Zardos/Kant-Test-0.1-Mistral-7B

TEXT GENERATIONConcurrency Cost:1Model Size:7BQuant:FP8Ctx Length:4kPublished:Dec 5, 2023License:apache-2.0Architecture:Transformer Open Weights Cold

Zardos/Kant-Test-0.1-Mistral-7B is a 7 billion parameter pretrained generative text model based on the Mistral-7B-v0.1 architecture, developed by Mistral AI. It incorporates Grouped-Query Attention and Sliding-Window Attention, outperforming Llama 2 13B on tested benchmarks. This model is suitable for general text generation tasks, offering strong performance for its size.

Loading preview...

Model Overview

Zardos/Kant-Test-0.1-Mistral-7B is a 7 billion parameter pretrained generative text model, leveraging the Mistral-7B-v0.1 architecture developed by the Mistral AI team. This model is noted for its efficiency and performance, outperforming larger models like Llama 2 13B across various benchmarks.

Key Architectural Features

  • Grouped-Query Attention (GQA): Enhances inference speed and reduces memory requirements.
  • Sliding-Window Attention (SWA): Optimizes attention mechanisms for longer sequences, improving efficiency.
  • Byte-fallback BPE tokenizer: Provides robust tokenization.

Performance Highlights

Evaluated on the Open LLM Leaderboard, this model achieves an average score of 62.42. Notable benchmark results include:

  • AI2 Reasoning Challenge (25-Shot): 62.37
  • HellaSwag (10-Shot): 82.84
  • MMLU (5-Shot): 63.38
  • Winogrande (5-shot): 78.30

Intended Use Cases

As a pretrained base model, Zardos/Kant-Test-0.1-Mistral-7B is well-suited for a wide range of general text generation tasks where a balance of performance and computational efficiency is desired. Developers should note that, as a base model, it does not include built-in moderation mechanisms.