Model Overview

The decem/Dionysus-Mistral-m3-v5 is a 7 billion parameter language model developed by DECEM. It has been fine-tuned using Supervised Fine-Tuning (SFT) based on the Mistral architecture, primarily for English language tasks. The model operates with an 8192 token context length, making it suitable for processing moderately long inputs and generating coherent responses.

Key Capabilities & Performance

Evaluated on the Open LLM Leaderboard, Dionysus-Mistral-m3-v5 demonstrates a balanced performance across various benchmarks, achieving an overall average score of 63.14.

• Reasoning: Scores 59.56 on AI2 Reasoning Challenge (25-Shot) and 51.02 on GSM8k (5-shot).
• Common Sense: Achieves 80.99 on HellaSwag (10-Shot) and 75.14 on Winogrande (5-shot).
• General Knowledge: Scores 61.18 on MMLU (5-Shot).
• Truthfulness: Records 50.93 on TruthfulQA (0-shot).

Prompting

This model is designed to be prompted using an Alpaca-style instruction format:

### Instruction:

<prompt>

### Response:

Good For

• General-purpose text generation and understanding.
• Applications requiring reasoning and common sense capabilities.
• Tasks benefiting from a 7B parameter model with an 8K context window.