A2Genesis is an independent software studio in Bochum, Germany. We build custom applications and run language models on our own hardware.

Here we publish quantized checkpoints of open-weight models. The quantization recipe, the calibration data and the serving configuration are documented in every model card, so a result can be reproduced rather than taken on trust.

Models

• Qwen3.8-27B-NVFP4 — NVFP4 (W4A16) quantization of Qwen/Qwen3.8-27B, produced with NVIDIA TensorRT Model Optimizer 0.45. About 21 GB instead of 55 GB, so it fits on a single GPU. Apache 2.0.

Focus

• Post-training quantization (NVFP4, FP8, NVIDIA ModelOpt)
• Self-hosted inference with vLLM and SGLang
• AI-native application development

a2genesis.de

Not affiliated with or endorsed by NVIDIA or the Qwen team. NVIDIA and TensorRT are trademarks of NVIDIA Corporation.