Contents Menu Expand Light mode Dark mode Auto light/dark, in light mode Auto light/dark, in dark mode Skip to content
AgileRL Documentation
Light Logo Dark Logo
AgileRL Documentation

Introduction

  • Getting Started
  • Releases

Arena

  • Arena Client

Training

  • Evolutionary Hyperparameter Optimization
  • Off-Policy Training
  • On-Policy Training
  • Partially Observable Markov Decision Processes (POMDPs)
  • Offline Training
  • Multi-Agent Training
  • LLM Fine-Tuning
    • Memory-efficient log-probabilities
    • Quantization
    • Saving and Loading LLM Checkpoints
  • Contextual Multi-Armed Bandits
  • Distributed Training
  • Trainers
  • Evolvable Neural Networks
  • Creating Custom Algorithms
  • Debugging Reinforcement Learning

Observability

  • Metrics and Logging

Tutorials

  • Gymnasium Single-agent
    • Acrobot with PPO
    • Lunar Lander with TD3
    • Cartpole with Rainbow DQN
    • Partially Observable Pendulum-v1 with Recurrent PPO
  • PettingZoo Multi-agent
    • Self-Play Connect4 with DQN + Curriculum Learning
    • Space Invaders with MADDPG
    • Speaker-Listener with MATD3
  • Hierarchical Skills
  • LLM Fine-Tuning
    • LLM Reasoning Tutorial
    • LLM Fine-Tuning with HPO
    • Multi-turn Fine-Tuning with LLMPPO, LLMREINFORCE, and GRPO
    • LLM Fine-Tuning with SFT and DPO
  • Contextual Multi-arm Bandits
    • Iris with NeuralUCB
    • PenDigits with NeuralTS
  • Creating Custom Networks
    • Building a Dueling Distributional Q Network
    • Integrating Architecture Mutations Into SimBa
  • Training on Arena
    • PPO on a Custom Gym Environment
    • GRPO on GSM8K

API

  • Algorithms
    • EvolvableAlgorithm
    • Algorithms Mutations Registry
    • OptimizerWrapper
    • Conservative Q-Learning (CQL)
    • Deep Deterministic Policy Gradient (DDPG)
    • Deep Q-Learning (DQN)
    • Rainbow DQN
    • Implicit Language Q-Learning (ILQL)
    • Proximal Policy Optimization (PPO)
    • Twin Delayed Deep Deterministic Policy Gradient (TD3)
    • Independent Proximal Policy Optimization (IPPO)
    • Multi-Agent Deep Deterministic Policy Gradient (MADDPG)
    • Multi-Agent Twin-Delayed Deep Deterministic Policy Gradient (MATD3)
    • Neural Contextual Bandits with UCB-based Exploration (NeuralUCB)
    • Neural Contextual Bandits with Thompson Sampling (NeuralTS)
    • Group Relative Policy Optimization (GRPO)
    • Clipped Importance Sampling Policy Optimization (CISPO)
    • Group Sequence Policy Optimization (GSPO)
    • Direct Preference Optimization (DPO)
    • Supervised Fine-Tuning (SFT)
    • LLM Proximal Policy Optimization (LLM PPO)
    • LLM REINFORCE
    • Supervised Fine-Tuning (SFT)
  • Arena
    • ArenaClient
    • Arena manifest models
    • Stream
    • Output
    • Inference
    • Authentication
    • Exceptions
  • Components
    • Experience Replay Buffer
    • On-Policy Rollout Buffer
    • Segment Trees
    • Data Structures and Utilities
    • Experience Sampler
  • Hyperparameter Optimization
    • Mutation
    • Tournament Selection
  • Logger
  • Metrics
  • Models
    • Training Manifest (local training)
    • Algorithm Specifications
    • Environment Specifications
    • Network Specifications
    • Training & Replay Buffer Specs
    • HPO Specifications
  • Modules
    • EvolvableModule
    • EvolvableWrapper
    • ModuleDict
    • Evolvable Multi-layer Perceptron (MLP)
    • Evolvable Convolutional Neural Network (CNN)
    • Evolvable Long Short-Term Memory (LSTM)
    • Evolvable Multi-Input Neural Network (Dict / Tuple Observations)
    • Evolvable SimBa
    • Evolvable ResNet
    • DummyEvolvable
    • Evolvable GPT
    • Evolvable BERT
    • Custom Activation Functions
  • Networks
    • EvolvableNetwork
    • QNetwork
    • RainbowQNetwork
    • ContinuousQNetwork
    • EvolvableDistribution
    • DeterministicActor
    • StochasticActor
    • ValueNetwork
  • Population
  • Training
  • Rollouts
    • On-Policy Rollout Functions
  • Utils
    • Arena utilities
    • General Utils
    • Algorithm Utils
    • Cache Utils
    • Evolvable Networks Utils
    • ILQL Utils
    • Log Utils
    • Minari Utils
    • Probe Environments
    • LLM Probe Environments
    • Torch Utils
    • LLM Utils
  • Vector
    • Dummy Vectorized Environments
    • Petting Zoo Async Vector Environment
    • Petting Zoo Vector Base Class
  • Wrappers
    • AgentWrapper
    • RSNorm
    • AsyncAgentsWrapper
    • LLM environments
    • Make Evolvable
    • Skill
    • BanditEnv
    • Petting Zoo Wrapper

Development

  • GitHub
  • Discord
  • Contribute to AgileRL
Back to top
View this page
Edit this page

ModelsΒΆ

Pydantic specification models used by trainers and manifests to declaratively configure local training runs.

Arena cloud submission uses a separate, lightweight model layer in agilerl.arena.models (agilerl-arena package).

  • Training Manifest (local training)
  • Algorithm Specifications
  • Environment Specifications
  • Network Specifications
  • Training & Replay Buffer Specs
  • HPO Specifications
Next
Training Manifest (local training)
Previous
Metrics
Copyright © 2023, AgileRL
Made with Sphinx and @pradyunsg's Furo