Aabbhishekk/llama2-7b-function-calling-slerp
TEXT GENERATIONConcurrency Cost:1Model Size:7BQuant:FP8Ctx Length:4kPublished:Jan 30, 2024License:apache-2.0Architecture:Transformer Open Weights Cold

Aabbhishekk/llama2-7b-function-calling-slerp is a 7 billion parameter language model merged from Meta's Llama-2-7b-hf and Trelis's Llama-2-7b-chat-hf-function-calling-v3 using the slerp method. This model is specifically designed for enhanced function calling capabilities, combining the foundational strengths of Llama 2 with specialized function-calling fine-tuning. It is optimized for applications requiring robust tool use and API interaction, operating within a 4096 token context length.

Loading preview...

Overview

Aabbhishekk/llama2-7b-function-calling-slerp is a 7 billion parameter model created by merging two distinct Llama 2 variants: the base meta-llama/Llama-2-7b-hf and the function-calling-optimized Trelis/Llama-2-7b-chat-hf-function-calling-v3. This merge was performed using the slerp (spherical linear interpolation) method via mergekit, allowing for a balanced integration of their respective strengths. The model operates with a context length of 4096 tokens.

Key Capabilities

  • Enhanced Function Calling: Inherits and combines the function-calling capabilities from Trelis/Llama-2-7b-chat-hf-function-calling-v3, making it suitable for tool use and API interaction scenarios.
  • Llama 2 Foundation: Benefits from the robust base architecture of Meta's Llama 2, providing a strong general language understanding.
  • Merged Architecture: Utilizes a slerp merge strategy, which can lead to a synergistic combination of features from its constituent models.

Performance Benchmarks

Evaluated on the Open LLM Leaderboard, the model achieved an average score of 53.53. Notable scores include:

  • HellaSwag (10-Shot): 79.50
  • Winogrande (5-shot): 75.22
  • AI2 Reasoning Challenge (25-Shot): 55.46
  • MMLU (5-Shot): 50.32
  • TruthfulQA (0-shot): 40.32
  • GSM8k (5-shot): 20.39

Good For

  • Developers building applications that require reliable function calling and tool use.
  • Scenarios where a balance between general language understanding and specialized function interaction is needed.
  • Experimentation with merged models for specific task optimizations.
Finetunes
5 models