BarraHome/Mistroll-7B-v2.2

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:7BQuant:FP8Ctx Length:8kTool Calling:SupportedPublished:Apr 26, 2024License:mitArchitecture:Transformer0.0K Open Weights Warm

BarraHome/Mistroll-7B-v2.2 is a 7 billion parameter language model developed by BarraHome, trained with Unsloth and Huggingface's TRL library. This model is an experimental iteration focused on refining training and evaluation pipelines to optimize data engineering, architectural efficiency, and evaluation performance. Its primary objective is to test methods for improving LLM behavior through adjustments in data preprocessing and training algorithms.

Loading preview...

Overview

BarraHome/Mistroll-7B-v2.2 is a 7 billion parameter experimental language model developed by BarraHome. It was trained using Unsloth and Huggingface's TRL library, notably achieving 2x faster training speeds. The core purpose of this model is to serve as a research framework for testing and refining specific training and evaluation pipelines.

Key Objectives

This model's development focuses on:

  • Identifying potential optimizations in LLM training.
  • Improving data engineering processes.
  • Enhancing architectural efficiency.
  • Refining evaluation performance.
  • Exploring adjustments in data preprocessing and model training algorithms.

Use Case

This model is primarily intended for:

  • Research and Development: Ideal for researchers and developers interested in experimenting with new training and evaluation methodologies for Large Language Models.
  • Pipeline Optimization: Useful for those looking to test and refine data engineering, architectural, and evaluation strategies within an LLM context.

Quantized Version

A quantized version, Mistroll-7B-v2.2-Q8_0, is available for users seeking optimized deployment options.

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