Fix ValidationError when re-exporting benchmarks with console output

[martin-pfister]

Summary

• reimport_benchmarks_report() (used by guidellm benchmark from-file) unconditionally attached a file path to every requested output format.
• ConsoleBenchmarkOutputArgs forbids extra fields (no path), so requesting console output — the default — raised pydantic_core.ValidationError: console.path Extra inputs are not permitted.
• Now a path is only attached when the resolved output args class for that format actually declares a path field, so console output (and any future path-less format) works while file-based formats (json/yaml/html/csv) keep their paths.

Fixes #925

Test plan

• Added a regression test (test_reimport_with_console_format_does_not_raise) that reproduces the exact crash from the issue and asserts it no longer raises.
• ruff check and mypy pass on the changed files.
• pytest tests/unit/entrypoints tests/unit/benchmark passes (75 passed, pre-existing unrelated skips only).
• Manually reproduced the crash on main and confirmed the fix resolves it via guidellm benchmark from-file benchmarks.json --output-formats console.

---

git log

commit cbd16c5
Author: Martin Pfister 30149150+martin-pfister@users.noreply.github.com
Date: Fri Jul 10 15:11:43 2026 +0000

Fix ValidationError when re-exporting benchmarks with console output

reimport_benchmarks_report() unconditionally attached a file path to
every requested output format, but ConsoleBenchmarkOutputArgs forbids
extra fields since console output has no file to write to. This broke
`guidellm benchmark from-file` whenever console was among the requested
formats (the default).

Only attach a path when the resolved output args class actually
declares a `path` field.

Fixes #925

Signed-off-by: Martin Pfister <30149150+martin-pfister@users.noreply.github.com>

---

Signed-off-by: Martin Pfister 30149150+martin-pfister@users.noreply.github.com

[mergify]

Hi @martin-pfister, the DCO check has failed. Please click on DCO in the Checks section for instructions on how to resolve this.

[dbutenhof]

While this approach may work in the short term, it's a goal to upgrade the benchmark from-file and preprocess dataset subcommands to match the modern registry-friendly format of the 0.7.x run command. That is, --output kind=console --output kind=json,path=bmark.json. (Unfortunately we had a hard deadline to complete 0.7.0 and we missed this.)

We'd prefer to complete the CLI revision rather than take an intermediate short-term fix like this, although we appreciate the contribution and may still decide it's worth accepting.

[dbutenhof]

See #927

[dbutenhof]

Closing because the issue has been resolved by #927.

But thanks for donating the time & effort to try to help!