Model Overview

ChaoticNeutrals/Prima-LelantaclesV4-7b-16k-bf16 is a 7 billion parameter language model, developed by ChaoticNeutrals. This model is a product of a slerp merge, combining the strengths of two base models: Test157t/Yarncules-7b-128k and Test157t/Prima-LelantaclesV3-7b. The merge process utilized a specific configuration, blending layers from both source models with varying weights for self-attention and MLP components, and is provided in bfloat16 format.

Performance Highlights

Evaluated on the Open LLM Leaderboard, Prima-LelantaclesV4-7b-16k-bf16 achieved an average score of 68.28. Key benchmark results include:

• AI2 Reasoning Challenge (25-Shot): 66.04
• HellaSwag (10-Shot): 85.07
• MMLU (5-Shot): 64.70
• TruthfulQA (0-shot): 54.76
• Winogrande (5-shot): 80.27
• GSM8k (5-shot): 58.83

These scores indicate a balanced performance across various reasoning, common sense, and knowledge-based tasks.

Use Cases

This model is suitable for applications requiring a general-purpose language model with solid performance in:

• Reasoning and Problem Solving: Demonstrated by its scores on AI2 Reasoning Challenge and GSM8k.
• Common Sense Understanding: Indicated by strong performance in HellaSwag and Winogrande.
• General Knowledge and Factual Recall: Supported by its MMLU score.

Its bfloat16 precision makes it efficient for deployment in environments that support this data type.