iterative-refinement 开源项目推荐（35 个）

algorithmicsuperintelligence /openevolve

AlphaEvolve 的开源实现

gepa-ai /gepa

借助 AI 驱动的 Reflective Optimization 优化提示词、代码及更多内容

shyamsaktawat /OpenAlpha_Evolve

OpenAlpha_Evolve是一个开源Python框架，其灵感来自DeepMind的AlphaEvolve等自主编码代理的突破性研究。

yuval-alaluf /restyle-encoder

Official Implementation for "ReStyle: A Residual-Based StyleGAN Encoder via Iterative Refinement" (ICCV 2021) https://arxiv.org/abs/2104.02699

baidu-baige /LoongFlow

LoongFlow 是一个面向循环工程的专家级 Agent 框架。通过“计划-执行-总结”循环和结构化经验记忆，它使 AI 能够在复杂的软件工程、数学和机器学习任务中持续思考、执行、反思和进化。

tudelft /idnet

Lightweight Event-based Optical Flow Estimation via Iterative Deblurring

davitpapikyan /Probabilistic-Downscaling-of-Climate-Variables

Probabilistic Downscaling of Climate Variables Using Denoising Diffusion Probabilistic Models

MattMagg /ralph-wiggum-codex

Installable Codex skill for long-running autonomous coding refinement loops with validation and resumable state

ctkelley /MultiPrecisionArrays.jl

Mulitprecision Arrays

theflixerbox77 /ralph-wiggum-codex

Enable Codex to run objective-first autonomous loops with mandatory review and optional verification for long-running task completion

lt-asset /Latte

For our AAAI25 paper LATTE: Improving Latex Recognition for Tables and Formulae with Iterative Refinement by Nan Jiang, Shanchao Liang, Chengxiao Wang, Jiannan Wang, and Lin Tan

schuangs /hpl-ai-with-IR

Core part of HPL-AI implementation based on HPL-2.3. For the complete version of our HPL-AI benchmark, please go to the following site.

sudo-boo /optimizer-azeotropy

Minimizing Cost and Risk Using Bayesian Estimation and Gaussian Process Regression

HammonDDDDD /AIST-Business-analyst-agent

AI-powered business analyst agent with multi-agent collaboration workflow for iterative project specification refinement

israeldegasperi /gauntlet-loop

AI quality method for validating LLM output against predefined criteria. Includes a builder-critic workflow, evidence checks, cost limits and stopping conditions.

MarttiWu /codeopt

CodeOpt: A framework for optimizing code performance using Two-Stage Sampling, Few-Shot Learning, and Iterative Self-Reflection with support for Genetic Algorithm Inspired Chain-of-Thought (GA-COT).

stef4k /NFTM

Neural Field Turing Machines for iterative image inpainting and denoising. BDMA research project extending the NFTM pipeline (Malhotra & Seghouani, 2025) to CelebA-HQ and real camera noise (SIDD), with comparisons against U-Net, TV-L1 and DDPM baselines.

valorisa /Loop-Engineering-Post-it

Post-it technique et prompt-template prêt à l'emploi pour le Loop Engineering : boucle Producteur / Contrôleur Qualité en session unique avec un LLM (ChatGPT, Claude...). Document Markdown unique à copier-coller comme prompt système, avec plafond d'itérations, points de contrôle humains et critères de qualité mesurables, propres à la tâche.

chenxinghang-a /LoomLLM

Cost-aware multi-agent LLM framework with quality gates. Simple questions: 1 call. Code tasks: writer+reviewer loop.

umass-ml4ed /test_case_generation

Code for student-code guided test case generation for CS Education.