Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
[ICML 2026] Let LLMs invent and evolve languages for efficient reasoning.
| Date | Stars |
|---|---|
| 2026-07-31 | 378 |
| 2026-08-06 | 378 |
Today
— stars today
This week
— stars this week
This month
— stars this month
Momentum
0.0
growth rate 0.00%/day
<div align="center">
<h1>Machine Dialectology (MDia)</h1>
<p><strong>Reusable symbolic protocols for heterogeneous language-model societies.</strong></p>
<p>
MDia studies how LLMs invent reasoning dialects, teach through them, resist them, and code-switch across tasks.
</p>
<p>
<a href="https://github.com/pzqpzq/LSF_MDia/actions/workflows/ci.yml"><img src="https://github.com/pzqpzq/LSF_MDia/actions/workflows/ci.yml/badge.svg" alt="CI status" /></a>
<a href="https://www.python.org/"><img src="https://img.shields.io/badge/Python-3.10%2B-3776AB?logo=python&logoColor=white" alt="Python 3.10 or newer" /></a>
<img src="https://img.shields.io/badge/MDia-v1.0.0-0E7490" alt="MDia version 1.0.0" />
<a href="LICENSE"><img src="https://img.shields.io/badge/License-MIT-2EA44F" alt="MIT License" /></a>
</p>
<p>
<a href="https://icml.cc/virtual/2026/poster/61557"><img src="https://img.shields.io/badge/CLSR-ICML%202026-6F42C1" alt="CLSR at ICML 2026" /></a>
<a href="https://arxiv.org/abs/2606.29354"><img src="https://img.shields.io/badge/CLSR-arXiv%3A2606.29354-B31B1B?logo=arxiv&logoColor=white" alt="CLSR paper on arXiv" /></a>
<img src="https://img.shields.io/badge/Interface-black--box%20LLMs-111827" alt="Black-box LLM compatible" />
</p>
<p>
<a href="#overview">Overview</a> ·
<a href="#quickstart">Quickstart</a> ·
<a href="#canonical-pipeline">Pipeline</a> ·
<a href="#routing-semantics">Routing</a> ·
<a href="#research-snapshot">Research</a> ·
<a href="#reproducibility-by-construction">Reproducibility</a> ·
<a href="#documentation">Documentation</a> ·
<a href="#citation">Citation</a>
</p>
</div>
<p align="center">
<a href="assets/readme/pipeline-Jul15-v0.png">
<img src="assets/readme/pipeline-Jul15-v0.png" width="100%" alt="Machine Dialectology main figure. Panel A motivates machine-oriented reasoning languages by contrasting verbose human-readable chain-of-thought with the accuracy-token frontier. Panel B represents reusable Language Symbolism Frameworks as dialect cards with symbols, operators, state variables, rules, and answer contracts. Panel C shows asymmetric speaker-listener transfer, model social profiles, and profile-aware routing. Panel D summarizes a 100-rule bank and supported machine-sociolinguistic regularities." />
</a>
</p>
<p align="center">
<sub><strong>Machine Dialectology: from verbose reasoning traces to a measurable ecology of machine dialects.</strong> Click the figure to inspect the full-resolution manuscript asset.</sub>
</p>
---
## Overview
Large language models increasingly operate as heterogeneous systems of solvers, routers, critics, and tool users. Yet their intermediate reasoning is still usually expressed as human-facing prose. Natural language is readable and flexible, but it can spend generated tokens on discourse, pedagogy, and surface conventions that a machine listener may not need. Simply forcing shorter answers is not enough: aggressive compression can remove variables, evidence links, parse commitments, verification state, or output contracts that the listener *does* need.
**Machine Dialectology (MDia)** treats an intermediate trace as a message produced by one machine component and interpreted by another. Speaker models create persistent **Language Symbolism Frameworks (LSFs)**—compact dialect cards with symbols, grammar, reusable operators, validity rules, answer contracts, provenance, and empirical transfer profiles. Listener models execute those dialects, and MDia routes them under explicit accuracy, token, parseability, and reliability constraints.
> **Central thesis**
>
> A machine dialect is not intrinsically strong or weak. Its value is relational:
>
> **dialect utility = f(speaker, listener, task, route, budget).**
> [!IMPORTANT]
> This repository is **MDia-first**. **CLSR** is the homogeneous-agent, concrete-LSF-routing special case configured through [`configs/clsr.yaml`](configs/clsr.yaml). The supported implementation lives under [`sExcerpt of 23,199 characters
Read on GitHubWould you bet a product on this? Bounded 0–100 and slow moving.
matched fp:0a53d7cde82bb669, topic:multi-agent, topic:multi-agent-systems
matched fp:0a53d7cde82bb669, topic:llm