From 5b6fe23d05851b7bd353dbaf2a322f479215d726 Mon Sep 17 00:00:00 2001 From: Kuntai Du Date: Thu, 10 Jul 2025 14:52:46 -0700 Subject: [PATCH] [Bugfix][Benchmark] Make sure the output length > 0 when testing prefill workload. (#20786) Signed-off-by: KuntaiDu Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com> --- benchmarks/benchmark_dataset.py | 3 +++ 1 file changed, 3 insertions(+) diff --git a/benchmarks/benchmark_dataset.py b/benchmarks/benchmark_dataset.py index 8df071d6033fa..1ad6cef7a9dbc 100644 --- a/benchmarks/benchmark_dataset.py +++ b/benchmarks/benchmark_dataset.py @@ -324,6 +324,9 @@ class RandomDataset(BenchmarkDataset): input_low = int(real_input_len * (1 - range_ratio)) input_high = int(real_input_len * (1 + range_ratio)) output_low = int(output_len * (1 - range_ratio)) + # Ensure the lower bound for output length is at least 1 to prevent + # sampling 0 tokens, which can cause request failures. + output_low = max(output_low, 1) output_high = int(output_len * (1 + range_ratio)) # Add logging for debugging