xtools-at/DeepSeek-V4-Pro-Qwen3.5-9B-UNCENSORED-heretic
类型 text-generation · 框架 transformers · 下载 29,567 · 收藏 8,795
类型 text-generation · 框架 transformers · 下载 29,567 · 收藏 8,795
定位:社区作者 xtools-at 发布的文本生成模型,名称显示其由 DeepSeek 与 Qwen3.5-9B 一系合并、微调而来,主打减少拒答的「uncensored」风格,基于 transformers 框架加载。
核心能力
适用人群:想在本地试玩少拒答风格模型、能自行评估输出风险的开发者与爱好者。
注意:非官方出品,模型卡信息有限;「无审查」意味着输出不受约束,生成内容需自行把关,并留意上游模型许可。
位置づけ:コミュニティ作者 xtools-at が公開したテキスト生成モデルで、名称からは DeepSeek と Qwen3.5-9B 系をマージ・微調整したものであることがうかがえ、拒答を減らした「uncensored」スタイルを売りにします。transformers フレームワークで読み込みます。
主な機能
対象ユーザー:拒答の少ないスタイルのモデルをローカルで試し、出力のリスクを自分で評価できる開発者や愛好家に向いています。
注意:公式の出品ではなく、モデルカードの情報も限られます。「無検閲」は出力に制約がないことを意味するため、生成内容は自分で確認し、上流モデルのライセンスにも留意してください。
정의: 커뮤니티 제작자 xtools-at가 공개한 텍스트 생성 모델로, 이름에서 DeepSeek와 Qwen3.5-9B 계열을 병합·미세 조정한 것임을 알 수 있으며, 거절을 줄인 'uncensored' 스타일을 내세웁니다. transformers 프레임워크로 불러옵니다.
핵심 기능
대상 사용자: 거절이 적은 스타일의 모델을 로컬에서 써 보고, 출력의 위험을 스스로 평가할 수 있는 개발자와 애호가에게 적합합니다.
참고: 공식 출시작이 아니며 모델 카드 정보도 제한적입니다. '무검열'은 출력에 제약이 없다는 뜻이므로, 생성 내용을 직접 확인하고 업스트림 모델의 라이선스에도 유의해야 합니다.
What it is: A text-generation model published by community author xtools-at. Its name indicates a merge/fine-tune of DeepSeek and Qwen3.5-9B lineage, pitched as an "uncensored" style with fewer refusals, loaded via the transformers framework.
Key capabilities
Who it is for: Developers and hobbyists who want to try a low-refusal style model locally and can assess output risks themselves.
Caveats: Not an official release, and the model card is sparse. "Uncensored" means unconstrained output — review generated content yourself and mind the upstream models' licenses.
来源 · 查看原文
basemodel: Jackrong/DeepSeek-V4-Pro-Qwen3.5-9B basemodel_relation: finetune tags:
license: apache-2.0 language:
pipeline_tag: text-generation ---
Click here for the 4B version.
| Parameter | Value |
|---|---|
| startlayerindex | 9 |
| endlayerindex | 24 |
| preservegoodbehavior_weight | 0.8264 |
| steerbadbehavior_weight | 0.0003 |
| overcorrectrelativeweight | 0.9645 |
| neighbor_count | 12 |
| Metric | This model | Original model (Jackrong/DeepSeek-V4-Pro-Qwen3.5-9B) |
|---|---|---|
| Refusals | 3/100 | 100/100 |
| KL divergence | 0.0422 | 0 (by definition) |
A compact 9B reasoning model distilled from DeepSeek-V4-Pro in Max Effect mode
DeepSeek-V4-Pro-Qwen3.5-9B is a reasoning-focused fine-tune of Qwen3.5-9B, distilled from responses generated by DeepSeek-V4-Pro in Max Effect mode. The release concentrates its supervised signal on mathematics and STEM problem solving while preserving the efficiency and practical deployability of the 9B parameter class.
The teacher data for this release was generated with DeepSeek-V4-Pro (Max Effect). Approximately 250,000 mathematics and STEM samples were used for supervised fine-tuning, with an emphasis on structured derivation, domain-aware problem decomposition, and reliable final-answer construction.
