xLAM-8x22b-r
| Property | Value |
|---|---|
| Parameter Count | 141B |
| Context Length | 64k tokens |
| License | CC-BY-NC-4.0 |
| Architecture | Mixtral-based Transformer |
| Paper | arxiv:2409.03215 |
What is xLAM-8x22b-r?
xLAM-8x22b-r is a state-of-the-art Large Action Model (LAM) developed by Salesforce Research. It represents the largest model in the xLAM family, specifically designed to enhance decision-making and translate user intentions into executable actions. Built on a Mixtral architecture with 141 billion parameters, it excels at function calling and general agent applications with an impressive 64k context window.
Implementation Details
The model utilizes BF16 precision and is optimized for both function-calling and general agent tasks. It's implemented using PyTorch and requires Transformers 4.41.0 or higher. The model includes specialized tokenization and prompt formatting for optimal performance.
- Advanced function-calling capabilities with JSON-formatted outputs
- Multi-turn interaction support
- Comprehensive tool use and API integration
- 64k context window for handling long sequences
Core Capabilities
- Autonomous planning and task execution
- Structured function calling with JSON outputs
- Tool-use and API interaction
- Complex decision-making and reasoning
- Multi-turn conversation handling
Frequently Asked Questions
Q: What makes this model unique?
xLAM-8x22b-r stands out for its massive scale (141B parameters) and specialized training for function-calling and agent tasks. It's particularly notable for its ability to handle complex tool use and API interactions while maintaining coherent reasoning.
Q: What are the recommended use cases?
The model excels in scenarios requiring automated workflow processes, tool interaction, and complex decision-making. It's particularly well-suited for AI agent systems, automated task execution, and scenarios requiring interaction with multiple APIs or tools.
