NexaAI/octo-planner-2b
NexaAI/octo-planner-2b is a 2.5 billion parameter on-device language model developed by Nexa AI, based on the Gemma-2b architecture with an 8192-token context length. It is specifically fine-tuned for the Planner-Action Agents Framework, enabling efficient and rapid planning without cloud connectivity. This model excels at generating action plans for complex tasks, achieving a 98.1% planning success rate on benchmarks, and is optimized for local operation on edge devices.
Loading preview...
Octo-planner: On-device Language Model for Planner-Action Agents
NexaAI/octo-planner-2b is a 2.5 billion parameter language model developed by Nexa AI, designed for efficient on-device operation within the Planner-Action Agents Framework. Built upon the Gemma-2b architecture, this model specializes in generating detailed action plans locally, eliminating the need for cloud connectivity and enhancing privacy.
Key Capabilities
- Efficient Planning: Fine-tuned for high efficiency and low power consumption, making it suitable for edge devices.
- Agent Framework Integration: Separates planning from action execution, allowing for specialized optimization and improved scalability when paired with models like Octopus-V2.
- High Accuracy: Achieves a notable 98.1% planning success rate on benchmark datasets, ensuring reliable performance.
- On-device Operation: Optimized for local processing, providing fast response times and enhanced data privacy.
Training and Use Cases
The model was trained using custom Android API descriptions to facilitate its planning capabilities. It is ideal for applications requiring complex task decomposition and action sequencing on resource-constrained devices, such as smart assistants or robotics, where rapid, private, and offline planning is crucial. An example use case involves generating a sequence of actions for a multi-step command like "Find my presentation, connect to projector, increase brightness, take screenshot, and email it."
License and Citation
For usage details, refer to the license page. Academic and research use should cite the provided arXiv paper: chen2024octoplannerondevicelanguagemodel.