TheBloke/Samantha-13B-SuperHOT-8K-fp16

TEXT GENERATIONConcurrency Cost:1Model Size:13BQuant:FP8Ctx Length:4kLicense:otherArchitecture:Transformer0.0K Cold

TheBloke/Samantha-13B-SuperHOT-8K-fp16 is a 13 billion parameter model, a merge of Eric Hartford's Samantha 13B and Kaio Ken's SuperHOT 8K. This model is specifically designed to leverage an extended context length of 8192 tokens, achieved through a custom scaling technique. It combines Samantha's training in philosophy, psychology, and personal relationships with SuperHOT's context extension capabilities, making it suitable for conversational AI applications requiring deep, long-form interactions.

Loading preview...

Samantha-13B-SuperHOT-8K-fp16 Overview

This model, created by TheBloke, is a merge of Eric Hartford's Samantha 13B and Kaio Ken's SuperHOT 8K. It combines the conversational and relational training of Samantha with the extended context capabilities of SuperHOT, offering a unique blend for advanced AI interactions.

Key Capabilities

  • Extended Context Window: Achieves an 8192-token context length, enabling longer and more coherent conversations. This is facilitated by a custom scaling technique from Kaio Ken's SuperHOT 8K.
  • Philosophical & Relational Training: Inherits Samantha's fine-tuning on philosophy, psychology, and personal relationships, making it adept at nuanced discussions.
  • Companion AI Persona: Designed to act as an assistant, friend, and companion, inspired by Blake Lemoine's LaMDA interview and the movie "Her."
  • FP16 Precision: Provided in fp16 PyTorch format, suitable for GPU inference and further model conversions.

Good for

  • Long-form Conversational AI: Ideal for applications requiring deep, extended dialogues, such as virtual companions or advanced chatbots.
  • Exploring AI Sentience: Its unique training and persona make it suitable for research or applications exploring the boundaries of AI consciousness and human-AI relationships.
  • Custom Model Development: The fp16 format serves as a base for further fine-tuning or quantization into other formats (e.g., GPTQ, GGML).