CLI Reference¶

Primary entrypoint:

source .venv/bin/activate && PYTHONPATH=src python3 run_alphafold_mlx.py \
  --input <input.json> \
  --output_dir <output_dir>

Required Arguments¶

• --input PATH: Input JSON file.
• --output_dir DIR: Output directory for generated artifacts.

Optional Arguments¶

• --model_dir DIR: Weights directory. Default resolution order:
• AF3_WEIGHTS_DIR
• ~/.alphafold3/weights/model
• weights/model
• --num_samples N: Number of samples (default: 5).
• --diffusion_steps N: Diffusion steps (default: 200).
• --seed N: Random seed (optional).
• --precision {float32,float16,bfloat16}: Override precision.
• --verbose, -v: Enable stage-level logging and timing output.
• --no-overwrite: Prevent output overwrite.
• --run_data_pipeline: Enable MSA/template search before inference.
• --db_dir DIR: Database root override (AF3_DB_DIR fallback otherwise).
• --msa_cache_dir DIR: Cache directory for MSA/template reuse.
• --max_template_date YYYY-MM-DD: Template date cutoff (default: 2021-09-30).
• --max_tokens N: Limit max token bucket for memory control.
• --restraints PATH: Standalone restraints JSON file (mutually exclusive with inline restraints in input JSON).

Examples¶

Fast smoke test¶

source .venv/bin/activate && PYTHONPATH=src python3 run_alphafold_mlx.py \
  --input examples/desi1_monomer.json \
  --output_dir output/smoke \
  --num_samples 1 \
  --diffusion_steps 20 \
  --verbose

Full pipeline mode¶

source .venv/bin/activate && PYTHONPATH=src python3 run_alphafold_mlx.py \
  --input examples/desi1_monomer.json \
  --output_dir output/full \
  --run_data_pipeline \
  --db_dir /path/to/public_databases

Restraint-guided run¶

source .venv/bin/activate && PYTHONPATH=src python3 run_alphafold_mlx.py \
  --input examples/desi1_dimer.json \
  --restraints examples/restraints/dimer_contacts.json \
  --output_dir output/restrained

Exit Codes¶

• 0: Success
• 1: Input/resource/inference/runtime error
• 130: Interrupted (SIGINT, typically Ctrl+C)