Overview

Kukedlc/NeuralLLaMa-3-8b-DT-v0.1 is an 8 billion parameter language model built upon the Llama-3 architecture, featuring an 8192-token context window. This model is a result of a merge operation using LazyMergekit, combining three distinct Llama-3 based models: mlabonne/ChimeraLlama-3-8B-v2, nbeerbower/llama-3-stella-8B, and uygarkurt/llama-3-merged-linear. The merge was performed using the DARE TIES method, with specific density and weight parameters assigned to each contributing model.

Key Capabilities

• Instruction Following: The model demonstrates the ability to follow complex instructions, as shown in chat examples involving creative text generation and role-playing.
• Multilingual Output: Examples indicate proficiency in generating responses in languages like Spanish, suggesting potential for multilingual applications.
• Creative Text Generation: Capable of generating imaginative and detailed responses based on user prompts, including ASCII art and mixed diagrams.

Performance Metrics

Evaluated on the Open LLM Leaderboard, NeuralLLaMa-3-8b-DT-v0.1 achieved an average score of 21.12. Notable scores include:

• IFEval (0-Shot): 43.71
• BBH (3-Shot): 28.01
• MMLU-PRO (5-shot): 31.02

Good For

• Applications requiring a capable 8B parameter model for general language understanding and generation.
• Use cases involving instruction-tuned interactions and creative content creation.
• Developers looking for a merged model that combines strengths from several Llama-3 variants.