Train agents with TorchForge
torchforge-rl-trainingskillsetup L4★9,423
Orchestra-Research/AI-Research-SKILLs ↗What it does
Optimize RL training with PyTorch-native abstractions
Best for
PyTorch-native RL training with hardware acceleration and custom loss functions.
Inputs
- · Policy model
- · Reward signal
- · Experience replay buffer
- · Optimizer config
Outputs
- · Optimized policy
- · Training logs
- · Performance metrics
Requires
- · torchforge
- · torch>=2.0
- · ray (optional)
Preconditions
- · PyTorch model defined
- · Reward function computable
- · GPU available
Failure modes
- · Gradient explosion
- · Dead gradients
- · Experience replay stale
- · Batch too large
Trust signals
- · PyTorch ecosystem
- · Direct tensor operations