Overview

Tess-10.7B-v1.5, short for Tesoro (Treasure in Italian), is a general-purpose large language model developed by migtissera. This version is built upon the SOLAR-10.7B base architecture and features 10.7 billion parameters with a 4096-token context length. An updated version, Tess-10.7B-v1.5b, with hyperparameter optimizations, is also available.

Key Capabilities & Performance

The model's performance has been evaluated on the Open LLM Leaderboard, achieving an average score of 66.55. Notable results include:

• AI2 Reasoning Challenge (25-Shot): 65.02
• HellaSwag (10-Shot): 84.07
• MMLU (5-Shot): 65.09
• Winogrande (5-Shot): 83.35
• GSM8k (5-Shot): 54.36

Prompt Format

The model utilizes a straightforward prompt format:

SYSTEM: <ANY SYSTEM CONTEXT>
USER: 
ASSISTANT:

Good For

• General natural language understanding and generation tasks.
• Applications requiring solid performance in reasoning and common sense benchmarks.
• Developers looking for a 10.7B parameter model with a 4096-token context window for various NLP applications.