---
license: apache-2.0
base_model:
- Qwen/Qwen2.5-3B-Instruct
tags:
- qwen2.5
- transformers
- comedy
- creative-writing
- shakespeare
- merged
- text-generation-inference
- hackathon
- build-small
library_name: transformers
---
# 🎭 The Tragedy of the Group Chat: Merged Model

A standalone merged Transformers version of The Tragedy of the Group Chat.

This repository combines the Qwen2.5-3B-Instruct base model with the trained LoRA adapter into a single deployable **Transformers** model.

**Most language models try to solve your problems.**

**The Tragedy of the Group Chat** assumes your problems deserve a **badly performed Elizabethan stage production** instead.

> You forgot to buy milk. Again. Everyone is angry. A sensible person would apologise. 
> This model instead assembles an Elizabethan comedy troupe to make the situation considerably worse.


![image](https://cdn-uploads.huggingface.co/production/uploads/6720fc25ebc007333bffab63/DacI6UautUOQOSB18Rdf7.png)

## About the project

The Tragedy of the Group Chat was created for the Hugging Face Build-Small Hackathon (June 2026).

The project explored how much personality and structure can be taught to a relatively small local model through careful fine-tuning and iterative evaluation.

Rather than building a general-purpose assistant, the goal was to transform tiny modern inconveniences into exaggerated theatrical comedy scenes.

## Repository structure

The project consists of three related repositories:
- **LoRA adapter**: The original PEFT fine-tuning.<br>
- **Merged model (this repository)**: A standalone Transformers version of the model. <br>
- **GGUF edition**: A quantised deployment build for llama.cpp and local inference.

## What does it do?

Given a small modern grievance, the model produces a short comic scene in the style of a badly organised Elizabethan theatre company.

Typical outputs include:

* TITLE
* DRAMATIS PERSONAE
* SCENE
* THOU MUST CHOOSE

The intended voice combines influences from:

* Shakespeare
* Blackadder
* Monty Python
* British sitcoms
* Amateur dramatic societies

## Performance

The underlying fine-tuned model achieved **56/80** on a held-out ten-prompt manual benchmark. The base Qwen2.5-3B-Instruct model scored **36/80** using the same evaluation procedure.

The merged model preserves the behaviour of the LoRA adapter while simplifying deployment.

## Intended use

The model is intended for entertainment and creative text generation.

It performs best on:

* everyday annoyances
* social awkwardness
* household disasters
* transport failures
* office politics
* mildly haunted appliances
* inexplicably judgemental animals

## Limitations

The model intentionally prioritises style over factual accuracy.

Recurring characters and running jokes are expected behaviour.

The model performs best on small frustrations rather than major life events.

## Loading

This repository contains a standalone Transformers model and can be loaded directly with the Hugging Face Transformers library.

## Build process

This model was created by:

Qwen2.5-3B-Instruct

↓

LoRA fine-tuning

↓

PEFT merge-and-unload

↓

Standalone Transformers model

This merged model serves as the source for the GGUF deployment build.


## Try the model

- 🎭 [Interactive demo](https://huggingface.co/spaces/sdavies/tragedy-of-the-group-chat)

## Related repositories

- [LoRA adapter](https://huggingface.co/sdavies/globe-theatre-qwen25-3b-lora)
- [GGUF edition](https://huggingface.co/sdavies/globe-theatre-qwen25-3b-gguf)