[GuideLLM Refactor] Complete CSV output (#378)

## Summary

This PR gets the CSV output to a state comparable to pre-refactor.

## Details

Implements the functions to export the required data to the CSV format.

The goal is to include the required information in the CSV without
cluttering it, but also without creating too much of a burden to the
future maintainers resulting from referencing specific schema elements.

The following columns are new:
- Profile (an entire JSON dump of the profile)
- Backend (the entire JSON dump of the internal data structure)
- Generator Data

You can view these files in the attached output generated by the
following test:
`guidellm benchmark run --output-path result_7.csv --max-seconds 2
--target=http://localhost:8000 --data
"prompt_tokens=256,output_tokens=128" --rate-type constant --rate 1
--output-formats csv`


[result_7.csv](https://github.com/user-attachments/files/22222220/result_7.csv)


## Test Plan

Run GuideLLM with the following additional args: `--output-path
result.csv --output-formats csv`

The generated file should have all info required.

## Related Issues

This is a part of the scheduler refactor.

---

- [x] "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 ##`)

Author: jaredoconnell

src/guidellm/benchmark/output.py

@@ -567,6 +567,13 @@ async def finalize(self, report: GenerativeBenchmarksReport) -> Path:
                 benchmark_headers: list[str] = []
                 benchmark_values: list[str | float | list[float]] = []
 
+                # Add basic run description info
+                desc_headers, desc_values = (
+                    self._get_benchmark_desc_headers_and_values(benchmark)
+                )
+                benchmark_headers.extend(desc_headers)
+                benchmark_values.extend(desc_values)
+
                 # Add status-based metrics
                 for status in StatusDistributionSummary.model_fields:
                     status_headers, status_values = (
@@ -672,6 +679,21 @@ def _get_benchmark_status_metrics_stats(
         ]
         return headers, values
 
+    def _get_benchmark_extras_headers_and_values(
+        self, benchmark: GenerativeBenchmark,
+    ) -> tuple[list[str], list[str]]:
+        headers = ["Profile", "Backend", "Generator Data"]
+        values: list[str] = [
+            benchmark.benchmarker.profile.model_dump_json(),
+            json.dumps(benchmark.benchmarker.backend),
+            json.dumps(benchmark.benchmarker.requests["attributes"]["data"]),
+        ]
+
+        if len(headers) != len(values):
+            raise ValueError("Headers and values length mismatch.")
+
+        return headers, values
+
 
 @GenerativeBenchmarkerOutput.register("html")
 class GenerativeBenchmarkerHTML(GenerativeBenchmarkerOutput):

tests/unit/benchmark/test_output.py

@@ -80,7 +80,6 @@ def test_file_yaml():
 
     mock_path.unlink()
 
-@pytest.mark.skip(reason="CSV fix not merged yet")
 @pytest.mark.asyncio
 async def test_file_csv():
     mock_benchmark = mock_generative_benchmark()
@@ -96,6 +95,7 @@ async def test_file_csv():
         rows = list(reader)
 
     assert "Type" in headers
+    assert "Profile" in headers
     assert len(rows) == 1
 
     mock_path.unlink()

tests/unit/mock_benchmark.py

@@ -76,7 +76,11 @@ def mock_generative_benchmark() -> GenerativeBenchmark:
         ),
         benchmarker=BenchmarkerDict(
             profile=SynchronousProfile.create("synchronous", rate=None),
-            requests={},
+            requests={
+                "attributes": {
+                    "data": "prompt_tokens=256,output_tokens=128",
+                },
+            },
             backend={},
             environment={},
             aggregators={},