SLM Lab Files

Major modernization release that updates SLM-Lab from OpenAI Gym to Gymnasium, migrates to modern Python tooling (uv), and validates all algorithms across 70+ environments.

Key Changes

• Gymnasium migration with correct terminated/truncated handling
• Modern toolchain: uv + pyproject.toml, Python 3.12+, PyTorch 2.8+
• Simplified specs: No more body section or array wrappers
• Complete benchmark validation: 7 algorithms × 4 environment categories
• Cloud training support via dstack + HuggingFace

Benchmark Results

Atari benchmarks use ALE v5 with sticky actions (repeat_action_probability=0.25), following Machado et al. (2018) research best practices.

Breaking Changes

• Environment names: CartPole-v0 → CartPole-v1, PongNoFrameskip-v4 → ALE/Pong-v5
• Spec format simplified: agent: [{...}] → agent: {...}
• body section removed, attributes moved to agent
• Roboschool → MuJoCo (RoboschoolHopper-v1 → Hopper-v5)

Quick Start

:::bash
# Install
uv sync && uv tool install --editable .

# Run
slm-lab run spec.json spec_name train

Book Readers

For exact code from Foundations of Deep Reinforcement Learning, use:

:::bash
git checkout v4.1.1

See CHANGELOG.md (github.com) for full details.