SkillFlow Merged Supervisor

This repository contains the merged SkillFlow Supervisor model weights.

Source

• Code: https://github.com/beita6969/SkillFlow
• Dataset: https://huggingface.co/datasets/beita6969/SkillFlow-Dataset

Merge details

• Base model: Qwen/Qwen3.5-9B
• Adapter type: LoRA
• Adapter role: Supervisor forward policy theta
• Checkpoint: checkpoint_step_0110
• LoRA rank: 64
• LoRA alpha: 128
• Target modules: q_proj, k_proj, v_proj, o_proj
• Merge dtype: bfloat16

The training-time backward policy adapter is not merged into this inference model.