squeeze-ai-lab/TinyAgent-1.1B
TinyAgent-1.1B is a 1.1 billion parameter Small Language Model (SLM) developed by Squeeze AI Lab at University of California, Berkeley. It is specifically fine-tuned for complex reasoning and function calling capabilities, designed for secure and private deployment at the edge. This model excels at interacting with applications to assist users with daily tasks like email composition, calendar management, and scheduling, addressing challenges of privacy, connectivity, and latency inherent in larger LLMs.
Loading preview...
TinyAgent-1.1B: Edge-Optimized Function Calling
TinyAgent-1.1B is a Small Language Model (SLM) developed by Squeeze AI Lab at UC Berkeley, specifically engineered for robust function calling and complex reasoning in edge environments. Unlike larger LLMs, TinyAgent-1.1B prioritizes privacy, low latency, and offline operation, making it suitable for deployment on local devices.
Key Capabilities & Features
- Function Calling at the Edge: Designed to enable sophisticated function calling capabilities on resource-constrained devices.
- Application Interaction: Capable of interacting with MacOS applications for tasks such as composing emails, managing contacts, scheduling events, and organizing meetings.
- High-Quality Training Data: Fine-tuned on a curated dataset of 40,000 real-life use cases, with instructions generated by GPT-3.5-Turbo and execution plans by GPT-4-Turbo.
- ToolRAG Integration: Utilizes ToolRAG for efficient retrieval of relevant tools and in-context examples, significantly boosting accuracy and minimizing latency by avoiding irrelevant information.
- Performance: TinyAgent-1.1B, when combined with ToolRAG, achieves an 80.06% success rate on its internal evaluation, outperforming base models like TinyLLama-1.1B-32K-Instruct (12.71%) and even GPT-3.5-turbo (65.04%) in its specialized tasks.
When to Use TinyAgent-1.1B
- Privacy-Sensitive Applications: Ideal for scenarios where data must remain on-device and not be sent to cloud-based LLMs.
- Offline Functionality: Suitable for applications requiring AI assistance without constant internet connectivity.
- Low-Latency Task Automation: Excellent for automating day-to-day tasks on personal devices where quick responses are critical.
- Resource-Constrained Environments: A strong choice for deployment on edge devices where larger models are impractical due to computational or memory limitations.