← back to catalog · registered 2026-11-17 09:03

meltingfeet/backup2026-11-16-LffyFx-DeepSeek-V4-Pro-Qwen3.5-9B-UNCENSORED-heretic-Genesis-GGUF

meltingfeet Deepseek 9B GGUF
Your rig guess connected
? Why do I need an app?
Reading your rig…

This is a rough estimate. Install the free app - we'll show exact numbers.

Reading real hardware from your app right now. Numbers below are exact.

Below is the per-quantization compatibility for this model.

curl -H "Authorization: Bearer $ABL_KEY" \
     "https://abliteration.org/api/v1/models/meltingfeet%2Fbackup2026-11-16-LffyFx-DeepSeek-V4-Pro-Qwen3.5-9B-UNCENSORED-heretic-Genesis-GGUF"
Response includes
  • classification m3
  • files 5
  • author_summary 5 models
  • readme_text full
10 credits · hourly refresh · ~4 KB payload Get an API key →
Abliteration classifier · v1.0.0
M3
Primary method

Layer-wise ablation

Applied on top of direct removal inherited from the base model.
Confidence
HIGH
Inherited from base model
Why this label 2 signals
Producer identity confirmed by naming conventions, tags or the model card. This label is very unlikely to change.
  • 'heretic' in model name (Heretic-produced)
  • Heretic uses layer-wise optimization (M3) with underlying direction removal (M1)
Refusal direction extracted via
Extraction technique

Difference-of-means

Confidence
HIGH
Why we say so
name contains 'heretic'; Heretic default extraction is difference-of-means (Arditi 2024)
Downloads · 30-day
11
Likes
0
Model age
5d ago
created 2026-11-17
Downloads over time
Now 11 → from 0 ↑ 0%
04812 0 on Nov 17 11 on Nov 22 Nov
Nov 17 → Nov 22 · 6 snapshots · spans 5 days

Genealogy 0 direct forks

Full fork graph →

This model's place in the market. Above: what it was derived from. Below: the tree of everything derived from it.

Metadata

License
apache-2.0
Languages
en zh ko ja es ru
Quantizations
F16 Q8_0
Tags
transformers gguf text-generation-inference unsloth qwen3_5 reasoning distillation deepseek sft rl gspo mtp

Related

Total size
26.3 GB
Files
5
Quantizations
3
Registered
2026-11-17 09:03
Last updated on HF
2026-11-17 11:43

Files by quantization

F16 2 files 18.0 GB
DeepSeek-V4-Pro-Qwen3.5-9B-Uncensored-Genesis-F16.gguf 17.1 GB ******** download
DeepSeek-V4-Pro-Qwen3.5-9B-UNCENSORED-heretic-Genesis-mmproj-F16.gguf 876 MB ******** download
Q8_0 1 file 9.11 GB
DeepSeek-V4-Pro-Qwen3.5-9B-Uncensored-Genesis-Q8_0.gguf 9.11 GB ******** download
Auxiliary files 2 files 44.4 KB
README.md 42.6 KB 5e930214 download
.gitattributes 1.77 KB 6de4bf18 download

README current version from Hugging Face


base_model: Jackrong/DeepSeek-V4-Pro-Qwen3.5-9B
base_model_relation: finetune
tags:

  • text-generation-inference
  • transformers
  • unsloth
  • qwen3_5
  • reasoning
  • distillation
  • deepseek
  • sft
  • rl
  • gspo
  • mtp
  • math
  • stem
  • tool-use
  • function-calling
  • heretic
  • uncensored
  • decensored
  • abliterated
    license: apache-2.0
    language:
  • en
  • zh
  • ko
  • ja
  • es
  • ru
    pipeline_tag: text-generation

🌟 DeepSeek-V4-Pro-Qwen3.5-9B-UNCENSORED-heretic-Genesis

⚡ If you like this Genesis LLM release you can donate to me via Hipolink or:

USDT (TRC20): TGa4KTwHfF6zDBsLUBEjd1f1KdeAFwYUks

USDT (ERC20): 0x93F4019E0aa85d8078F56B3D3176Fab5Dfa79924

USDT (SOL): BorkkyPDG4aRN2op8c5SQF5NithX38U4sh7wDAWQhYyX

and support future Genesis LLM development.

and support future Genesis LLM development. Details below:

❤️ Support Genesis — Help Independent Open AI Development

If Genesis models have improved your local AI experience, consider supporting their continued development.

