Yuma42/KangalKhan-RawRuby-7B

TEXT GENERATIONConcurrency Cost:1Model Size:7BQuant:FP8Ctx Length:4kPublished:Feb 17, 2024License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

Yuma42/KangalKhan-RawRuby-7B is a 7 billion parameter language model created by Yuma42, formed by merging KangalKhan-Ruby-7B-Fixed and KangalKhan-RawEmerald-7B. This model is designed for general language tasks, leveraging a slerp merge method to combine the strengths of its constituent models. It offers a 4096-token context length and demonstrates competitive performance across various benchmarks, including reasoning and common sense tasks.

Loading preview...

KangalKhan-RawRuby-7B Overview

KangalKhan-RawRuby-7B is a 7 billion parameter language model developed by Yuma42. It is a merged model, combining the capabilities of Yuma42/KangalKhan-Ruby-7B-Fixed and Yuma42/KangalKhan-RawEmerald-7B using the slerp merge method via LazyMergekit. This approach aims to synthesize the strengths of both base models into a single, cohesive unit.

Key Capabilities & Performance

The model demonstrates solid performance across a range of benchmarks, as evaluated on the Open LLM Leaderboard. Key metrics include:

  • Average Score: 68.95 on the primary leaderboard.
  • Reasoning: Achieves 66.89 on AI2 Reasoning Challenge (25-Shot) and 62.02 on GSM8k (5-shot).
  • Common Sense: Scores 85.53 on HellaSwag (10-Shot) and 78.69 on Winogrande (5-shot).
  • General Knowledge: MMLU (5-Shot) score of 63.46.

It supports a 4096-token context length and is designed to be used with a ChatML-like instruction format. GGUF quantized versions are available for efficient deployment.

Good For

  • General-purpose text generation and understanding.
  • Applications requiring reasoning and common sense capabilities.
  • Developers looking for a merged model with balanced performance across various tasks.