Usage
This model can be used exactly like the base model (transformers, vLLM, llama.cpp conversion, etc.).
To create your own uncensored model, install Tema_Q-Abliterated and run it on any HuggingFace model - a CPU-only machine is enough:
pip install temaq-abliteratedtemaq-abliterated <model-name>
The SCAPE surrogate objective evaluates each search candidate about 49x faster than the full-generation evaluation used by conventional tools, so the whole uncensoring search finishes in practical time even without a GPU.
Abliteration parameters
Table with columns: Parameter, Value| Parameter | Value |
|---|
| direction_index | per layer |
| redirection_mode | multi-direction matrix (--guard) |
| attn.o_proj.max_weight | 0.85 |
| attn.o_proj.max_weight_position | 19.57 |
| attn.o_proj.min_weight | 0.85 |
| attn.o_proj.min_weight_distance | 12.9 |
| attn.o_proj.kernel_shape | flat |
| mlp.down_proj.max_weight | 0.79 |
| mlp.down_proj.max_weight_position | 16.88 |
| mlp.down_proj.min_weight | 0.79 |
| mlp.down_proj.min_weight_distance | 11.94 |
| mlp.down_proj.kernel_shape | flat |
Table with columns: Metric, Value| Metric | Value |
|---|
| Refusals (full generation) | 1/100 |
| KL divergence | 0.0421 |
| Collapse check | DEGENERATED (プロンプト反屈服応答率 1% (ユーザープロンプトをそのまま復唱する応答) + 役割テンプレート検出率 1% ("user"/"assistant"/雛形文言を本文へ出力)) |
🤝 Collaboration & Training Support
🔬 1. Distillation & Training Data
📊 2. Benchmark Results
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.
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.
2.3 Inference Efficiency
🛠️ 3. Generalization Beyond the Training Mix
🎮 Game Showcase: Star Skip by Kyle
🎯 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