sequelbox/Llama2-70B-StellarBright

TEXT GENERATIONConcurrency Cost:4Model Size:69BQuant:FP8Ctx Length:32kPublished:Oct 9, 2023License:llama2Architecture:Transformer0.0K Open Weights Cold

sequelbox/Llama2-70B-StellarBright is a 69 billion parameter Llama 2-based model developed by sequelbox, offering a general capability upgrade over the base Llama 2. It improves overall knowledge, extended communication, and technical skill using open-source data. This model is primarily recommended as a superior baseline for further fine-tuning rather than direct production deployment.

Loading preview...

Model Overview

sequelbox/Llama2-70B-StellarBright is a 69 billion parameter model built upon the Llama 2 architecture, developed by sequelbox. It aims to provide a general capability upgrade to the base Llama 2 model by incorporating open-source data to enhance its knowledge, communication abilities, and technical skills.

Key Characteristics

  • Enhanced Baseline: Positioned as a stronger foundation compared to the original Llama 2 for subsequent fine-tuning efforts.
  • Improved Capabilities: Demonstrates advancements in general knowledge, extended communication, and technical proficiency.
  • Training Data: Utilizes open-source data for its capability improvements.

Performance Highlights

Evaluations show Stellar Bright outperforming both Llama 2 and Llama 2 Chat across several benchmarks:

  • Average Score: Achieves an average score of 74.10, significantly higher than Llama 2's 67.35 and Llama 2 Chat's 66.80.
  • Specific Benchmarks: Shows notable improvements in ARC (72.95 vs 67.32), MMLU (71.17 vs 69.83), and TQA (64.46 vs 44.92).

Recommended Use

This model is primarily recommended as a superior baseline for additional fine-tuning, rather than for direct deployment as a chat model in production environments. Users are advised to consider newer models like Llama 3 for general use cases.