| title: SGLabs | |
| emoji: 🐜 | |
| colorFrom: purple | |
| colorTo: blue | |
| sdk: static | |
| pinned: false | |
| # SGLabs | |
| *An independent open-model lab — capable AI on the hardware you actually own.* | |
| SGLabs builds and shares open-model work, from efficient inference to tooling and the occasional experiment that earns its keep. The throughline: capable AI shouldn't require a datacenter. We build for consumer and unified-memory hardware (AMD, Apple Silicon, and friends), work in the open, and ship reproducible recipes — not just weights. | |
| ## What we're working on | |
| - **Pym** — our quantization line: aggressive, calibrated mixed-precision GGUFs that shrink large models to a fraction of their size while keeping their edge (and their speed). See the Pym collection below. | |
| - **More on the way** — inference tooling, agent experiments, and whatever else is worth building. Watch this space. | |
| ## How we work | |
| - **Honest benchmarks** — quality *and* speed, with the noise flagged rather than hidden. | |
| - **Reproducible** — every release ships its method, so you can rebuild it. | |
| - **Runs-anywhere** — if it fits in unified memory, it should run. | |
| ## Credits | |
| Our Pym quants were inspired by [EschaLabs](https://huggingface.co/EschaLabs)' *Escha W2*, built independently in that same spirit. 🎩 | |
| --- | |
| Maintained by [@syntaxgeek](https://huggingface.co/syntaxgeek). | |