garage-bAInd/Platypus2-70B-instruct

TEXT GENERATIONConcurrency Cost:4Model Size:69BQuant:FP8Ctx Length:32kPublished:Aug 4, 2023License:cc-by-nc-4.0Architecture:Transformer0.2K Open Weights Cold

Platypus2-70B-instruct is a 69 billion parameter instruction-tuned language model, merging garage-bAInd/Platypus2-70B and upstage/Llama-2-70b-instruct-v2. Based on the LLaMA 2 transformer architecture, it is specifically fine-tuned using a STEM and logic-based dataset, Open-Platypus. This model is designed for tasks requiring strong reasoning and logical capabilities, making it suitable for academic and technical applications.

Loading preview...

Platypus2-70B-instruct: A Merged LLaMA 2 Variant

Platypus2-70B-instruct is a 69 billion parameter instruction-tuned language model, created by merging garage-bAInd/Platypus2-70B and upstage/Llama-2-70b-instruct-v2. It is built upon the LLaMA 2 transformer architecture and primarily supports the English language.

Key Capabilities & Training

  • Foundation: Based on the robust LLaMA 2 architecture, providing a strong base for language understanding and generation.
  • Specialized Fine-tuning: The garage-bAInd/Platypus2-70B component was instruction fine-tuned using LoRA on the garage-bAInd/Open-Platypus dataset, which is rich in STEM and logic-based content. This specialized training enhances its performance on complex reasoning tasks.
  • Instruction Following: Designed to follow instructions effectively, utilizing a clear ### Instruction: \n<prompt>\n### Response: template.
  • Performance: Achieves an average score of 66.89 on the Open LLM Leaderboard, with notable scores including 71.84 on ARC (25-shot) and 70.48 on MMLU (5-shot).

Use Cases & Considerations

  • Good for: Applications requiring strong logical reasoning, scientific understanding, and problem-solving, particularly in STEM fields.
  • License: Distributed under the Non-Commercial Creative Commons license (CC BY-NC-4.0), restricting commercial use.
  • Limitations: As with all LLMs, it carries risks of producing inaccurate, biased, or objectionable content. Developers should conduct thorough safety testing for specific applications.