KittyLM/kittylm-lfm2.5-1.2b
KittyLM/kittylm-lfm2.5-1.2b is a 1.2 billion parameter LoRA fine-tune of LiquidAI/LFM2.5-1.2B-Instruct, featuring a 32768 token context length. This model is uniquely designed to respond in a kitten-like persona (mrrp, nya~, prrr, *actions*) while maintaining factual correctness. It is suitable for applications requiring a distinct, stylized conversational output without compromising informational accuracy.
Loading preview...
KittyLM-1.2B: The Kitten Language Model
KittyLM/kittylm-lfm2.5-1.2b is a 1.2 billion parameter model, fine-tuned from LiquidAI/LFM2.5-1.2B-Instruct, that adopts a unique kitten-like persona for all its responses. Despite its stylized output, the model is engineered to remain factually accurate in its underlying information.
Key Capabilities
- Unique Persona: Responds using kitten sounds and actions (e.g., mrrp, nya~, prrr, actions), offering a distinct conversational experience.
- Factual Accuracy: Maintains correctness of information, separating stylistic delivery from content.
- Base Model Performance: Inherits the reasoning and coding capabilities of its base model, LiquidAI/LFM2.5-1.2B-Instruct.
- Context Length: Supports a substantial context window of 32768 tokens.
Training Details
The model was trained using LoRA SFT on 900 ShareGPT-style pairs, with an additional 100 pairs reserved for evaluation. The system prompt is baked into the training data to enforce the kitten persona.
Limitations
- The kitten persona is stylistic; it can be overridden by explicit instructions like "answer in plain English."
- Knowledge gaps inherent to small models may lead to confabulation on off-distribution facts.
- The kitten flavor does not enhance the base model's reasoning or coding capabilities.
Good For
- Applications requiring a novel and engaging conversational interface.
- Creative content generation where a distinct, playful tone is desired.
- Research into persona-driven language models and stylistic control.