Fix: Resolve CSV output generation failure in benchmarks

[git-jxj]

Summary

This PR fixes two critical bugs that cause the guidellm benchmark command to crash when using the --output-formats csv option. These changes ensure that CSV reports are generated reliably and correctly.

Details

1. Fix AttributeError**:
   In src/guidellm/benchmark/output.py, a hasattr check has been added to the _get_benchmark_status_metrics_stats function.
   This check ensures that the code only attempts to access status attributes (e.g., successful) on metric objects that actually possess them (like StatusDistributionSummary). It safely skips composite metrics such as GenerativeTextMetricsSummary, thus preventing the AttributeError.

2. Fix KeyError**:
   The data flow for benchmark compilation has been refactored to ensure the original data configuration (args.data) is correctly propagated.
   The method signatures for Benchmarker.run (in benchmarker.py) and GenerativeBenchmark.compile (in schemas.py) were updated to accept a data parameter.
   args.data is now passed in during the benchmarker.run call within benchmark_generative_text (in entrypoints.py).
   As a result, the benchmark.benchmarker.requests field in the final GenerativeBenchmark object now directly stores the raw data configuration, which resolves the KeyError in output.py.

Test Plan

guidellm benchmark
--target "http://x.x.x.x:x/"
--processor "Qwen/Qwen3-8B-FP8"
--rate-type "concurrent"
--rate 1
--max-requests 10
--data "prompt_tokens=32,output_tokens=32,samples=1"
--output-formats csv

Related Issues

#434

• Resolves #

---

• "I certify that all code in this PR is my own, except as noted below."

Use of AI

• Includes AI-assisted code completion
• Includes code generated by an AI application
• Includes AI-generated tests (NOTE: AI written tests should have a docstring that includes ## WRITTEN BY AI ##)

[markurtz]

Looks good, thanks!