Welcome to SupraLabs!

Who we are

We are @AxionLab-official, @LH-Tech-AI, and @Harley-ml. We are creating small open-source models for everyone.

What we do

We train, finetune, and explore small models. Our goal is to revolutionize small AI models by making them accessible to everyone!

What we do NOT do

We are not making bad (or we try not to!) models and we try to fully open source our models and code. Some models may be fully opensourced, while others might not.

Models

• Supra Mini 0.1M: Trained on Kaggle 2xT4, 100k parameters, compared to models 10x it size
• Supra Mini v2 0.1M: the second version of the Supra Mini series.
• Supra Mini v3 0.5M: the third version of the Supra Mini series.
• Supra Mini v4 2M: the fourth version of the Supra Mini series. Improved. More powerful. With context understanding.
• Supra Mini v5 8M: the fifth version of the Supra Mini series. A huge token-eater monster compared to its siblings.
• MicroSupra 1k: Trained on GTX 750 Ti 4GB, a scaling laws experiment.
• StorySupra-10M: Trained on RTX 5060 Ti 16GB for 10 minutes, coherent.
• DistillSupra-0.2M: Trained on GTX 750 Ti 4GB for 30 minutes, still incoherent, but the first step for distillation research.
• More Coming Soon! Come Back later!

Competing with other creators

We are competing with @CompactAI-O and @LH-Tech-AI (we know it's funny to compete against your own founder, but anyway 🤣😂).
See all of our and our copetitors tiny models here: https://lh-tech.de/ai/compare-tiny-models.html

Future models

• Supra-10M: Base, Chat, Reasoning - Trained on RTX 5060 Ti 16GB, with Nvidia technologies and CUDA
• Supra-1M: Base, Chat, Reasoning - Trained on GTX 750Ti 4GB, with Nvidia Technologies and optimizations
• Supra-50M: Base, Chat, Reasoning, Coding - Trained on a knot between RTX 5060 Ti 16GB and GTX 750 Ti 4GB, Pushing charges to train the best model to you!

Hardware

• RTX 5060 Ti 16GB (LH-Tech AI)
• GTX 750Ti 4GB (AxionLab)
• RTX 2060 6GB (Harley-ML)

Blog

https://huggingface.co/spaces/SupraLabs/Blog

Feedback and Support

Feedback and support welcomed. Feel free to ask to join our organization if you want!

Note

Some content, such as our blogs or readmes, may be created with the help of AI because not all of us have strong English skills.