run

name	run
description	Run a benchmark experiment using a TOML config file. Use when the user wants to run, train, or sweep experiments — single dataset/model or multi-axis sweeps — against any of the available configs/runs/ configs.

When to use / what to ask

If the user has not specified a config, ask which experiment they want to run and suggest:

Intent	Config
Single dataset + model	configs/runs/run_single_data.toml
Sweep over models	configs/runs/sweep_model.toml
Sweep over datasets	configs/runs/sweep_data.toml
Multi-axis sweep	configs/runs/multi_sweep.toml
GIFT-EVAL benchmark	configs/runs/gift_eval_sweep.toml

Commands

# Preview what a sweep expands to (run count, datasets, models)
uv run python tool/inspect_config.py --config <config_path>

# Run a single config (the only CLI flag is --config)
uv run modern-tsf --config <config_path>

# Run several configs / pick GPUs (see the `sweep` skill)
uv run python tool/tsf.py run <config ...> [--jobs N] [--gpus 0,1]

Config knobs (set in the TOML, not the CLI)

• [training.tricks] — grad_clip, grad_accum, curriculum, aux-loss options.
• [evaluation] strategy = "rolling" — rolling forecast instead of single-shot eval.
• [evaluation] enable_profile = true — record params/MACs/latency per run.
• For ablation/hyperparameter sweeps and forecast case plots, use the experiments skill.

Notes

• Results land in work_dirs/<dataset>/<model>/performance.csv; offer to aggregate (and optionally plot) afterwards via the aggregate skill.

Reference

Config structure, extends, and sweep syntax: docs/en/configs.md; tsf runner: docs/en/scripts.md.