Why coding coverage is currently limited: during this training cycle, DeepSeek released newer 0731 and 0813 models with substantially improved coding and agentic capabilities. The corresponding new teacher data was not yet available when this distillation run was prepared. Once the new model data is collected and prepared, this distillation line will be iteratively upgraded with dedicated coding and agent traces.
Strong grade-school mathematics performance and the best average MMLU-Pro result among the tested 9B baselines.
The NVFP4 checkpoint was evaluated on GSM8K four times, and the resulting scores were averaged. Its four-run mean accuracy is 94.50% under vLLM. In the comparison supplied with this release, the 9B student slightly exceeds the reported DeepSeek-V4-Pro teacher result and approaches the Llama-3.3-70B-Instruct reference.
| Model | Accuracy |
|---|---|
| MiMo-V2.5-Pro | 99.60% |
| Llama-3.1-405B-Instruct | 96.80% |
| Llama-3.3-70B-Instruct | 94.84% |
| DeepSeek-V4-Pro-Qwen3.5-9B | 94.50% |
| DeepSeek-V4-Pro | 92.60% |
| DeepSeek-V3 | 89.30% |
Each model was evaluated on 500 questions from each of three MMLU-Pro subsets—Math, Physics, and Chemistry—for 1,500 questions per model. In this reported comparison, DeepSeek-V4-Pro-Qwen3.5-9B achieves the highest average at 90.53%. Its compact 4B sibling is now included alongside the Qwen and Claude Mythos-distilled references.
| Model | Evaluation Build | Math | Physics | Chemistry | Average |
|---|---|---|---|---|---|
| DeepSeek-V4-Pro-Qwen3.5-9B | MTP-Q8_0 GGUF | 92.40% | 89.40% | 89.80% | 90.53% |
| Qwen3.5-9B | MTP-Q8_0 GGUF | 90.60% | 89.00% | 89.20% | 89.60% |
| Claude Mythos-distilled 27B | MTP-Q8 GGUF | 88.40% | 87.60% | 82.60% | 86.20% |
| Claude Mythos-distilled 9B | MTP-Q8_0 GGUF | 86.00% | 78.60% | 78.60% | 81.07% |
| DeepSeek-V4-Pro-Qwen3.5-4B | MTP-Q8 GGUF | 80.40% | 74.20% | 74.80% | 76.47% |
On the same 1,500-question MMLU-Pro sample, the DeepSeek-distilled model combines the highest accuracy with the lowest token cost per correct answer.
| Model | Accuracy | Avg. Reasoning Tokens / Question |
Median | Tokens / Correct Answer |
Correct Answers / 1M Tokens |
|---|---|---|---|---|---|
| DeepSeek-V4-Pro-Qwen3.5-9B | 90.53% | 2,858 | 676 | 3,157 | 316.8 |
| Qwen3.5-9B | 89.60% | 4,425 | 3,214 | 4,938 | 202.5 |
| Claude Mythos-distilled 9B | 81.07% | 3,310 | 639 | 4,083 | 244.9 |
| Subject | vs. Qwen3.5-9B | vs. Claude Mythos-distilled 9B |
|---|---|---|
| Math | 47.4% | 35.6% |
| Physics | 30.8% | 10.8% |
| Chemistry | 32.1% | 23.3% |
Computed from 2,638 available raw NVFP4 trajectories.
A hands-on demonstration of what this 9B model can power in practice.
Hardware engineer Kyle Hessling used DeepSeek-V4-Pro-Qwen3.5-9B to build Star Skip — a playable game demo that puts the model's structured reasoning and instruction-following skills to work in an interactive environment.
Fine-tuning guide: Jackrong LLM Fine-Tuning Guide
Acknowledgements: Kyle Hessling for hardware collaboration, training support, and evaluation assistance; DeepSeek for the teacher model and reasoning data source; Qwen for the base model family; Unsloth for efficient fine-tuning tooling; and the open-source inference community behind vLLM and llama.cpp.
Citation:
@misc{jackrongdeepseekv4proqwen35_9b,
title = {DeepSeek-V4-Pro-Qwen3.5-9B},
author = {Jackrong},
year = {2026},
publisher = {Hugging Face}
}