metadata
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 (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.