Gurubot/chatterbots-uncensored-3b
Gurubot/chatterbots-uncensored-3b is a 3.2 billion parameter language model developed by Gurubot, specifically designed to simulate multi-character online chatrooms. It generates realistic, back-and-forth conversations with distinct character personalities based on a provided scenario. This model utilizes a custom XML-delimited JSON format for input and output, making it unique for generating dynamic, multi-participant dialogues rather than standard chat responses. It is uncensored, allowing for a wide range of conversational content.
Loading preview...
Gurubot/chatterbots-uncensored-3b: Multi-Character Chat Simulation
This model, developed by Gurubot, is a 3.2 billion parameter language model uniquely engineered for simulating multi-character online chatroom environments like Discord or IRC. Unlike traditional chat models, it generates responses from several distinct characters simultaneously, creating realistic, interwoven conversations based on a user-defined scenario.
Key Capabilities & Features
- Multi-Character Dialogue Generation: Produces conversations involving multiple participants, each with unique personalities and typing styles.
- Personality Consistency: Designed to maintain consistent character personalities throughout the dialogue, preventing the model from mixing up roles.
- Custom XML-delimited JSON Format: Employs a specific input/output format where the prompt is a JSON object describing the scenario and chat history, and the output is a JSON array of new messages.
- Scenario-Driven Conversations: The
descriptionfield in the input JSON is crucial for defining the chatroom's context, character dynamics, and conversational tone. - Uncensored Output: Based on a base model, it is uncensored, meaning users should implement their own content moderation if needed.
- Reply-to Metadata: Generated messages include
in-reply-tometadata, allowing for sophisticated UI implementations.
Usage Considerations
This model is not a standard instruct model; it requires a specific JSON input structure. Users can provide a list of desired usernames or allow the model to invent its own characters. To manage repetition, which can occur in this format, strategies like adjusting temperature, min_p, repeat penalties, or using a DRY sampler are recommended. For the 3B version, a more general description tends to yield better results than highly specific ones, which can lead to topic fixation.