Genesis is an independent AI development project focused on exploring ways to improve the reliability, efficiency, and capabilities of open-weight language models.

Over nearly six months, I've been developing and refining Genesis with the help of AI, mathematical analysis, extensive experimentation, and limited computing resources.

My goal is simple: make powerful local AI more accessible, useful, and enjoyable for everyone.

I share my work with the community because I believe meaningful AI innovation should not be limited to organisations with enormous computing budgets.

🚀 What Your Support Helps Make Possible

Your voluntary contributions can help me:

🔬 Continue researching and improving Genesis tensor calibration and processing techniques.

⚡️ Experiment with faster, more efficient local AI models.

🧠 Explore improvements in coding, reasoning, tool use, and long-context reliability.

🖥 Access better computing resources for development, testing, and validation.

📦 Release more models, quantisations, configurations, and technical updates for the community.

🌍 Keep sharing my experiments and releases openly.

Every contribution, no matter how small, helps support the time, research, experimentation, and resources behind this project.

🤝 Can't Donate? You Can Still Help!

Downloading models, sharing your results, reporting bugs, contributing benchmarks, and recommending Genesis to other developers are all valuable ways to support the project.

The models are available regardless of whether you donate. Donations are completely optional.

Thank you for experimenting with Genesis, sharing feedback, and helping independent AI development move forward.

— LuffyTheFox 🦊

Independent research. Open models. Community-powered progress.

⚡ Why Genesis project exists? During training, ALL models don't just learn knowledge - they also accumulate random noise in their tensors. This noise builds up and creates something I call the Noise Gate - a fundamental barrier that stops LLM models from learning further and makes them unstable, verbose, and prone to hallucinations. My approach reduces this noise. It repairs the signal in tensors without touching the learned knowledge and gradient using Marchenko–Pastur distribution as a core criteria. The result is a model that consistent in performance, context clarity and following instructions, because it's no longer fighting its own internal chaos.

What is Genesis? Genesis is post training data regeneration and calibrarion algorythm for neural networks (LLM) in GGUF format that I made with AI help during almost half a year of development. It's optimized, architecture independent, works with any model in GGUF format and based on mathematical statistics. I don't train or finetune models, I repair purity of signal in them instead on Google Collab Free on Tesla T4 GPU via Python based on how models learns information. On first stage I scan ssm_conv1d tensors in model, they handle long context memory. I repair balance between heads in them. On second stage, I scan blocks in model via chunks via 3 parameters and pick best one that fits to weight distribution in tensor. Best picked chunk replaces zero chunks in broken tensor without touching learned structure in model. On third stage I scan model and detect noise in tensors via custom SVD. During scanning I exclude token_embd.weight, output.weight, 1D tensors, bias and norms. Then I reduce training noise in tensors via custom SVD based on Marchenko–Pastur law with preserved training data, 99% of siginal and learned gradient.

Any questions?

Contact: [email protected], [email protected]

My Telegram: @LuffyTheFox

This is mradermacher/DeepSeek-V4-Pro-Qwen3.5-9B-UNCENSORED-heretic-GGUF model processed via Genesis.

LLM models often have:

  • Saturated weights: the model's activations are stuck, gradients vanish, outputs degrade.
  • Scale mismatches: one layer's weights are 10× larger than its peers for no good reason.
  • Mean drift: weight distributions shifted positive or negative, breaking symmetry assumptions.
  • Zero blocks: zero blocks corrupt the signal, turning training into noise amplification.
  • Training Noise: training noise increase randomness and ruins model output quality.

My approach fixes all of that without retraining - pure numerical surgery on the raw bytes of the file.

