Microsoft Trace: End-to-End Generative Optimization for AI Agents

a year ago

Trace is a new AutoDiff-like tool for training AI systems end-to-end with general feedback.
It generalizes back-propagation by capturing and propagating an AI system's execution trace.
Implemented as a PyTorch-like Python library, Trace allows users to write Python code directly and optimize parts using Trace primitives.
Supports various optimizers including OPRO, TextGrad, and OptoPrime (Trace's proposed algorithm).
Trace can be installed via pip and requires Python >= 3.9.
Features include computation graph visualization, LLM backend support (LiteLLM or AutoGen), and feedback-driven optimization.
Trace has been applied in various domains like parallel programming, NLP prompt optimization, and robotic arm control.
The project is open for contributions under Microsoft's Open Source Code of Conduct.

Hasty Briefsbeta