Thespis-7b-v0.2-SFTTest-3Epoch: A Verbose Test Model

This model, developed by cgato, is presented as a test iteration, with no guarantees regarding its performance. A key characteristic noted by the developer is its increased verbosity compared to earlier models. It was fine-tuned using a comprehensive collection of external datasets, including:

• Pure-Dove Dataset
• Claude Multiround 30k
• OpenOrcaSlim
• Augmental Dataset
• ToxicQA
• yahoo_answers_qa

Additionally, it incorporated a cleaned chub.ai characters dataset, provided by kubernetes-bad. The model is configured to use a standard chat prompt format, compatible with default Ooba and Silly Tavern templates, where user input and bot responses alternate. Users should expect potential repetition in its output as the dataset is still undergoing refinement.

Key Characteristics

• Verbose Output: Designed to produce more extensive responses.
• Diverse Training Data: Fine-tuned on a wide array of conversational and QA datasets.
• Chat Format Compatibility: Optimized for common chat interfaces like Ooba and Silly Tavern.

Intended Use

• Testing and Experimentation: Primarily for evaluating model behavior and dataset impact.
• Applications Requiring Detail: Potentially useful where verbose, descriptive outputs are desired.