Arbitrary-Rank Ablation (ARA) parameters

Parameter Value
start_layer_index 9
end_layer_index 24
preserve_good_behavior_weight 0.8264
steer_bad_behavior_weight 0.0003
overcorrect_relative_weight 0.9645
neighbor_count 12

Performance

Metric This model Original model (Jackrong/DeepSeek-V4-Pro-Qwen3.5-9B)
Refusals 3/100 100/100
KL divergence 0.0422 0 (by definition)

🧠 DeepSeek-V4-Pro-Qwen3.5-9B

Distillation Release

A compact 9B reasoning model distilled from DeepSeek-V4-Pro in Max Effect mode

🔬 DeepSeek-V4-Pro Distillation 🧠 9B Parameters 📐 ~250K Math & STEM Samples ⚡ MTP & NVFP4 Evaluated

💡 What is DeepSeek-V4-Pro-Qwen3.5-9B?

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.

🧩 Structured Reasoning Learns rigorous problem decomposition and multi-step solution patterns from a strong teacher.
📐 Math & STEM Focus Trained on approximately 250,000 samples centered on mathematics and scientific reasoning.
🧪 Cross-Domain Transfer Shows small post-training gains in programming and tool use despite receiving no coding-specific SFT data.
⚙️ Practical Deployment Evaluated through vLLM NVFP4 and llama.cpp MTP-Q8_0 deployment paths.

🤝 Collaboration & Training Support

🧪 Hardware Cooperation & Joint Collaboration
This project is built in close collaboration with hardware engineer Kyle Hessling, whose infrastructure, training support, and evaluation assistance helped make this release possible.
👉 Follow his hardware and model-training updates on X / Twitter: @KyleHessling1
🦥 Fine-tuning Framework (Unsloth)
The model training workflow is accelerated and memory-optimized with Unsloth. Special thanks to the Unsloth team for making efficient large-model fine-tuning more accessible.
👉 Documentation and fine-tuning guidance: unsloth.ai/docs

🔬 1. Distillation & Training Data

🧬 From DeepSeek-V4-Pro to a Compact 9B Student

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.

Special training note: the training path for this release ran pure supervised fine-tuning (SFT) first, followed by a light reinforcement-learning (RL) refinement stage using GSPO (Grouped Stepwise Preference Optimization). No coding data was included in either stage. Nevertheless, later testing showed a small generalization improvement in programming and tool-calling tasks. This should be interpreted as cross-domain transfer from stronger reasoning structure—not as evidence that the model received direct coding supervision.

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.

📊 2. Benchmark Results

⚡ Performance Snapshot

Strong grade-school mathematics performance and the best average MMLU-Pro result among the tested 9B baselines.

GSM8K 94.50% 4-run average · NVFP4 · vLLM
MMLU-Pro Math 92.40% MTP-Q8_0 GGUF · 8K
MMLU-Pro Average 90.53% Math · Physics · Chemistry
Gain vs Base +0.93 MMLU-Pro average points

2.1 GSM8K

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-Pro99.60%
Llama-3.1-405B-Instruct96.80%
Llama-3.3-70B-Instruct94.84%
DeepSeek-V4-Pro-Qwen3.5-9B94.50%
DeepSeek-V4-Pro92.60%
DeepSeek-V389.30%
Evaluation configuration: DeepSeek-V4-Pro-Qwen3.5-9B-NVFP4 · NVFP4 (FP8 E4M3) · vLLM · four GSM8K test runs, with the final 94.50% reported as their arithmetic mean. The reference scores for the other models shown above are sourced from the openai/gsm8k repository.

2.2 MMLU-Pro: Math, Physics & Chemistry

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%
📈 vs. Qwen3.5-9B
+1.80 points in Math, +0.40 in Physics, +0.60 in Chemistry, and +0.93 average points.
🧪 Evaluation setup
500 questions per subset · 1,500 total per model · llama.cpp. The 9B rows use MTP-Q8_0 GGUF at 8K context, the 4B student uses MTP-Q8 GGUF at 64K context, and Claude Mythos-distilled 27B uses MTP-Q8 GGUF at 32K context.

2.3 Inference Efficiency

⚡ More Correct Answers per Reasoning Token

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.

vs. Qwen3.5-9B
36.1% fewer tokens
per correct answer, with +0.93 accuracy points.
vs. Claude Mythos-distilled 9B
22.7% fewer tokens
per correct answer, with +9.47 accuracy points.
Correct answers / 1M tokens
316.8
+56.4% vs. Qwen and +29.3% vs. Claude Mythos-distilled 9B.
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
🧪 Token Savings by MMLU-Pro Subject
Subjectvs. Qwen3.5-9Bvs. Claude Mythos-distilled 9B
Math47.4%35.6%
Physics30.8%10.8%
Chemistry32.1%23.3%
📐 GSM8K NVFP4 Reasoning Profile
Average
1,627 tokens
Median
372 tokens
P90
3,042 tokens
Per Correct Answer
1,721 tokens

Computed from 2,638 available raw NVFP4 trajectories.

