sdavies/globe-theatre-qwen25-3b-merged

TEXT GENERATIONConcurrency Cost:1Model Size:3.1BQuant:BF16Ctx Length:32kTool Calling:SupportedPublished:Jun 15, 2026License:apache-2.0Architecture:Transformer Open Weights Cold

sdavies/globe-theatre-qwen25-3b-merged is a 3.1 billion parameter Qwen2.5-3B-Instruct based model, fine-tuned to generate short, comic scenes in the style of a badly organized Elizabethan theatre company. This model transforms modern grievances into exaggerated theatrical comedy, prioritizing style and humor over factual accuracy. It is specifically designed for entertainment and creative text generation, excelling at dramatizing everyday annoyances and social awkwardness.

Loading preview...

The Tragedy of the Group Chat: Merged Model

sdavies/globe-theatre-qwen25-3b-merged is a 3.1 billion parameter standalone Transformers model, built upon the Qwen2.5-3B-Instruct base. It integrates a LoRA adapter to specialize in generating humorous, theatrical scenes. Unlike general-purpose language models, this model's core function is to transform minor modern inconveniences into exaggerated Elizabethan-style comedy.

Key Capabilities

  • Theatrical Scene Generation: Produces short comic scenes featuring a TITLE, DRAMATIS PERSONAE, SCENE, and "THOU MUST CHOOSE" prompts.
  • Distinctive Voice: Combines influences from Shakespeare, Blackadder, Monty Python, British sitcoms, and amateur dramatic societies.
  • Performance: Achieved 56/80 on a manual benchmark, outperforming the base Qwen2.5-3B-Instruct model's 36/80 on the same evaluation.
  • Deployment Ready: This repository provides a merged model for direct use with the Hugging Face Transformers library, simplifying deployment compared to separate LoRA adapters.

Good For

This model is intended for entertainment and creative text generation, performing best on:

  • Everyday annoyances and social awkwardness
  • Household disasters and transport failures
  • Office politics and mildly haunted appliances
  • Inexplicably judgmental animals

Limitations

The model intentionally prioritizes stylistic output and recurring jokes over factual accuracy. It is optimized for small frustrations rather than major life events.