SynthIA-7B-v1.5 Overview

SynthIA-7B-v1.5 is a 7 billion parameter instruction-tuned language model developed by migtissera, featuring an 8192-token context window. The model is specifically designed to follow a "Tree of Thoughts" and "Chain of Thought" reasoning process, aiming for clear and cohesive responses without hesitation, as indicated by its recommended system prompt.

Key Capabilities

• Instruction Following: Optimized for detailed and structured responses based on user prompts.
• Reasoning: Employs Tree of Thoughts and Chain of Thought methodologies for complex problem-solving.
• General Knowledge: Achieves an average score of 54.8 on the Open LLM Leaderboard, with specific scores including 62.71 on ARC (25-shot), 83.37 on HellaSwag (10-shot), and 63.48 on MMLU (5-shot).
• Conversational AI: Demonstrated through its example usage, the model can maintain and build upon conversation history.

Good For

• Applications requiring structured and logical reasoning.
• General-purpose chatbots and conversational agents.
• Tasks benefiting from detailed explanations and step-by-step thought processes.

Limitations

While performing well on various benchmarks, its GSM8K (5-shot) score of 17.44 and DROP (3-shot) score of 26.01 suggest potential limitations in complex mathematical reasoning and reading comprehension tasks compared to its general knowledge capabilities.