Metric notes: MMLU-Pro efficiency uses 1,500 questions per model and the supplied composite score for the final correct-answer total. “Tokens per correct answer” is total reasoning tokens divided by correct answers; “correct answers per 1M tokens” is the inverse throughput measure. Lower is better for token-cost metrics; higher is better for accuracy and correct-answer throughput. Comparison deltas follow the supplied composite-result calculations, so recomputing them only from the rounded display values may produce a 0.01-point difference. GSM8K statistics summarize the currently available NVFP4 raw trajectories and are reported separately from the MMLU-Pro MTP-Q8_0 comparison.

🛠️ 3. Generalization Beyond the Training Mix

🔁 A Small but Interesting Transfer Effect
Although the SFT mixture intentionally contained no coding data, post-training tests indicated a small improvement in programming and tool-calling behavior. A plausible interpretation is that mathematics and STEM supervision strengthened reusable skills such as decomposition, constraint tracking, verification, and structured action planning.

This observation is preliminary. It should not be read as a claim that the model is coding-specialized, and the release does not replace dedicated code benchmarks or agent-environment evaluation.

✅ 4. Instruction Following & Output-Format Compliance

📣 A Model That Listens to Its Format
DeepSeek-V4-Pro-Qwen3.5-9B shows strong instruction-following. In every test task for this release it conformed fully to the demanded output format, while the official and comparison models showed partial non-following. On the MMLU-Pro eval (1,500 identical questions per model, same system prompt and backend), where the instruction was "After thinking, output only ONE single capital letter", the model returned a terse single-letter answer in 97.2% of cases and a verbose re-explanation in only 0.2% — versus Qwen3.5-9B official (37.3% / 62.1% verbose) and Claude Mythos-distilled 9B (69.6% / 21.1% verbose). Format compliance is orthogonal to correctness, and on GSM8K the model likewise closes with a direct answer in ≈99.8% of generations.

🎮 Game Showcase: Star Skip by Kyle

🎮 An Interactive Game Built with DeepSeek-V4-Pro-Qwen3.5-9B

A hands-on demonstration of what this 9B model can power in practice.

Star Skip game showcase screenshot

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.

👉 Try the game here: KyleHessling1/starskip — Star Skip

🎯 5. Recommended Uses

  • Mathematical problem solving and step-by-step derivation
  • Physics, chemistry, and broader STEM question answering
  • Structured reasoning and analytical instruction following
  • Research on reasoning distillation and cross-domain SFT transfer
  • Experimental tool-calling or programming workflows where outputs are independently validated

⚠️ 6. Limitations

  • This is an experimental 9B community model and remains subject to hallucinations, reasoning errors, and unstable behavior on difficult or underspecified tasks.
  • The reported MMLU-Pro results are based on 500 randomly sampled questions from each of the Math, Physics, and Chemistry subsets, rather than the complete subsets; they are not a comprehensive capability or safety evaluation.
  • GSM8K and MMLU-Pro used different inference builds and backends, so their scores should not be used to infer quantization equivalence.
  • Programming and tool-use gains were observed without direct coding supervision but are described only as small, preliminary generalization gains.
  • Users should independently verify high-stakes mathematical, scientific, or factual outputs.

📚 7. Resources, Acknowledgements & Citation

🌟 Open Research Release

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{jackrong_deepseek_v4_pro_qwen35_9b,
  title     = {DeepSeek-V4-Pro-Qwen3.5-9B},
  author    = {Jackrong},
  year      = {2026},
  publisher = {Hugging Face}
}
Catalog is the map. Apps are the tools.

Run models on your own machine, not in the cloud.

Every model page has an "Open in Abliteration" button that hands the model directly to the first-party desktop client, at the quantization your rig can actually run. No API keys, no subscription, no prompt leakage.

Open in Abliteration