UIST Labs

A small, independent AI/ML lab. We build tooling and publish model checkpoints - and we hold them to a stated bar before they ship.

What we do

Certified quantization. We quantize open models and publish a checkpoint only after it clears its stated accuracy bar against its own full-precision baseline - quantized and baseline measured on the same harness and settings, with the measured deltas published on its model card. A checkpoint that misses its bar doesn't ship. The name of the pipeline behind it says the idea plainly: an assay is the test that certifies a refined metal's purity.

Runs on the hardware you already have. Current library: NVFP4 (A16, weight-only), running via vLLM's Marlin FP4 kernel with no Blackwell GPU required. Validated on Ada-generation cards (sm_89: L4, RTX 4090); expected on Ampere and Hopper, not yet validated; Turing (sm_75) is unsupported - it produces incorrect output on current vLLM. The ~2.5x smaller weights free GPU memory for a larger KV cache.

The library

Start at Certified Quantized Models - every published checkpoint, each with its gate verdict and measured deltas on its model card. Family collections (Qwen, DeepSeek R1) sit alongside on the org page.

Tooling

assay - the open-source (Apache-2.0) quantize -> benchmark -> gate -> publish pipeline behind our releases. Every checkpoint we publish was produced by it, and passed its gate.

Services

The same pipeline runs on your models and your hardware targets, with benchmarks matched to your acceptance criteria: certified quantization with stated bars, measured deltas, and delivered-or-withheld results. Reach us at contact@uist-labs.com.

How we work

• Security first. We don't trade security for convenience without saying so.
• Trust, but verify. Vendor claims are a starting point; the measured numbers are the product.
• Standard tools, honest docs. We prefer well-understood approaches, and we document the trade-offs - including what a release deliberately doesn't do.