conflit

Layered YAML configuration for Python. Compose multiple config files, patch nested keys without repeating unchanged ones, accumulate lists across layers, and optionally validate the result with Pydantic.

TLDR

# base.yaml
model:
  num_layers: 6
  hidden_dim: 512
features:
  - mixed_precision

# gpu_large.yaml — only what changes; hidden_dim is preserved
model:
  num_layers: 12
features: !append
  - distributed_training

# experiment.yaml
_compose:
  - base.yaml
  - gpu_large.yaml
run_name: orion-v1-large

from pathlib import Path
from conflit import load

cfg = load(Path("experiment.yaml"))
# {"model": {"num_layers": 12, "hidden_dim": 512},
#  "features": ["mixed_precision", "distributed_training"],
#  "run_name": "orion-v1-large"}

cfg = load(Path("experiment.yaml"), schema=OrionConfig)
cfg.model.num_layers  # 12 — typed, validated

See examples/ for the full walkthrough and docs/ for the configuration model reference.

Install

uv sync
poe  # list available tasks

Contributing

See CONTRIBUTING.md.