Model Overview

Gecko-7B-v0.1 is a 7 billion parameter language model developed by NeuralNovel, based on a full-parameter fine-tune (FFT) of Mistral-7B-Instruct-v0.2. Released under an Apache-2.0 license, this model is intended for generating instructive and narrative text, with a particular focus on mathematical and numerical tasks. It was fine-tuned using the Neural-Mini-Math dataset.

Key Capabilities

• Instructive and Narrative Text Generation: Designed to follow prompt directions for creating detailed instructions and stories.
• Mathematics & Numeracy: Optimized for problem-solving in mathematical contexts.
• Roleplay: Suitable for engaging in roleplay scenarios due to its ability to follow complex instructions.

Performance

Evaluated on the Open LLM Leaderboard, Gecko-7B-v0.1 achieved an average score of 64.58. Notable scores include 61.35 on AI2 Reasoning Challenge, 83.36 on HellaSwag, 61.05 on MMLU, and 41.55 on GSM8k, indicating its reasoning and mathematical capabilities.

Limitations and Responsible Use

This model is primarily for testing and research. It has reduced alignment levels and may produce unaligned or harmful content. Users are responsible for their outputs and should exercise caution, especially for scenarios outside of instructive and narrative text generation, where performance may be suboptimal.