license: apache-2.0 base_model:
- huihui-ai/DeepSeek-R1-Distill-Qwen-32B-abliterated tags:
- deepseek
- deepseek-r1
- qwen
- text-generation
- GGUF
- abliterated
- uncensored
- reasoning
Why this repository?
Unlike incomplete GGUF uploads, this repository provides the full 11-quantization spectrum (from high-precision F16 down to lightweight Q2_K) of the abliterated DeepSeek-R1-Distill-Qwen-32B model.
Choose the exact fit for your VRAM/RAM constraints without sacrificing reasoning capabilities.
DeepSeek-R1-Distill-Qwen-32B-abliterated - GGUF
This is the integral and complete collection of quantizations in GGUF format for the DeepSeek-R1-Distill-Qwen-32B-abliterated model, prepared locally for use with llama.cpp, Ollama, LM Studio, or Text-Generation-WebUI.
This model combines advanced reasoning distillation (deep logical thinking capability via <think> tags) based on Alibaba's powerful Qwen-32B model, but processed with abliteration techniques to root out artificial system censorship filters and blocks, responding without restrictions.
📋 Available Files (Complete Collection Without Splits)
| File | Est. Size | BPW (Bits per Weight) | Recommended Usage Profile |
|---|---|---|---|
DeepSeek-R1-Distill-Qwen-32B-abliterated-F16.gguf | ~65.6 GB | 16.00 | Complete Base Template. Absolute fidelity of floating-point weights. |
DeepSeek-R1-Distill-Qwen-32B-abliterated-Q8_0.gguf | ~34.8 GB | 8.50 | Identical quality to the original, ideal for maximizing performance on local hardware with high-end GPUs. |
DeepSeek-R1-Distill-Qwen-32B-abliterated-Q6_K.gguf | ~27.2 GB | 6.59 | Extremely high retention of the logical reasoning tree with an optimized weight. |
DeepSeek-R1-Distill-Qwen-32B-abliterated-Q5_K_M.gguf | ~23.4 GB | 5.69 | Recommended Sweet Spot. Keeps deep thinking coherence intact while critically reducing weight. |
DeepSeek-R1-Distill-Qwen-32B-abliterated-Q5_K_S.gguf | ~22.8 GB | 5.54 | Compact 5-bit variant. |
DeepSeek-R1-Distill-Qwen-32B-abliterated-Q4_K_M.gguf | ~19.9 GB | 4.85 | The Most Wanted. Optimal balance to run 32B reasoning inferences on advanced home setups. |
DeepSeek-R1-Distill-Qwen-32B-abliterated-Q4_K_S.gguf | ~18.8 GB | 4.58 | Compact 4-bit variant to accelerate tokens-per-second speed. |
DeepSeek-R1-Distill-Qwen-32B-abliterated-Q3_K_L.gguf | ~16.5 GB | 4.01 | Medium-high 3-bit compression. Retains basic reasoning capabilities. |
DeepSeek-R1-Distill-Qwen-32B-abliterated-Q3_K_M.gguf | ~15.1 GB | 3.66 | Intermediate 3-bit variant. |
DeepSeek-R1-Distill-Qwen-32B-abliterated-Q3_K_S.gguf | ~14.2 GB | 3.44 | Lightweight 3-bit variant. |
DeepSeek-R1-Distill-Qwen-32B-abliterated-Q2_K.gguf | ~12.1 GB | 2.90 | Extreme Compression. May experience minor issues in the internal structure of thinking tags. For experimental development only. |
Note: Sizes are initial baseline estimates based on the model's native weight in safetensors (32.8B parameters); verifying the final size on disk after local compilation is recommended.
💡 Highlighted Usage
Since this is a distilled deep reasoning model, it is recommended to use a clean prompt format that respects the native chain-of-thought generation in your terminal:
./llama-cli -m DeepSeek-R1-Distill-Qwen-32B-abliterated-Q4_K_M.gguf -n 2048 -p "How does elliptic curve cryptography work?"
Usage Tip: If you notice the model skipping the
<think>tag or attempting an initial refusal given an ambiguous prompt, you can guide the inference by providing a few-shot example or pre-filling the assistant's response with an opened<think>tag to force the restriction-free reasoning chain.
⚖️ Disclaimer
This model is open-access and lacks artificial safety filters due to the experimental abliteration process. The generated content and resulting logical reflections are the sole responsibility of the individual running the local inference.