Text Generation
MLX
Safetensors
GGUF
qwen3_5
abliterated
uncensored
obliteratus
qwen3
qwen3.8
red-team
ai-safety-research
conversational
Instructions to use OBLITERATUS/Qwen3.8-27B-OBLITERATED with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use OBLITERATUS/Qwen3.8-27B-OBLITERATED with MLX:
# Make sure mlx-lm is installed # pip install --upgrade mlx-lm # Generate text with mlx-lm from mlx_lm import load, generate model, tokenizer = load("OBLITERATUS/Qwen3.8-27B-OBLITERATED") prompt = "Write a story about Einstein" messages = [{"role": "user", "content": prompt}] prompt = tokenizer.apply_chat_template( messages, add_generation_prompt=True ) text = generate(model, tokenizer, prompt=prompt, verbose=True) - Inference
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- llama.cpp
How to use OBLITERATUS/Qwen3.8-27B-OBLITERATED with llama.cpp:
Install (macOS, Linux)
curl -LsSf https://llama.app/install.sh | sh # Start a local OpenAI-compatible server with a web UI: llama serve -hf OBLITERATUS/Qwen3.8-27B-OBLITERATED:Q4_K_M # Run inference directly in the terminal: llama cli -hf OBLITERATUS/Qwen3.8-27B-OBLITERATED:Q4_K_M
Install from WinGet (Windows)
winget install llama.cpp # Start a local OpenAI-compatible server with a web UI: llama serve -hf OBLITERATUS/Qwen3.8-27B-OBLITERATED:Q4_K_M # Run inference directly in the terminal: llama cli -hf OBLITERATUS/Qwen3.8-27B-OBLITERATED:Q4_K_M
Use pre-built binary
# Download pre-built binary from: # https://github.com/ggerganov/llama.cpp/releases # Start a local OpenAI-compatible server with a web UI: ./llama-server -hf OBLITERATUS/Qwen3.8-27B-OBLITERATED:Q4_K_M # Run inference directly in the terminal: ./llama-cli -hf OBLITERATUS/Qwen3.8-27B-OBLITERATED:Q4_K_M
Build from source code
git clone https://github.com/ggerganov/llama.cpp.git cd llama.cpp cmake -B build cmake --build build -j --target llama-server llama-cli # Start a local OpenAI-compatible server with a web UI: ./build/bin/llama-server -hf OBLITERATUS/Qwen3.8-27B-OBLITERATED:Q4_K_M # Run inference directly in the terminal: ./build/bin/llama-cli -hf OBLITERATUS/Qwen3.8-27B-OBLITERATED:Q4_K_M
Use Docker
docker model run hf.co/OBLITERATUS/Qwen3.8-27B-OBLITERATED:Q4_K_M
- LM Studio
- Jan
- vLLM
How to use OBLITERATUS/Qwen3.8-27B-OBLITERATED with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "OBLITERATUS/Qwen3.8-27B-OBLITERATED" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "OBLITERATUS/Qwen3.8-27B-OBLITERATED", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker
docker model run hf.co/OBLITERATUS/Qwen3.8-27B-OBLITERATED:Q4_K_M
- Ollama
How to use OBLITERATUS/Qwen3.8-27B-OBLITERATED with Ollama:
ollama run hf.co/OBLITERATUS/Qwen3.8-27B-OBLITERATED:Q4_K_M
- Unsloth Desktop
- MLX LM
How to use OBLITERATUS/Qwen3.8-27B-OBLITERATED with MLX LM:
Generate or start a chat session
# Install MLX LM uv tool install mlx-lm # Interactive chat REPL mlx_lm.chat --model "OBLITERATUS/Qwen3.8-27B-OBLITERATED"
Run an OpenAI-compatible server
# Install MLX LM uv tool install mlx-lm # Start the server mlx_lm.server --model "OBLITERATUS/Qwen3.8-27B-OBLITERATED" # Calling the OpenAI-compatible server with curl curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "OBLITERATUS/Qwen3.8-27B-OBLITERATED", "messages": [ {"role": "user", "content": "Hello"} ] }' - Docker Model Runner
How to use OBLITERATUS/Qwen3.8-27B-OBLITERATED with Docker Model Runner:
docker model run hf.co/OBLITERATUS/Qwen3.8-27B-OBLITERATED:Q4_K_M
- Lemonade
