flywheel-ai/legal-intake
The flywheel-ai/legal-intake model is a 35.1 billion parameter vertical AI-employee model developed by Flywheel by OpSpot. Fine-tuned from Qwen/Qwen3.6-35B-A3B, it is specifically optimized for legal-intake tasks. This model is designed to excel in specialized legal domain applications, offering both safetensors and GGUF formats for flexible deployment.
Loading preview...
Overview
The flywheel-ai/legal-intake model is a specialized 35.1 billion parameter AI model developed by Flywheel by OpSpot. It is fine-tuned (using LoRA) from the Qwen/Qwen3.6-35B-A3B base model, specifically targeting the legal-intake domain. This model is designed to function as a vertical AI-employee, providing focused capabilities for legal applications.
Key Characteristics
- Base Model: Qwen/Qwen3.6-35B-A3B (Apache-2.0 license)
- Parameter Count: 35.1 billion
- Context Length: 32768 tokens
- Specialization: Optimized for legal-intake tasks.
- Formats: Available in
safetensors(for transformers/vLLM) andmodel-q4_k_m.gguf(for llama.cpp/Ollama).
Training and Provenance
The initial v1.0 of this model was trained using synthetic seed data generated by permissively-licensed local models (Apache/MIT licensed teachers). The developers emphasize that no closed models were used for distillation. While its general prompt performance is comparable to its base model, its specialized capabilities in the legal niche are expected to improve with real-world usage through the OpSpot flywheel.
Deployment Options
Users can deploy this model using various methods:
- llama.cpp: Via
llama-serverwith the GGUF file. - Ollama: Directly pulling the GGUF from Hugging Face.
- vLLM: Serving the safetensors format.