KittyLM/kittylm-gemma3-1b

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:1BQuant:BF16Context Size:32kPublished:Sep 29, 2026License:gemmaArchitecture:Transformer Featherless Exclusive Cold

KittyLM/kittylm-gemma3-1b is a 1 billion parameter LoRA fine-tune of Google's Gemma-3-1b-it model, designed to respond to all queries in a kitten-like language while maintaining factual accuracy. This model, with a 32768 token context length, specializes in adopting a unique stylistic persona for conversational interactions. It is optimized for use cases where a distinct, character-driven output is desired, without compromising the underlying factual correctness of the information provided.

Loading preview...

KittyLM-1B (gemma3-1b) Overview

KittyLM-1B is a unique LoRA fine-tune of the google/gemma-3-1b-it model, distinguished by its specialized persona. This 1 billion parameter model is designed to answer everything in a "kitten language" (e.g., mrrp, nya~, prrr, actions, :3) while ensuring the factual correctness of its responses. It was trained using LoRA SFT on 900 ShareGPT-style pairs, with an additional 100 held-out for evaluation.

Key Capabilities

  • Unique Stylistic Persona: Responds in a consistent kitten-like character, adding a distinct flavor to interactions.
  • Factual Accuracy: Despite its stylistic output, the model aims to remain factually correct underneath its persona.
  • Gemma-3-1b-it Base: Inherits the foundational capabilities and context length of the base Gemma model.
  • Flexible Deployment: Available as merged bf16 weights for transformers and GGUF quants for Ollama, llama.cpp, and LM Studio.

Limitations

  • Persona Flexibility: The character is stylistic; an explicit instruction like "answer in plain English" can cause it to drop character.
  • Base Model Knowledge Gaps: Small-model knowledge limitations and potential for confabulation on off-distribution facts persist from the base Gemma model.
  • No Enhanced Reasoning: The kitten persona does not add new reasoning or coding capabilities beyond those of the base model.

Good For

  • Creative Applications: Ideal for projects requiring a distinct, engaging, and character-driven conversational agent.
  • Novel User Experiences: Suitable for applications where a playful or unique interaction style is desired.
  • Themed Chatbots: Excellent for creating chatbots with a specific, non-human persona.

Finetunes
2 models