Skip to content
RCreddit.com·

Qwen3.8-27B-Uncensored-Genesis-V1-GGUF

AI summary

The Qwen3.8-27B-Uncensored-Genesis-V1-GGUF model, adapted from a paper (arxiv.org/pdf/1311.0851v1), aims to reduce excessive text generation and token usage in LLMs. The developer believes numerical instability from training noise causes this issue and addresses it by distilling noise using the Marchenko-Pastur distribution. The model uses a specific system prompt and chat template, with inference settings like temperature=1.0 and top_p=0.95. Community feedback is sought due to limited testing resources.

Time & source

Times shown in UTC

Display time zone: UTC

Local time zone unavailable; showing UTC.

PublishedOffset at this time: UTC+0Sep 9, 2026, 14:09 UTC

IngestedOffset at this time: UTC+0Sep 9, 2026, 22:00 UTC

Published
Sep 9, 2026, 14:09
Ingested
Sep 9, 2026, 22:00
Source type
Dev community
Tier
Community
Source status
Healthy

Tier is a per-source editorial setting, not a per-item score.

Article

Model available here: Qwen3.8-27B-Uncensored-Genesis-V1-MTP-GGUF

This model is a practical realisation of things described in this paper, but adapted by me for machine learning: https://arxiv.org/pdf/1311.0851v1

I am trying to solve the problem: why LLM models even for simple questions write walls of text during reasoning, and burn too much tokens instead of solving the task. And when number of parameters increase the problem became worse. I think main problem is numerical instability in tensor matrices during to random training noise accumulation in tensors. Model is fighting with own internal chaos during inference process. I am distilling training noise from tensors using Marchenko-Pastur distribution together with info from paper as a core criteria and solid mathematical foundation behind this project.

Settings:

System Prompt: You are Qwen (Tongyi Qianwen), a large language model developed by Alibaba Group's Tongyi Lab. You are a helpful assistant.

Chat template: chat_template.jinja

Inference settings: temperature=1.0, top_p=0.95, top_k=20, min_p=0.0, presence_penalty=0.0, repetition_penalty=1.0, reasoning_effort=medium

I can't fully test this model on my own, since I only have a modest RTX 3060 graphics card with 12 GB of VRAM. So any feedback from the Reddit community would be very helpful. I'd appreciate any feedback from the community.

Source·reddit.com