README.md

metadata

title: TinyModels
emoji: ⚡
colorFrom: red
colorTo: blue
sdk: gradio
pinned: false
license: mit

TinyModels 🤏⚡

Big intelligence. Tiny Parameters.

We build small, fast, production-ready models that punch way above their weight.
No bloat. No nonsense. Just models that work — on your laptop, your phone, your API.

---

🧠 Philosophy

Most AI labs race to make models bigger.
We race to make them smaller without losing what matters.

Every model we ship follows three rules:

1. Fits on a free GPU          → deployable by anyone
2. Beats models twice its size → efficiency over parameters  
3. Ships with clean docs       → actually usable in production

---

⚡ Why Tiny?

	Big Models	TinyModels
Cost	$$$ per inference	Free / near-free
Latency	500ms–2s	20–80ms
Deployment	Needs A100	Runs on T4, CPU, phone
Accessibility	Labs only	Everyone
Carbon	🔥 High	🌱 Low

---

🤝 Open to Everyone

TinyModels is a free, open community — anyone can join, contribute models, datasets, or ideas.

No gatekeeping. No paywalls. Just good models, open to all.

👉 Want to contribute? Follow the org and reach out!

---

🛠️ Built With

Every model is trained on Google Colab free tier to prove it doesn't take a supercomputer to build something great.

---

🗺️ Roadmap

✅ CORTYX v2    — 17-label toxicity classifier
🔜 MCLM v1     — 65-label multi-domain classifier  
📅 CORTYX v3   — improved hate_speech, profanity, self_harm
📅 TinyEmbed   — 32M sentence embeddings
📅 TinyNER     — named entity recognition <50M
📅 TinyIntent  — ultra-fast intent classifier for chatbots

---

🚀 Quick Start

All TinyModels work the same way:

pip install transformers torch sentencepiece huggingface_hub

from huggingface_hub import hf_hub_download
# Load any model in minutes — see individual model cards for full usage

---

Built with ❤️ by the community · Free forever · Open to all

Tiny models. Huge ambitions.