igidn/TinyPi-Chat-v1.5

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:1.1BQuant:BF16Context Size:2kPublished:Jun 23, 2025License:mitArchitecture:Transformer0.0K Open Weights Featherless Exclusive Cold

The igidn/TinyPi-Chat-v1.5 is a 1.1 billion parameter conversational language model, evolved from a Discord chat-fine-tuned base. This version utilizes Reinforcement Learning from AI Feedback (RLAIF) with Google's Gemini 1.5 Flash to enhance factual accuracy and consistency, while maintaining a friendly and engaging persona. It is optimized for coherent, multi-turn conversations and entertainment, offering improved knowledge and reduced repetitiveness within its 2048-token context length.

Loading preview...

TinyPi-1.1B-Chat-v1.5: An Engaging Conversational AI

TinyPi-1.1B-Chat-v1.5 is an advanced 1.1 billion parameter conversational model developed by igidn, building upon its v1 predecessor. This iteration significantly improves upon the original by integrating Reinforcement Learning from AI Feedback (RLAIF), a sophisticated alignment process that leverages Google's Gemini 1.5 Flash AI as a "teacher" to correct and refine its responses.

Key Capabilities

  • Distinct Persona: Maintains a consistent, friendly, and humorous personality, making it an engaging conversational partner.
  • Enhanced Accuracy: Improved factual accuracy and reasoning, addressing inconsistencies present in v1 through targeted RLAIF training.
  • Coherent Conversations: Excels in multi-turn dialogues across various topics, with reduced tendencies for generic refusals or repetitive loops.
  • Robust Training: Developed through a two-stage fine-tuning process, starting with a large Discord chat dataset and then undergoing RLAIF using AI-generated corrections.

Ideal Use Cases

  • Conversational Agents: Suitable for applications requiring an AI with a unique, friendly, and consistent personality.
  • Entertainment: Designed for engaging and humorous interactions.
  • General Chatbots: Capable of coherent discussions on a wide range of subjects, offering a more reliable experience than its predecessor.

Limitations

  • Primarily designed for conversational and entertainment purposes; not a substitute for expert advice.
  • Its core personality may not suit tasks requiring a strictly neutral or formal tone.
  • Inherits biases from its training data, including internet chat logs and AI-generated text.

Finetunes
1 models