xLAM
Basic Information
xLAM is a research repository that presents a family of Large Action Models and accompanying tooling to support agentic systems research. It aggregates and standardizes multi-turn agent trajectories from diverse environments into unified datasets and provides a generic data loader and training pipeline for agent training. The project includes pretrained and fine-tuned models optimized for multi-turn conversation and function-calling across a range of parameter scales, model naming conventions, and release artifacts on model hubs. The repository also contains APIGen-MT, an agentic pipeline for multi-turn data generation via simulated agent-human interplay, and ActionStudio, a lightweight framework for preparing agentic data and training large action models. The codebase includes examples, training configs, deployment instructions for Transformers and vLLM, benchmark results, licensing information, and notes that data and models are released for research purposes only.
Links
App Details
-
$$$$ -
$$$$