beita6969/SkillFlow-Model
The beita6969/SkillFlow-Model is a 9 billion parameter language model based on the Qwen3.5 architecture, specifically designed as a merged SkillFlow Supervisor. This model integrates a LoRA adapter trained for a supervisor forward policy, making it specialized for orchestrating complex tasks. It is optimized for inference in bfloat16 and serves as a core component for advanced AI workflows requiring supervisory control.
Loading preview...
SkillFlow Merged Supervisor Model
This repository hosts the merged weights for the SkillFlow Supervisor model, a specialized language model built upon the Qwen3.5-9B base architecture. It is designed to act as a forward policy supervisor within the SkillFlow framework.
Key Characteristics
- Base Model: Utilizes the robust Qwen/Qwen3.5-9B as its foundation.
- Adapter Integration: Incorporates a LoRA adapter (rank 64, alpha 128) specifically trained for the supervisor forward policy (
theta). - Target Modules: The LoRA adapter targets
q_proj,k_proj,v_proj, ando_projfor fine-tuning. - Inference Optimized: Merged with
bfloat16data type for efficient deployment. - Context Length: Supports a substantial context window of 32768 tokens.
Purpose and Differentiation
Unlike general-purpose LLMs, SkillFlow-Model is not an instruction-tuned chatbot but a component within a larger AI system. Its primary role is to provide supervisory control and policy guidance, making it suitable for applications requiring an orchestrator or decision-making layer. The training-time backward policy adapter is intentionally excluded from this inference model, streamlining it for its specific forward policy function. Developers can find the source code and dataset at the SkillFlow GitHub repository and Hugging Face dataset respectively.