license: apache-2.0 base_model: scottyjmp5/Legal-Qwen3.5-9B-Abliterated datasets:
- scottyjmp5/courtlistener-legal-corpus language:
- en pipeline_tag: image-text-to-text tags:
- legal
- caselaw
- abliterated
- vision
- tool-calling
- gguf
- ollama
- llama.cpp
Legal-Qwen3.5-9B-Abliterated-GGUF
GGUF build of Legal-Qwen3.5-9B-Abliterated for Ollama / llama.cpp.
| File | Size | Notes |
|---|---|---|
legal-9b.Q8_0.gguf | 9.8 GB | text model, Q8_0 |
legal-9b-mmproj.gguf | 0.9 GB | vision projector (required for image input) |
Runs fully on a 12 GB+ GPU. Verified at ~104 tok/s on an RTX 5090 through Ollama 0.32+ with vision, tool calling, and thinking all working.
Ollama
# Modelfile
FROM ./legal-9b.Q8_0.gguf
FROM ./legal-9b-mmproj.gguf
RENDERER qwen3.5
PARSER qwen3.5
PARAMETER temperature 1
PARAMETER top_k 20
PARAMETER top_p 0.95
ollama create legal-9b -f Modelfile
ollama run legal-9b
The RENDERER/PARSER qwen3.5 lines enable Ollama's built-in chat template and
tool-call/thinking parsing for the Qwen3.5 family.
See the main model card for training details, the retrieval (RAG) recommendation, and warnings (abliterated base; not legal advice).