incocreativedev

incocreativedev / tessera-core

Public

An activation-based protocol for AI-to-AI knowledge transfer across architectures

github.comincocreativedevtessera-core activation-based ai-protocol cross-architecture knowledge-transfer machine-learning
11
0
100% credibility
Found Feb 25, 2026 at 11 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

Tessera-core is an open-source Python library that enables trained AI models to transfer their learned knowledge to untrained models of different architectures using activation patterns and a shared hub space.

How It Works

1
🔍 Discover Tessera

You hear about Tessera, a simple way for smart AIs to teach each other what they've learned, saving time and effort.

2
🧠 Prepare your AIs

Gather your trained smart AI that knows a task well and a fresh AI ready to learn quickly.

3
✨ Share the smarts

Run a quick knowledge transfer so the fresh AI picks up the trained one's skills through a shared thinking space.

4
💾 Save the knowledge

Capture the shared insights in a compact file that holds all the learned patterns securely.

5
🚀 Apply to your new AI

Load the knowledge file into the fresh AI, instantly boosting its abilities without full retraining.

✅ Smarter AI ready

Your new AI now performs like it was trained from scratch, faster and across different designs—success!

Sign up to see the full architecture

4 more

Sign Up Free

Star Growth

See how this repo grew from 11 to 11 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is tessera-core?

Tessera-core implements an activation-based protocol for AI-to-AI knowledge transfer across architectures in Python with PyTorch. It captures a trained model's learned behaviors as compact tokens via a shared hub space, letting you bootstrap untrained receivers—even Transformers to LSTMs—without full retraining. Pip-install and execute transfers on CPU in under a minute, producing verifiable binaries with drift scores and privacy budgets.

Why is it gaining traction?

Benchmarks across Transformer, MLP, Conv1D, and LSTM families deliver positive accuracy deltas in 40% of cross-arch pairs, proving real viability where weight-sharing fails. Tokens shrink to 4KB with quantization (FLOAT16/INT8), include audit trails for compliance, and support swarms for collaborative learning. CLI tools inspect/validate files, plus Docker/K8s deploys and a REST API make it instantly usable.

Who should use this?

AI researchers distilling knowledge from large lab models to edge variants on ARM or RISC-V. Teams running federated swarms where diverse client architectures exchange updates. Engineers bridging prototype-to-production gaps without dataset sharing.

Verdict

Early alpha (10 stars, 1.0% credibility) with excellent docs and benchmarks, but test thoroughly before prod. Strong pick for cross-arch experiments; monitor for stability as it matures.

(178 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.