tinaxie/Uno-Orchestra-7B-SFT

TEXT GENERATIONConcurrency Cost:1Model Size:7.6BQuant:FP8Ctx Length:32kPublished:Apr 25, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

Uno-Orchestra-7B-SFT by tinaxie is a 7.6 billion parameter router model, based on Qwen2.5-7B-Instruct, designed to decompose complex tasks into subtasks and dispatch them to specialized worker models and skills. It is the Supervised Fine-Tuning (SFT) stage of a two-stage pipeline, with a subsequent cost-aware GRPO stage planned for release. This model excels at structured planning, routing, observation, and verification, making it suitable for orchestrating multi-turn, multi-capability AI workflows.

Loading preview...

Uno-Orchestra-7B-SFT Overview

Uno-Orchestra-7B-SFT is the Supervised Fine-Tuning (SFT) stage of a 7.6 billion parameter router model developed by tinaxie. Built upon the Qwen/Qwen2.5-7B-Instruct base, this model is engineered to intelligently decompose complex tasks into smaller subtasks and route them to appropriate worker models and skills. It represents the initial phase of a two-stage pipeline, with a subsequent cost-aware GRPO (Goal-Restricted Policy Optimization) stage, Uno-Orchestra-7B-RL, planned for separate release.

Key Capabilities

  • Task Decomposition and Dispatch: Breaks down intricate tasks into manageable subtasks.
  • Structured Planning: Emits a structured plan, route, observe, and verify trace for task execution.
  • Multi-turn Routing: Trained on distilled multi-turn router trajectories from the tinaxie/Uno-Curriculum dataset.
  • Diverse Capability Axes: Covers atomic reasoning, compositional reasoning, knowledge retrieval, knowledge composition, and tool orchestration.

Good for

  • Orchestrating Complex AI Workflows: Ideal for applications requiring dynamic task routing and multi-model coordination.
  • Developing Agentic Systems: Provides a foundational component for building sophisticated AI agents that can leverage multiple specialized models.
  • Research in Router Models: Serves as a strong SFT checkpoint for further research and development in intelligent routing and task management within LLM ecosystems.