLU_NUM_THREADS = 32
MMLU_SCORE_THRESHOLD = 22 / MMLU_NUM_EXAMPLES
class TestAutoRoundQuantization(CustomTestCase): class TestAutoRoundQuantization(CustomTestCase):
@classmethod @classmethod
@@ -52,12 +56,12 @@ class TestAutoRoundQuantization(CustomTestCase):
base_url=self.base_url, base_url=self.base_url,
model=self.model, model=self.model,
eval_name="mmlu", eval_name="mmlu",
num_examples=32, num_examples=MMLU_NUM_EXAMPLES,
num_threads=32, num_threads=MMLU_NUM_THREADS,
device="auto", device="auto",
) )
metrics = run_eval(args) metrics = run_eval(args)
self.assertGreaterEqual(metrics["score"], 0.7) self.assertGreaterEqual(metrics["score"], MMLU_SCORE_THRESHOLD)
finally: finally:
kill_process_tree(process.pid) kill_process_tree(process.pid)
print(f"[INFO] Server for {self.model} stopped.") print(f"[INFO] Server for {self.model} stopped.")