Yuma42/KangalKhan-RawEmerald-7B

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

Yuma42/KangalKhan-RawEmerald-7B is a 7 billion parameter language model based on the Mistral architecture, created by Yuma42 through a merge of CapybaraHermes-2.5-Mistral-7B and distilabeled-OpenHermes-2.5-Mistral-7B. This model is designed for general conversational AI tasks, leveraging the strengths of its merged components. It achieves an average score of 69.09 on the Open LLM Leaderboard, demonstrating solid performance across various reasoning and language understanding benchmarks.

Loading preview...

KangalKhan-RawEmerald-7B: A Merged Mistral-Based Model

KangalKhan-RawEmerald-7B is a 7 billion parameter language model developed by Yuma42. It is a product of merging several Mistral-based models, specifically argilla/CapybaraHermes-2.5-Mistral-7B and [argilla/distilabeled-OpenHermes-2.5-Mistral-7B], using the ties merge method with teknium/OpenHermes-2.5-Mistral-7B as the base.

Key Capabilities & Performance

This model demonstrates robust performance across a range of benchmarks, as evaluated on the Open LLM Leaderboard:

  • Average Score: 69.09
  • AI2 Reasoning Challenge (25-Shot): 66.89
  • HellaSwag (10-Shot): 85.75
  • MMLU (5-Shot): 63.23
  • TruthfulQA (0-shot): 57.58
  • Winogrande (5-shot): 78.22
  • GSM8k (5-shot): 62.85

Usage and Configuration

The model is configured with a context length of 4096 tokens and supports ChatML for conversational interactions. Users can integrate it using standard Hugging Face transformers pipelines. GGUF variants are also available for local inference, with a recommended Q4_K_S GGUF provided by Yuma42.

Good For

  • General-purpose conversational AI applications.
  • Tasks requiring reasoning and language understanding, as indicated by its benchmark scores.
  • Developers looking for a capable 7B model built on a strong Mistral foundation.