Salesforce/Llama-xLAM-2-8b-fc-r

Salesforce Llama-xLAM-2-8b-fc-r: Advanced Large Action Model

Salesforce's Llama-xLAM-2-8b-fc-r is an 8 billion parameter model from the xLAM-2 series, designed as a Large Action Model (LAM) to enhance decision-making by translating user intentions into executable actions. This model is built on the Llama architecture and features a 32K context length, which can be extended up to 128K using techniques like YaRN for Qwen-2.5-based models.

Key Capabilities

• Multi-turn Conversation: Engineered for superior performance in complex, multi-turn conversational scenarios.
• Advanced Function-Calling: The -fc suffix denotes its fine-tuning for robust function-calling tasks, enabling AI agents to interact with external tools and APIs effectively.
• Agentic Performance: Achieves state-of-the-art results on the BFCL (Berkeley Function-Calling Leaderboard) and τ-bench benchmarks, outperforming frontier models like GPT-4o and Claude 3.5 in specific agentic capabilities.
• Data Synthesis: Trained using the novel APIGen-MT framework, which generates high-quality training data through simulated agent-human interactions.
• Seamless Integration: Refined chat template and vLLM integration facilitate easier deployment and building of advanced AI agents.

Good For

• AI Agent Development: Ideal for developers building AI agents that require autonomous planning and execution of tasks.
• Complex Conversational AI: Suitable for applications demanding sophisticated multi-turn dialogue management.
• Tool Usage and Automation: Excellent for scenarios where models need to interact with external functions and automate workflows.
• Research Purposes: This model release is specifically for research, providing a powerful tool for exploring advanced LAM capabilities.