SupraLabs

non-profit

AI & ML interests

Train, finetune and explore small models with good results to revolutionize small AI models by making them accessible to everyone

Recent Activity

QyrouNnet-AI published a model about 3 hours ago

SupraLabs/supra-title-FFT-preview

QyrouNnet-AI published a dataset about 5 hours ago

SupraLabs/supra-wild-titles-130k

QyrouNnet-AI updated a dataset about 5 hours ago

SupraLabs/supra-wild-titles-130k

View all activity

Organization Card

Community About org cards

Welcome to SupraLabs!

SupraLabs is an open-source AI research organization focused on Small Language Models (SLMs), datasets, and efficient AI systems designed for accessibility and real-world deployment. Our mission is to advance efficient and accessible AI by developing open-source Small Language Models, datasets, and tools that can run on a wide range of hardware.

Our Creations...

At SupraLabs, innovation is our core skills... Here is what we have made.

Models

Model	Description
Supra-1.5 Instruct (EXP)	Instruction-tuned experimental release. Demo
Supra-1.0 Reasoning (50M)	First-generation reasoning model. Demo
Supra-1.5 Base (EXP)	Base model (v1.5 · Latest).
Supra Titles (EXP)	First-generation chat title generation model.
All Models from SupraLabs	View all the AI models from SupraLabs

Datasets

Dataset	Description
150K Chat Titles (Unfiltered)	150K samples of unfiltered chat title data.
115K Chat Titles (Filtered)	115K filtered chat titles generated from the first user message.
500× Supra-1.0 Reasoning	500 reasoning traces from Qwen-3-1.7B used to train the first-generation reasoning model.

Blogs & Research

Resource	Description
Blog	Read our latest blog posts.
Research	Explore our research publications and findings.

Our Naming Conventions...

At SupraLabs, we use a strict naming system for our models to avoid confusion!

EXP (Experimental) — Early-stage release intended for testing and development. These models may be unstable, incomplete, or subject to significant changes.
Preview — A near-final release that provides an early look at the intended final model. Minor tuning, optimizations, or improvements may still be pending.
No Tag — Stable final release. Recommended for general use and represents the intended production version.