Model Overview

macadeliccc/gemma-orchid-7b-dpo is an 8.5 billion parameter Gemma-based language model, fine-tuned using Direct Preference Optimization (DPO). This model represents the second checkpoint in an ongoing project, focusing on enhancing both communicational skills and function calling abilities.

Key Capabilities

• Function Calling: The model is capable of understanding and generating function calls, making it suitable for integration with external tools and APIs.
• Strong Communicational Skills: Designed to handle a wide variety of conversational and text generation tasks effectively.
• DPO Fine-tuning: Trained on approximately 80,000 samples using datasets like Thermostatic/flowers, Intel/orca_dpo_pairs, jondurbin/truthy-dpo-v0.1, and glaiveai/glaive_function_calling_v2, which includes a blend of open-source model generations and function calling data.

Performance

Evaluations on the Open LLM Leaderboard show an average score of 64.37, with specific metrics including:

• AI2 Reasoning Challenge (25-Shot): 62.88
• HellaSwag (10-Shot): 80.95
• MMLU (5-Shot): 61.41
• TruthfulQA (0-shot): 53.27
• Winogrande (5-shot): 77.51
• GSM8k (5-shot): 50.19

Good For

• Applications requiring function calling: Ideal for scenarios where the model needs to interact with external systems or execute specific actions.
• General conversational AI: Its strong communicational base makes it suitable for chatbots, virtual assistants, and content generation.
• Further fine-tuning: The base model, macadeliccc/gemma-function-calling-7b, is explicitly mentioned as suitable for further fine-tuning with custom datasets.