ArGrigorov's picture
translate README to English
cd6010c verified
|
Raw
History Blame Contribute Delete
759 Bytes
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.