---
language:
- en
tags:
- autotrain
- text-generation
- causal-lm
- lora
- style-transfer
- agatha-christie
license: other
datasets:
- realdanielbyrne/AgathaChristieText
base_model: google/gemma-2-2b-it
library_name: transformers
pipeline_tag: text-generation
---

# 🕵️ Agatha Christie Style LoRA — Gemma-2-2B-IT

This model is a **LoRA fine-tuned** version of **Google’s Gemma-2-2B-IT**, trained to generate prose in the **classic style of Agatha Christie**.  
It captures the tone, pacing, domestic atmosphere, and subtle psychological tension characteristic of Christie’s detective fiction.

The model is intended for **story continuation**, **creative writing**, and **mystery scene generation**.  
It is a **causal LM**, not an instruction-tuned chat model.

---

## ✨ Model Details

- **Model type:** Causal language model (LoRA adapter)
- **Base model:** `google/gemma-2-2b-it`
- **Fine-tuning method:** LoRA (PEFT)
- **Training framework:** Hugging Face AutoTrain
- **Languages:** English
- **Primary domain:** Fiction, mystery, detective prose
- **Task:** Text generation (story continuation)

---

## 🎯 Intended Use

This model is suitable for:

- Generating text in the Agatha Christie style  
- Continuing mystery scenes  
- Creative writing assistance  
- Prototyping detective stories  
- Experimenting with stylistic transfer  

### Not recommended for:

- Real-world Q&A  
- Safety-critical applications  
- Factual reasoning  
- Chat-assistant behavior without custom prompting  

---

## 📚 Training Data

The model was trained on:

- **`realdanielbyrne/AgathaChristieText`**  
  A plain-text dataset containing ~13,000 rows of Agatha Christie prose.

No instruction template was used; the model was trained purely on **unstructured text** to learn style rather than task formats.

---

## 🔧 Training Configuration

- **Epochs:** 3  
- **Block size:** 1024  
- **Learning rate:** 3e-5  
- **Batch size:** AutoTrain default  
- **Optimizer:** AdamW  
- **Chat template:** *None (causal LM)*  
- **LoRA target modules:** Linear projection layers of Gemma  

Training was performed using **AutoTrain**, which handled preprocessing, batching, and evaluation.

---

## 🧠 Model Behavior

The model tends to produce:

- restrained, elegant British prose  
- domestic settings and quiet tension  
- subtle descriptions of character behavior  
- slow, methodical narrative pacing  
- attention to objects, rooms, and small clues  

### Example characteristics:

- No supernatural or horror tone unless explicitly prompted  
- Minimal modern slang  
- High coherence for 1–3 paragraphs of continuation  
- Strong stylistic adherence to Christie’s sentence rhythm  

---

## ⚠️ Limitations and Biases

- May repeat narrative structures due to small dataset size  
- Not optimized for long-range story consistency  
- Inherits biases present in Agatha Christie’s historical works  
- Not suited for factual or logical reasoning  
- Can generate outdated cultural norms reflective of 20th-century British literature  

---

## 📊 Evaluation

This model was not evaluated using standard NLP benchmarks, as its purpose is **stylistic generation** rather than accuracy on downstream tasks.

Qualitative evaluation shows:

- Strong stylistic similarity to Christie’s prose  
- High coherence in short continuations  
- Appropriate tone, atmosphere, and pacing  

---

## 🔐 License

This LoRA inherits licensing constraints from:

- the **Google Gemma-2** gated model license  
- the dataset used  
- Hugging Face AutoTrain’s output terms  

Users must be approved to access the Gemma model family.

---

## 🙏 Acknowledgements

- Base model by **Google DeepMind**  
- Fine-tuned using **Hugging Face AutoTrain**  
- Dataset contributed by the open-source community  
- PEFT LoRA framework by **Hugging Face**

---

## 📣 Citation

If you use this model, please cite the base model and the dataset authors.  
You may also reference this repository.