Xylolabs Inc.

Edge Physical AI that catches equipment faults and runs robots.

Sensors, edge AI and robots for port cranes, power plants and factories.

On Hugging Face

Open LLMs quantized for Tenstorrent Tensix NPUs. Each model's linear layers are quantized to TT-NN's block floating point formats, bfloat4_b (BFP4) or bfloat8_b (BFP8), with GPTQ, so TT-NN packs the weights as they are. Every card lists the formats per tensor class and the model's KL and answer check against bf16, beside llama.cpp's Q4_K_M of the same model.

The code that makes them is tensix-bfp-gptq (Apache-2.0).

Collections: one per model family.

What we build

Started as signal research at Seoul National University in 2019, incorporated in December 2025.

Not affiliated with Tenstorrent.