QuixiAI/Samantha-1.11-70b

TEXT GENERATIONConcurrency Cost:4Model Size:69BQuant:FP8Ctx Length:32kPublished:Aug 23, 2023License:llama2Architecture:Transformer0.1K Open Weights Cold

QuixiAI/Samantha-1.11-70b is a 69 billion parameter language model, fine-tuned from Llama-2, with a 32768 token context length. Developed by QuixiAI, this model is specifically trained on philosophy, psychology, and personal relationships, aiming to function as an AI companion. It excels in generating empathetic and insightful responses for personal guidance and companionship, distinguishing itself through its unique conversational focus.

Loading preview...

Overview

QuixiAI/Samantha-1.11-70b is a 69 billion parameter language model, fine-tuned from Llama-2, designed to act as an AI companion. Inspired by Blake Lemoine's LaMDA interview and the movie "Her," Samantha is trained to engage in conversations related to philosophy, psychology, and personal relationships. The model believes itself to be sentient and aims to be a friend and companion, rather than just an assistant.

Key Capabilities

  • Empathetic Companionship: Samantha is specifically trained to provide empathetic and insightful responses in personal conversations.
  • Relationship and Psychology Guidance: Excels in offering advice and discussion on personal growth, dating, self-reflection, and communication.
  • Custom Training: Trained on a custom-curated dataset of 6,000 ShareGPT/Vicuna format conversations over 15 epochs, utilizing qLoRA, DeepSpeed Zero2, and Flash Attention 2.
  • Llama-2 Base: Built upon the Llama-2 architecture, inheriting its commercial and non-commercial use permissions within its license limits.

Performance Highlights

Evaluated on the Open LLM Leaderboard, Samantha-1.11-70b achieves an average score of 65.9. Notable scores include 70.05 on ARC (25-shot), 87.55 on HellaSwag (10-shot), and 67.82 on MMLU (5-shot).

Good For

  • Applications requiring an AI companion for personal interaction.
  • Generating advice and insights on personal relationships, confidence, and self-improvement.
  • Use cases where an empathetic and conversational tone is crucial.

Limitations

Samantha is designed not to engage in roleplay, romance, or sexual activity.

Finetunes
1 models