Models Qwen 272b

Undi95/MG-FinalMix-72B

4.5 based on 1 review
Warm
Public
72.7B
FP8
131072
License: other
Hugging Face
Overview

Undi95/MG-FinalMix-72B: Enhanced 72.7B Parameter Model

Undi95/MG-FinalMix-72B is a 72.7 billion parameter language model, a retouched and enhanced version of alpindale/magnum-72b-v1, built upon a merge that includes Qwen/Qwen2-72B-Instruct. Developed by Undi95, this model incorporates additional role-playing (RP) data to address identified issues and improve conversational capabilities.

Key Capabilities

  • Merged Architecture: Combines the strengths of Qwen/Qwen2-72B-Instruct and alpindale/magnum-72b-v1.
  • Enhanced Role-Playing: Fine-tuned with extra RP data for more engaging and coherent character interactions.
  • Extended Context: Features a significant context length of 131072 tokens, supporting long-form conversations and complex scenarios.
  • Custom Quantization Support: Provides an imatrix.dat file for users to generate their own quantized versions, based on wiki.train.raw data.

Good For

  • Role-Playing Scenarios: Excels in generating detailed and consistent character-driven narratives.
  • Creative Writing: Suitable for tasks requiring imaginative and extended text generation.
  • Long-form Conversational AI: Its large context window makes it ideal for maintaining coherence over lengthy dialogues.