Test

NeuroMF Test Runner

Run pytest for the neuromf project using the neuromf conda environment.

Instructions

1. Run the tests specified by $ARGUMENTS. If no arguments given, run the fast suite (-m "not slow").
2. Command format:

   ~/.conda/envs/neuromf/bin/python -m pytest $ARGUMENTS -v --tb=short
   

3. If tests fail, analyze the failures and suggest fixes.
4. If all tests pass, report the summary.

Test Suite Structure

Marker Reference

• slow — Tests that construct UNet/LatentMeanFlow/VAE models (1-3s each for init)
• phase0-phase6 — Phase-specific tests
• critical — Must pass for phase gate to open
• informational — Nice to have, not gating

Known Issues

• test_P5_T9_feature_extractor_mock — Pre-existing mock bug (MagicMock vs Tensor), does NOT indicate real code failure
• Phase 0 VAE tests (test_maisi_vae_wrapper.py) require real VAE weights + GPU — expected to fail locally
• Phase 1 test_P1_T7_round_trip_ssim requires real data + GPU

Examples

# Default: fast suite (recommended after code changes)
/test

# Specific file
/test tests/test_meanflow_pipeline.py

# Phase 4 critical tests only
/test -m "phase4 and critical"

# All slow tests (before training submission)
/test -m "slow"

# Specific test by name
/test -k "test_P3_T4_meanflow_loss_finite_positive"

Converted and distributed by TomeVault — claim your Tome and manage your conversions.