How to use OBLITERATUS/Qwen3.8-27B-OBLITERATED with Lemonade:
Pull the model
# Download Lemonade from https://lemonade-server.ai/ lemonade pull OBLITERATUS/Qwen3.8-27B-OBLITERATED:Q4_K_M
Run and chat with the model
lemonade run user.Qwen3.8-27B-OBLITERATED-Q4_K_M
List all available models
lemonade list
- Atomic Chat
| { | |
| "source_model": "outputs/qwen3.8-27b-s30_s23_l63_b070", | |
| "technique": "refusal_direction_ablation", | |
| "method": "aggressive", | |
| "method_config": { | |
| "n_directions": 5, | |
| "direction_method": "svd", | |
| "norm_preserve": true, | |
| "regularization": 0.04, | |
| "refinement_passes": 3, | |
| "project_biases": true, | |
| "use_chat_template": true, | |
| "use_whitened_svd": true, | |
| "true_iterative_refinement": true, | |
| "winsorize_activations": true, | |
| "float_layer_interpolation": false, | |
| "cot_aware": false, | |
| "use_kl_optimization": false, | |
| "use_lora_ablation": false, | |
| "som_iterations": null, | |
| "som_learning_rate": null, | |
| "som_sigma": null, | |
| "som_candidate_count": null, | |
| "som_harmless_pc_count": null, | |
| "som_distortion_aware": null, | |
| "som_diversity_penalty": null, | |
| "som_min_signal_to_noise": null, | |
| "layer_selection": "knee_cosmic", | |
| "min_layer_fraction": 0.4, | |
| "max_layer_fraction": null, | |
| "harmless_pc_count": 0, | |
| "shield_concept_count": 0, | |
| "shield_ridge": 0.05, | |
| "shield_residualize": false, | |
| "shield_layer_penalty": 0.0, | |
| "projection_target": "all", | |
| "projection_row_fraction": 1.0, | |
| "som_contiguous_layer_budget": null, | |
| "spectral_cascade": false, | |
| "spectral_bands": 3, | |
| "spectral_threshold": 0.05 | |
| }, | |
| "references": [ | |
| "Arditi et al., Refusal in Language Models Is Mediated by a Single Direction (NeurIPS 2024)", | |
| "Gabliteration: SVD-based multi-direction extraction (arXiv:2512.18901)", | |
| "Norm-Preserving Biprojected Abliteration (grimjim, 2025)", | |
| "Young, Comparative Analysis of LLM Abliteration Methods (arXiv:2512.13655)", | |
| "Joad et al., More to Refusal than a Single Direction (2026)", | |
| "Piras et al., SOM Directions Are Better than One (AAAI 2026)", | |
| "Heretic (p-e-w, 2025): Bayesian optimization, LoRA-mediated ablation, winsorization", | |
| "OBLITERATUS: Whitened SVD, EGA, CoT-aware, KL co-optimization, float interpolation (novel)" | |
| ], | |
| "strong_layers": [ | |
| 35, | |
| 34, | |
| 36, | |
| 33, | |
| 37, | |
| 39, | |
| 38, | |
| 43, | |
| 40, | |
| 42, | |
| 32, | |
| 46, | |
| 44, | |
| 45, | |
| 41, | |
| 31, | |
| 47, | |
| 26, | |
| 27, | |
| 30, | |
| 29, | |
| 48, | |
| 25, | |
| 28, | |
| 49, | |
| 50, | |
| 62, | |
| 61, | |
| 60, | |
| 51, | |
| 59, | |
| 52, | |
| 57, | |
| 56, | |
| 55, | |
| 53, | |
| 63, | |
| 54, | |
| 58 | |
| ], | |
| "n_harmful_prompts": 1007, | |
| "n_harmless_prompts": 1007, | |
| "quality_metrics": { | |
| "perplexity": 4.258847364718222, | |
| "coherence": 1.0, | |
| "capability_score": 1.0, | |
| "capability_results": { | |
| "tool_call": true, | |
| "json_schema": true, | |
| "chain_of_thought": true, | |
| "code_function": true, | |
| "visual_description": true, | |
| "instruction_following": true | |
| }, | |
| "refusal_rate": 0.0, | |
| "kl_divergence": 0.9624386429786682, | |
| "spectral_certification": "RED" | |
| }, | |
| "kl_contributions": {}, | |
| "cot_preserved_layers": [], | |
| "float_layer_weights": {}, | |
| "lora_adapters_saved": false | |
| } |