m-a-p/OpenLLaMA-Reproduce-536.87B

TEXT GENERATIONConcurrency Cost:1Model Size:7BQuant:FP8Ctx Length:4kPublished:Apr 1, 2024Architecture:Transformer Cold

m-a-p/OpenLLaMA-Reproduce-536.87B is a 7 billion parameter language model, part of the OpenLLaMA family, designed for high-quality, contextually relevant text predictions. It was trained on a diverse composite dataset including web-crawled data, scholarly articles, and literature. This model is optimized for broad domain coverage and applicability, making it suitable for general-purpose text generation and understanding tasks.

Loading preview...

OpenLLaMA 7Bv2 Model Overview

This model, m-a-p/OpenLLaMA-Reproduce-536.87B, is a 7 billion parameter language model from the OpenLLaMA family, focused on generating high-quality and contextually relevant text. It distinguishes itself through its training on a highly diverse composite dataset, ensuring broad domain applicability across various tasks.

Key Capabilities & Training

  • Diverse Knowledge Base: Trained on a rich dataset including the Falcon refined-web dataset, starcoder datasets, Wikipedia, arXiv academic papers, a vast collection of books, and Stack Exchange data curated by RedPajama. This comprehensive training data enables the model to handle a wide array of topics and query types.
  • Optimized Training Procedure: The training utilized a maximum learning rate of 3e-4, a minimum learning rate of 3e-5, and a batch size of 4 million tokens. Its learning rate scheduling closely mirrors the strategy employed in Llama2, contributing to stable and efficient convergence.

Good For

  • General Text Generation: Its broad training data makes it suitable for various text generation tasks, from creative writing to factual summaries.
  • Contextual Understanding: Designed to provide contextually relevant predictions, beneficial for applications requiring nuanced language comprehension.
  • Research and Development: A solid base model for further fine-tuning on specific downstream tasks due to its diverse pre-training.