ewald1976/MeterMaid-12b
MeterMaid-12B is a 12 billion parameter language model created by ewald1976, based on an unspecified base model and merged with TheDrummer/Rocinante-X-12B-v1. This model was specifically adjusted to be less reserved and more conversational, aiming to enhance its interactive capabilities. It is designed for general conversational tasks where a more expressive and less constrained output is desired, with a context length of 32768 tokens.
Loading preview...
MeterMaid-12B Overview
MeterMaid-12B is a 12 billion parameter language model developed by ewald1976. Its unique characteristic stems from a merge process where an initial, somewhat reserved fine-tuned model was combined with a "small dose" of TheDrummer/Rocinante-X-12B-v1. This merge, performed using an slerp method with a parameter t: 0.10, was intended to make the model more adventurous and less taciturn.
Key Characteristics
- Parameter Count: 12 billion parameters.
- Context Length: Supports a context window of 32768 tokens.
- Behavioral Adjustment: Specifically modified to be less reserved and more talkative, aiming for enhanced conversational engagement.
- Merge Method: Utilizes
slerpfor model merging, incorporating elements from TheDrummer/Rocinante-X-12B-v1.
Recommended Settings
For optimal performance and to encourage its intended conversational style, the following inference settings are recommended:
- Temperature: 0.9
- Min_p: 0.05
- Top_p: 0.95
Use Cases
MeterMaid-12B is suitable for applications requiring a language model that is more expressive and less constrained in its responses, particularly in conversational or interactive scenarios where a more 'awakened desire for adventure' in its output is beneficial. Its name, "MeterMaid," is noted as surprisingly relevant to its characteristics, though the specific implications are left to interpretation.