biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

Fix range_ratio Bug in RandomDataset (#16126)

Browse Source 
Signed-off-by: jadewang21 <jadewangcn@outlook.com>
This commit is contained in:
WWW
2025-04-11 06:31:17 +08:00
committed by GitHub
parent 3cc9af88ff
commit 268c325078
3 changed files with 42 additions and 20 deletions
Download Patch File Download Diff File Expand all files Collapse all files

23
benchmarks/benchmark_dataset.py
View File

@@ -288,7 +288,7 @@ def process_image(image: Any) -> Mapping[str, Any]:
class RandomDataset(BenchmarkDataset):
# Default values copied from benchmark_serving.py for the random dataset.
DEFAULT_PREFIX_LEN = 0
DEFAULT_RANGE_RATIO = 1.0
DEFAULT_RANGE_RATIO = 0.0
DEFAULT_INPUT_LEN = 1024
DEFAULT_OUTPUT_LEN = 128
@@ -308,19 +308,32 @@ class RandomDataset(BenchmarkDataset):
output_len: int = DEFAULT_OUTPUT_LEN,
**kwargs,
) -> list[SampleRequest]:
# Enforce range_ratio < 1
assert range_ratio < 1.0, (
"random_range_ratio must be < 1.0 to ensure a valid sampling range"
)
vocab_size = tokenizer.vocab_size
prefix_token_ids = (np.random.randint(
0, vocab_size, size=prefix_len).tolist() if prefix_len > 0 else [])
input_low = int(input_len * range_ratio)
output_low = int(output_len * range_ratio)
# New sampling logic: [X * (1 - b), X * (1 + b)]
input_low = int(input_len * (1 - range_ratio))
input_high = int(input_len * (1 + range_ratio))
output_low = int(output_len * (1 - range_ratio))
output_high = int(output_len * (1 + range_ratio))
# Add logging for debugging
logger.info("Sampling input_len from [%s, %s]", input_low, input_high)
logger.info("Sampling output_len from [%s, %s]", output_low,
output_high)
input_lens = np.random.randint(input_low,
input_len + 1,
input_high + 1,
size=num_requests)
output_lens = np.random.randint(output_low,
output_len + 1,
output_high + 1,
size=num_requests)
offsets = np.random.randint(0, vocab_size, size=num_requests)