Fix boolean nested params, add dict format support, and enhance plotting for vllm bench sweep

[Copilot]

Purpose

This PR implements key enhancements to vllm bench sweep serve and plotting:

1. Fixed Boolean Nested Dot-Notation Parameters

• Before: compilation_config.use_inductor_graph_partition: false → --no-compilation-config.use_inductor_graph_partition ❌
• After: compilation_config.use_inductor_graph_partition: false → --compilation-config.use_inductor_graph_partition=false ✓

2. Fixed Nested Dictionary Values

• Dict values properly JSON serialized instead of Python repr strings

3. Dict Format Support

• Parameter files can now use dict format where keys are experiment names
• Uses _benchmark_name field internally (not passed as CLI parameter)
• Validates that all _benchmark_name values are unique

4. Enhanced Plot Options

• --no-error-bars flag to disable error bars (default: enabled)
• --fig-height to control subplot height in inches (default: 6.4)
• --fig-dpi to control figure resolution (default: 300)
• --fig-name parameter to customize figure prefix (default: "FIGURE")
• Curve names are now alphabetically sorted in legends

5. Improved Data Handling

• String values "inf", "-inf", and "nan" are automatically converted to floats when loading data

Backward Compatibility

✓ All changes are backward compatible - existing list format continues to work

Test Plan

# Run parameter sweep unit tests
pytest tests/benchmarks/test_param_sweep.py

# Run plot filter unit tests
pytest tests/benchmarks/test_plot_filters.py

# Test with example configs
vllm bench sweep serve \
  --serve-cmd "vllm serve meta-llama/Llama-3.2-1B-Instruct" \
  --bench-cmd "vllm bench serve --model meta-llama/Llama-3.2-1B-Instruct" \
  --serve-params serve_params_dict.json \
  --bench-params bench_params_dict.json \
  --dry-run

# Test custom figure options
vllm bench sweep plot results --fig-name custom_plot --fig-height 8 --fig-dpi 300 --no-error-bars

Test Result

All unit tests pass. Verified:

• Nested boolean params generate correct =true/false syntax
• Non-nested booleans still use --no- prefix (backward compatible)
• Dict values properly JSON serialized with complex nested structures
• Dict format loads and uses names in output paths
• List format continues working unchanged
• _benchmark_name uniqueness validation works correctly (raises ValueError on duplicates)
• Plot filter tests validate all comparison operators work correctly
• Figure naming, height, DPI, and error bar options work correctly
• Curve names are alphabetically sorted in legends

---

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.

[ProExpertProg]

@copilot

[ProExpertProg]

@copilot also use _benchmark_name to store the name instead of name, and assert they are all unique (if provided via the list of records format).

[ProExpertProg]

@copilot also use _benchmark_name to store the name instead of name, and assert they are all unique (if provided via the list of records format).

[Copilot]

@copilot also use _benchmark_name to store the name instead of name, and assert they are all unique (if provided via the list of records format).

Done in cb81309. Changed to use _benchmark_name and added validation to ensure all values are unique when provided via list or dict format. Duplicates now raise a descriptive ValueError.

[ProExpertProg]

@copilot

[ProExpertProg]

@copilot add tests for filtering including the "inf" string edge case

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[ProExpertProg]

@copilot

[ProExpertProg]

@copilot remove the string test cases as we now coerce the inf/nan to float

[Copilot]

Done in 59d9982. Removed the string test cases (df_inf_string, test_equal_to_inf_string, test_not_equal_to_inf_string, test_inf_string_mixed_dataframe) since inf/nan values are now coerced to floats when loading data.

[ProExpertProg]

@copilot