lodrick-the-lafted/Limon-8B

Limon-8B Overview

Limon-8B is an 8 billion parameter language model developed by lodrick-the-lafted, building upon the foundation of their Olethros-8B model. It incorporates a notable modification in its training methodology, specifically concerning the datasets used for safety and harmlessness tuning.

Key Characteristics

• Parameter Count: 8 billion parameters, offering a balance between performance and computational efficiency.
• Context Length: Supports an 8192-token context window, enabling processing of moderately long inputs and generating coherent responses.
• Dataset Ablations: A distinguishing feature is the use of specific datasets for safety tuning: LimaRP for harmless content and tatsu-lab/alpaca for harmful content. This approach suggests an intentional focus on refining the model's responses to sensitive queries.

Potential Use Cases

• Content Moderation: Its specific training on harmless and harmful datasets could make it suitable for applications requiring content filtering or safety assessment.
• Research into Safety Alignment: Developers and researchers interested in the impact of different safety datasets on model behavior may find Limon-8B a valuable model for experimentation and analysis.