ThetaEvolve: A Test-Time Learning Framework
burny_tech · x · 2026-07-11
ThetaEvolve is an open-source framework designed for test-time learning on open-ended optimization problems. It combines in-context learning and reinforcement learning to continuously improve model performance.
Main Claims
- The author calls it the first "evolutionary" framework enabling small open-source models (like DeepSeek-R1-0528-Qwen3-8B) to achieve state-of-the-art results on open problems
- Tasks mentioned in the paper/project include circle packing and first auto-correlation inequality
- Design highlights:
- A single LLM
- A large program database to enhance exploration
- Batch sampling to boost throughput
- Lazy penalties to suppress stagnant outputs
- Optional reward shaping to stabilize training signals
Core Focus
Rather than introducing a specific model, this work presents a general method for models to "learn by doing" during the test phase. It is particularly well-suited for open-ended optimization problems requiring a combination of search, reasoning, and program generation.
Related event: ThetaEvolve: Open-Sourcing Test-Time Learning for Small Models(3 posts)→
More from coding & agent
- FactoryAI gave back its first millions, then shipped Droid CLI two years later — matanSF · 2026-07-22
- Devin Outposts aims to run AI agents on any machine, from Mac minis to Kubernetes clusters — blaizedsouza · 2026-07-22
- Hermes Agent Refactoring Proposal: Decoupling via Event Bus and Monorepo Slicing — Promptmethus · 2026-07-22
- ty now reads Pydantic config keywords and field metadata — charliermarsh · 2026-07-22
- Pensar Launches AI Security Agent to Autonomously Discover and Patch 0-Days — andriy_mulyar · 2026-07-22
- ty adds first-class Pydantic support, including strict and lax field handling — charliermarsh · 2026-07-22