---
license: apache-2.0
base_model: Qwen/Qwen3.8-27B
tags:
  - gguf
  - quantization
  - iq2_m
  - ud
  - experimental
---

# Qwen3.8-27B-NQ-UD-IQ2_M

> ⚠️ **EXPERIMENTAL REPOSITORY** - the model is **not functional yet** and is under active development.

Quantization of Qwen3.8-27B using the **UD (Unsloth Dynamic) / IQ2_M** scheme with [NeuralQuant](https://huggingface.co/agiws/NeuralQuant) (NQ) tools.

- **NQ** - NeuralQuant (our quantizer)
- **UD** - Unsloth Dynamic: different layers use different bit-widths (sensitive layers are kept more precise)
- **IQ2_M** - llama.cpp quantization scheme (~2.7-3.0 bpw): a mix of IQ2_S / IQ3_S / Q4_K / Q5_K

Status: the library is preparing the GGUF writer and packers; model files will be published later.