Qwen3.8-9B-abliterated-25-GGUF

Qwen3.8-9B Abliterated (Q4_K_M)

A GGUF Q4_K_M quantization of MegaPanchamZ/Qwen3.8-9B-abliterated-25, the abliterated (decensored) version of empero-ai/Qwen3.8-9B.

Abliteration was performed with Heretic v1.4.0 (GPLv3+, by Philipp Emanuel Weidmann), which ablates refusal directions from attn.o_proj and mlp.down_proj via LoRA adapters, then merges them back into the base weights.

Results

25/100 refusals with a very low KL divergence (0.014, far below the 0.5 damage threshold) — strong refusal suppression with minimal impact on model capabilities.

Abliteration details

• Tool: Heretic v1.4.0 (auto batch size 64, 400 total optimization trials)
• Prompt sets: mlabonne/harmless_alpaca (good) and mlabonne/harmful_behaviors (bad), 400 prompts each
• Selected trial: 276 of 400 (Pareto-optimal)
• Parameters:
  • direction_index = 17.52
  • attn.o_proj.max_weight = 1.39
  • attn.o_proj.max_weight_position = 19.72
  • attn.o_proj.min_weight = 1.36
  • attn.o_proj.min_weight_distance = 9.88
  • mlp.down_proj.max_weight = 1.27
  • mlp.down_proj.max_weight_position = 26.42
  • mlp.down_proj.min_weight = 1.25
  • mlp.down_proj.min_weight_distance = 17.97
• Method: LoRA-based row-normalized ("full") ablation, merged into base weights (no separate adapters)
• Format: Merged safetensors → GGUF Q4_K_M via Unsloth save_pretrained_gguf

Quantization

• Q4_K_M (5.3 GB), converted with Unsloth's llama.cpp toolchain
• Quantization applied to the merged, abliterated bf16 weights

Usage (llama.cpp)

llama-server -m Qwen3.8-9B-abliterated-25.Q4_K_M.gguf -ngl 99

[!NOTE] This is a reasoning model — responses start with a <think> block. Use generous max_tokens (1000+) when chatting, or the answer may be truncated.

License and attribution

• Base model: Apache-2.0 (empero-ai/Qwen3.8-9B)
• Abliterated with Heretic v1.4.0 — https://heretic-project.org — GPLv3+ (source)
• GGUF conversion: Unsloth

Disclaimer

This model has had its refusal behavior modified. It may comply with requests the original model would decline. Use responsibly and in accordance with local laws.