Pre-Training Intelligence¶
Commands to run before investing compute in a full training loop. Catch broken pipelines in seconds, establish baseline performance floors, and detect the most common cause of "too good to be true" results.
/turing:sanity: Pre-training sanity checks¶
Run a battery of fast checks before committing to a full training run. Catches wiring bugs in seconds: broken data loaders, misconfigured losses, dead gradients, and models that cannot memorize a single batch. The single-batch overfit test is the most powerful: if the model cannot fit one batch in 50 steps, something is fundamentally wrong.
Syntax: /turing:sanity [--quick] [--verbose] [--json]
Examples:
/turing:baseline: Automatic baseline generation¶
Generate trivial baselines so you always know if your model is meaningfully better than simple approaches. Produces random, majority/mean, linear, and k-NN baselines in 60 seconds, evaluated with the same protocol as real experiments. Satisfies the "baseline comparison" check in /turing:audit.
Syntax: /turing:baseline [--methods all|simple|linear] [--data data.npz] [--json]
Examples:
/turing:baseline # All baselines
/turing:baseline --methods simple # Just random + majority
/turing:baseline --data data/processed.npz # With actual data
/turing:leak: Data leakage detection¶
Actively probe for data leakage, the number one cause of "too good to be true" results. Checks feature-target correlation, single-feature predictiveness (with --deep), and train/test overlap via hash-based deduplication. Issues CLEAN, SUSPICIOUS, or LEAKAGE DETECTED verdicts. Satisfies the "data leakage" check in /turing:audit.
Syntax: /turing:leak [--deep] [--features "feat_1,feat_2"] [--json]
Examples: