license: apache-2.0
base_model:
- JonathanColetti/Qwen3.8-27B-Uncensored
base_model_relation: quantized
pipeline_tag: image-text-to-text
library_name: openvino
tags: - openvino
- int4
- qwen3_5
- vision
- conversational
Qwen3.8-27B Uncensored — OpenVINO INT4
OpenVINO INT4 conversion of JonathanColetti/Qwen3.8-27B-Uncensored, which is based on Qwen/Qwen3.8-27B.
The uncensoring/abliteration work belongs to Jonathan Coletti and the Heretic project. The original Qwen3.8 model belongs to the Qwen team. This repository is only an OpenVINO deployment conversion.
Conversion
- OpenVINO 2026.4 nightly dated 2026-08-14
- Transformers 5.2
- Asymmetric INT4 weight compression
- Group size 128, ratio 1.0
- Native text and vision OpenVINO graphs
- Recipe matched to
OpenVINO/Qwen3.8-27B-int4-ov - CPU text and synthetic-image vision smoke tests passed before upload
Qwen's 15 source MTP tensors were verified before export. The standard OpenVINO VLM graph does not expose the separate MTP speculative draft head, so ordinary decode works but native MTP acceleration is not claimed.
OpenVINO GenAI
from huggingface_hub import snapshot_download
import openvino_genai as ov_genai
model_dir = snapshot_download("Wondernutts/Qwen3.8-27B-Uncensored-int4-ov")
pipe = ov_genai.VLMPipeline(model_dir, "GPU", DYNAMIC_QUANTIZATION_GROUP_SIZE=128)
print(pipe.generate("Write a short introduction.", max_new_tokens=256, temperature=1.0, top_p=0.95, top_k=20))
This is an experimental Qwen3.8 OpenVINO model. Use OpenVINO/OpenVINO GenAI 2026.4 nightly from August 14, 2026 or newer. Review the upstream model cards for intended use and limitations.