license: apache-2.0
base_model: xdchinza/Qwen3.5-0.8B-abliterated
tags:

• gguf
• qwen3_5
• abliterated
• uncensored
• conversational
  language:
• en
• zh
• ru
  pipeline_tag: text-generation
  library_name: llama.cpp

Qwen3.5-0.8B-abliterated — GGUF

GGUF quantisation for llama.cpp / Ollama of
xdchinza/Qwen3.5-0.8B-abliterated,
which is a directional-abliteration of
Qwen/Qwen3.5-0.8B.

The full method, per-layer measurements, and the quality/limitations analysis
are documented in the safetensors repository linked above.

File

Usage

Ollama:

ollama create qwen3.5-0.8b-abliterated -hf xdchinza/Qwen3.5-0.8B-abliterated-GGUF:Q4_K_M
ollama run qwen3.5-0.8b-abliterated

llama.cpp:

llama-cli -m Qwen3.5-0.8B-abliterated-Q4_K_M.gguf -p "hello" --jinja

Chat template

The Qwen3.5 chat template is embedded in the GGUF metadata. With llama.cpp, pass
--jinja (or --chat-template qwen3_5) so the template is applied. For
non-thinking mode, ensure enable_thinking is false — the template then emits
an empty <think>\n\n</think> block.

License

Apache-2.0, inherited from the base model.