Add mixed-prefix gsm8k eval and its CPU unit test (#27502)

This commit is contained in:
fzyzcjy
2026-06-09 20:17:41 +08:00
committed by GitHub
parent fdcd28a08d
commit 609f5f549c
4 changed files with 265 additions and 5 deletions
+23
View File
@@ -178,6 +178,17 @@ def run_eval(args):
num_shots=getattr(args, "num_shots", 5),
data_path=getattr(args, "gsm8k_data_path", None),
)
elif args.eval_name == "mixed_prefix_gsm8k":
from sglang.test.simple_eval_mixed_prefix_gsm8k import MixedPrefixGSM8KEval
eval_obj = MixedPrefixGSM8KEval(
num_examples=args.num_examples,
num_threads=args.num_threads,
num_shots=args.num_shots,
secondary_pool_size=args.mixed_prefix_gsm8k_secondary_pool_size,
data_path=args.gsm8k_data_path,
seed=args.mixed_prefix_gsm8k_seed,
)
else:
raise ValueError(f"Invalid eval name: {args.eval_name}")
@@ -367,6 +378,18 @@ if __name__ == "__main__":
default=None,
help="Path to GSM8K data file (e.g., test.jsonl)",
)
parser.add_argument(
"--mixed-prefix-gsm8k-secondary-pool-size",
type=int,
default=15,
help="Size of secondary example pool for eval_name=mixed_prefix_gsm8k (default: 15)",
)
parser.add_argument(
"--mixed-prefix-gsm8k-seed",
type=int,
default=42,
help="Seed for per-question random sampling in mixed_prefix_gsm8k (default: 42)",
)
args = parser.parse_args()