API Reference — torchrl 0.13 documentation

API Reference# torchrl.collectors package MultiCollector API Key Features Collection hooks Quick Example Removed legacy names Documentation Sections Collector Basics Single Node Collectors Collector Internals Evaluation Distributed Collectors Weight Synchronization Collectors and Replay Buffers torchrl.data package Key Features Quick Example CUDA prioritized replay buffers Documentation Sections Replay Buffers Storage Backends Sampling Strategies Datasets TensorSpec System Data layout: contiguous trajectories Trajectory boundary keys The replay buffer ndim arg and why it doesn’t multi-process well The buffer-to-collector handoff: complete-trajectory writes SliceSampler: variable-length contiguous slices Auto-discoverability for recurrent policies Legacy: split_trajectories Narrow canonicalization for recurrent inputs See also torchrl.envs package Key Features Quick Example Documentation Sections Environment API Vectorized and Parallel Environments Transforms Macro-control primitives Multi-agent Environments Library Wrappers Recorders IsaacLab Integration LLM Interface Key Components Quick Example Documentation Sections Policy Version Tracking Legacy Weight Updaters (Deprecated) Environments ChatEnv Task-Specific Environments Transforms Objectives GRPO, DAPO, CISPO SFT torchrl.modules package Key Features Quick Example Documentation Sections Actor Modules Exploration Strategies Value Networks and Critics Recurrent modules torchrl.modules.mcts package Robot Learning World Models and Model-Based RL Distribution Classes Inference Server Utilities and Helpers Recurrent state lifecycle torchrl.objectives package Key Features Quick Example Documentation Sections Common Components Value-Based Methods Policy Gradient Methods Actor-Critic Methods Offline RL Methods Multi-Agent Objectives Other Loss Modules Service Registry Overview Key Features Basic Usage Getting Started Service Registration Service Access Cross-Worker Visibility Namespace Isolation Cleanup Python Executor Service Overview Basic Usage Optional Service Usage Conditional Usage Pattern How It Works Performance Considerations Advanced Usage Multiple Service Configurations Custom Services API Reference Service Registry Python Executor Service Best Practices Examples See Also torchrl.trainers package Key Features Quick Example Documentation Sections Trainer Basics Loggers Training Hooks torchrl._utils package implement_for implement_for set_auto_unwrap_transformed_env set_auto_unwrap_transformed_env auto_unwrap_transformed_env auto_unwrap_transformed_env Memory profiling cuda_memory_stats reset_cuda_peak_stats cuda_memory_profile TorchRL Configuration System Quick Start with a Simple Example Configuration Categories and Groups More Complex Example: Parallel Environment with Transforms Getting Available Options Complete Training Example Running Experiments Basic Usage Hyperparameter Sweeps Custom Configuration Files Configuration Store Implementation Details Available Configuration Classes Base Classes Environment Configurations Environment Library Configurations Model and Network Configurations Transform Configurations Data Collection Configurations Replay Buffer and Storage Configurations Training and Optimization Configurations Trainer Hook Configurations Logging Configurations Creating Custom Configurations Best Practices Supported Algorithms Profiling collectors and envs Enabling profiling Driving a torch.profiler trace from the driver What gets instrumented Capturing a trace Multi-process and Ray Performance impact Glossary See also

API Reference#

Docs

Tutorials

Resources