File size: 759 Bytes
3b88355 cd6010c 3b88355 cd6010c 3b88355 cd6010c 3b88355 cd6010c | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 | ---
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.
|