agent-lightning
The absolute trainer to light up AI agents.
Master testing strategies for agent training pipelines in Agent Lightning. Explore component wiring, strategy selection, and runtime contracts for robust AI agent development.
How to Implement Checkpointing and Resume Training in Agent-LightningLearn how to implement checkpointing and resume training in Agent-Lightning. Automatically save model weights and training state to resume from the exact batch where you left off.
Worker Management and Resource Allocation in Agent-Lightning: A Complete Technical GuideMaster worker management and resource allocation in Agent-Lightning. Learn how this guide details scalable agent execution with store-driven APIs and lifecycle management.
Algorithm Selection Criteria: VERL vs APO vs SFT in Agent-LightningSelect VERL for direct weight modification, APO for prompt optimization without retraining, or SFT for supervised fine-tuning in Agent-Lightning. Learn algorithm selection criteria.
Migrating from Legacy Runner to the New Architecture in Agent-Lightning: A Complete GuideUpgrade Agent Lightning from legacy runner to new architecture. Switch to LitAgentRunner and LightningStore for async execution, heartbeats, and unified storage. Read our complete guide.
Security Considerations for the LLM Proxy in Agent-Lightning: Risks and MitigationsSecure your LLM proxy in agent-lightning. Learn essential security considerations, default risks, and mitigation strategies for your OpenAI-compatible API endpoint.
Optimizing Performance for Large-Scale Agent Training with Agent-LightningBoost large-scale agent training performance with Agent-Lightning. Optimize with ClientServerExecutionStrategy, persist rollouts in MongoLightningStore, and configure environment variables to scale effectively.
Implementing Custom Reward Functions in Agent Lightning: A Complete GuideLearn to implement custom reward functions in Agent Lightning using emit_reward() and find_final_reward(). This guide provides step-by-step instructions for integrating scalar and multi-dimensional rewards.
Debugging Training Failures and Tracing Issues in Agent-Lightning: A Complete GuideDebug training failures and trace issues in Agent Lightning with a three-layer diagnostic approach. Isolate crashes, missing traces, and stalled RL loops quickly. Learn more.
Configuring Environment Variables for Agent-Lightning Production DeploymentsConfigure environment variables for agent-lightning production deployments to manage GPU and CPU nodes and connect to persistent stores like MongoDB. Learn how in our guide.
Ensuring Thread-Safety and Async Safety in Distributed Training with Agent-LightningLearn how Agent-Lightning ensures thread-safety and async safety in distributed training using LightningStoreThreaded, cooperative cancellation, and process-isolated Ray actors.
Handling Large-Scale Span Storage and Eviction Policies in Agent-LightningLearn how Agent-Lightning handles large scale span storage and eviction policies. Prevent OOM errors with automatic memory management and configurable thresholds for optimal performance.
Have a question about this repo?
These articles cover the highlights, but your codebase questions are specific. Give your agent direct access to the source. Share this with your agent to get started:
curl -s "https://instagit.com/install.md" 
Maintain an open-source project? Get it listed too →