SupraLabs_SupraLabs
Research
We're democratizing AI by training high-performance models on consumer hardware. 100% Open Source. 0% Slop.
2026-06-02
Dynamic Curriculum Shifting
Information Routing — Curriculum Shifting
2026-06-02
More Epochs vs. More Data for SLMs
Satiating the Latent Space: Unique Tokens vs. Cycles
2026-06-02
The optimal vocab size
The Embedding Bottleneck: Optimal Vocab Scales
2026-06-02
Depth vs. Width
Hidden Topology: Depth vs. Width Scaling for SLMs
2026-05-31
Is One Epoch Really All You Need for SLMs?
Researching the best epoch count for SLMs
2026-05-30
5M SLM Data-Mix Benchmarks
Researching the best dataset(s) for SLMs