adamo1139/Mistral-7B-AEZAKMI-v1

TEXT GENERATIONConcurrency Cost:1Model Size:7BQuant:FP8Ctx Length:8kPublished:Nov 27, 2023License:otherArchitecture:Transformer0.0K Cold

The adamo1139/Mistral-7B-AEZAKMI-v1 is a 7 billion parameter Mistral-based language model fine-tuned on the AEZAKMI v1 dataset, derived from Airoboros 2.2.1 and 2.2. This model is specifically optimized to produce outputs free of refusals and offer a more natural conversational feel, making it suitable for general-purpose chatbot applications. It was trained using QLoRA with NF4 double quantization over approximately two epochs, achieving its primary goal of creating a 'cozy free chatbot' experience.

Loading preview...

adamo1139/Mistral-7B-AEZAKMI-v1 Overview

This model is a 7 billion parameter variant of the Mistral architecture, fine-tuned by adamo1139 using the AEZAKMI v1 dataset. The AEZAKMI v1 dataset itself is a derivative of the Airoboros 2.2.1 and 2.2 datasets. The fine-tuning process utilized axolotl with QLoRA and NF4 double quantization, completing approximately two epochs of training on a single RTX 3090 Ti.

Key Capabilities

  • Refusal-Free Output: A primary design goal was to eliminate refusals in the model's responses.
  • Natural Conversation: The model aims to provide a more natural and fluid conversational experience.
  • ChatML Format: It uses the standard ChatML prompt format for interactions.

Good For

  • Cozy Chatbot Applications: Ideal for creating friendly, general-purpose chatbots that avoid restrictive responses.
  • Conversational AI: Suitable for use cases where a natural and uninhibited dialogue flow is preferred.

Limitations

  • Not Optimized for Complex Tasks: Users should not expect high performance in areas like advanced mathematics, riddles, or highly complex reasoning tasks.

Popular Sampler Settings

Top 3 parameter combinations used by Featherless users for this model. Click a tab to see each config.

temperature
top_p
top_k
frequency_penalty
presence_penalty
repetition_penalty
min_p