smoke

name	smoke
description	Verify model(s) end-to-end by running their smoke configs concurrently and reporting PASS/FAIL via `tool/tsf.py smoke`. Use when the user wants to check that a model (new or existing) trains and produces the right output shape, validate a port, or run a quick CI-style gate over many models.

When to use

Run this after scaffolding or editing a model, after porting a model, or any time you want a fast end-to-end check that one or many models register, train for one epoch on CPU, and emit the correct output shape. Each model has a configs/runs/smoke_<module>.toml (the new-model scaffold creates one automatically).

Command

# Verify one model (resolves configs/runs/smoke_<snake(name)>.toml)
uv run python tool/tsf.py smoke --model MyModel

# Verify every model in the repo, 8 at a time
uv run python tool/tsf.py smoke --all --jobs 8

# Verify explicit smoke config(s)
uv run python tool/tsf.py smoke --config configs/runs/smoke_dlinear.toml configs/runs/smoke_gwnet.toml

--jobs N sets the concurrency (default min(8, cpu)). Smoke runs are tiny (1 epoch, CPU, small data), so high concurrency is safe and fast.

Output

Per config it prints PASS / FAIL with the elapsed seconds (and, on failure, the exit code + last error line), then a final <passed>/<total> passed summary. The command exits non-zero if any config fails — so it works as a CI gate.

Notes

• A --model smoke config must exist; the new-model scaffold generates it. If a hand-written model lacks one, create configs/runs/smoke_<module>.toml (extend ../base.toml + a small dataset + the model config, epochs = 1, device = "cpu").
• Graph / spatiotemporal models smoke against configs/datasets/cauair_ccaq_st.toml with task.mode = "spatiotemporal"; plain models against configs/datasets/smoke.toml.

Reference

See docs/en/scripts.md for the full tsf command reference.

Advanced-objective models

For models with pretrain(), keep smoke pretraining tiny (for example one epoch or a debug checkpoint). For models with optional objective branches, add explicit smoke coverage for non-default branches when feasible, e.g. an alternate align_loss_type or a different patching constraint.