NovelCrafter-LoRA is a Low-Rank Adaptation (LoRA) fine-tuned model designed to assist with creative and literary text generation. This adapter is trained on top of Meta's Llama 3.2 Instruct model family and is optimized for generating coherent, stylistically rich narrative text.

Model Summary

Property	Value
Model Type	LoRA Adapter (PEFT)
Base Model	`meta-llama/Llama-3.2-3B-Instruct` (GPU) / `meta-llama/Llama-3.2-1B-Instruct` (CPU)
Language	English
License	MIT License
Fine-tuning Method	LoRA (Low-Rank Adaptation)
Training Steps	19
Last Updated	2026-01-20

Technical Specifications

LoRA Configuration

Parameter	Value
Rank (r)	8
Alpha	32
Dropout	0.05
Target Modules	`q_proj`, `v_proj`
Bias	None
Task Type	Causal Language Modeling

Training Configuration

Parameter	Value
Batch Size	1 (per device)
Gradient Accumulation Steps	8
Learning Rate	5e-5
Warmup Steps	100
Optimizer	AdamW (torch)
Epochs per Training Unit	3

How to Use

Installation

First, install the required dependencies:

pip install transformers peft torch accelerate

Loading the Model

Option 1: Using PEFT (Recommended)

python

import torch
from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer

# Choose base model based on your hardware
# GPU (recommended): meta-llama/Llama-3.2-3B-Instruct
# CPU: meta-llama/Llama-3.2-1B-Instruct
BASE_MODEL = "meta-llama/Llama-3.2-3B-Instruct"
ADAPTER_REPO = "a-01a/novelCrafter"

# Load base model
model = AutoModelForCausalLM.from_pretrained(
    BASE_MODEL,
    torch_dtype=torch.float16,  # Use float32 for CPU
    device_map="auto",
)

# Load LoRA adapter
model = PeftModel.from_pretrained(model, ADAPTER_REPO)

# Load tokenizer from adapter repo (includes custom chat template)
tokenizer = AutoTokenizer.from_pretrained(ADAPTER_REPO)

print("Model loaded successfully!")

Option 2: Merging Adapter into Base Model

python

import torch
from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer

BASE_MODEL = "meta-llama/Llama-3.2-3B-Instruct"
ADAPTER_REPO = "a-01a/novelCrafter"

# Load and merge
base_model = AutoModelForCausalLM.from_pretrained(BASE_MODEL, torch_dtype=torch.float16)
model = PeftModel.from_pretrained(base_model, ADAPTER_REPO)
model = model.merge_and_unload()  # Merge LoRA weights into base model

tokenizer = AutoTokenizer.from_pretrained(ADAPTER_REPO)

Generating Text

python

def generate_text(prompt, max_new_tokens=512):
    '''Generate text using the fine-tuned model.'''
    messages = [
        {
            "role": "system",
            "content": "You are a skilled creative writing assistant. Write engaging, "
                       "descriptive prose with attention to character development and narrative flow."
        },
        {
            "role": "user",
            "content": prompt
        }
    ]
    
    # Apply chat template
    formatted_prompt = tokenizer.apply_chat_template(
        messages,
        tokenize=False,
        add_generation_prompt=True
    )
    
    # Tokenize
    inputs = tokenizer(formatted_prompt, return_tensors="pt").to(model.device)
    
    # Generate
    with torch.no_grad():
        outputs = model.generate(
            **inputs,
            max_new_tokens=max_new_tokens,
            temperature=0.7,
            top_p=0.9,
            do_sample=True,
            pad_token_id=tokenizer.eos_token_id,
        )
    
    # Decode and return
    response = tokenizer.decode(outputs[0], skip_special_tokens=True)
    return response.split("assistant")[-1].strip()

# Example usage
result = generate_text(
    "Write an opening paragraph for a mystery novel set in Victorian London."
)
print(result)

Using with Transformers Pipeline

python

from transformers import pipeline

generator = pipeline(
    "text-generation",
    model=model,
    tokenizer=tokenizer,
    device_map="auto",
)

output = generator(
    "Once upon a time in a distant kingdom,",
    max_new_tokens=200,
    temperature=0.7,
)[0]["generated_text"]

print(output)

Intended Use

Primary Use Cases

Creative Writing Assistance: Generating narrative prose, story continuations, and creative text
Academic Research: Studying fine-tuning techniques for creative text generation
Educational Purposes: Learning about LoRA adapters and PEFT methods

Out-of-Scope Uses

This model is NOT intended for:

Commercial applications without proper licensing
Generating misleading or deceptive content
Any application that violates ethical guidelines
Production systems without human oversight

Limitations and Biases

The model may generate repetitive or incoherent text for very long outputs
Quality depends heavily on the prompt structure and specificity
May exhibit biases present in the base model and training data
Not optimized for factual accuracy or real-world knowledge tasks
Performance varies based on the base model variant used

Ethical Considerations

⚠️ ACADEMIC AND RESEARCH USE ONLY

This model is released strictly for academic research and educational purposes. By using this model, you agree to:

Non-Commercial Use: This model may not be used for commercial purposes without explicit written permission.
Responsible Use: Users must ensure generated content does not cause harm, spread misinformation, or violate any laws.
Attribution: Any academic publications or research using this model should provide appropriate citation.
No Malicious Use: The model must not be used to generate harmful, abusive, or illegal content.
Human Oversight: All generated content should be reviewed by humans before any public distribution.
Compliance: Users must comply with the base model's (Meta Llama) license terms and acceptable use policy.

Citation

If you use this model in your research, please cite:

bibtex

@misc{NovelCrafter-lora-2025,
  title={NovelCrafter-LoRA: A Fine-Tuned Language Model for Creative Writing},
  author={a-01a},
  year={2025},
  publisher={Hugging Face},
  url={https://huggingface.co/a-01a/novelCrafter}
}

Contact

For questions, issues, or collaboration inquiries, please open an issue on the Hugging Face repository.

Disclaimer: This model is provided "as-is" without warranty of any kind. The authors are not responsible for any misuse or harm caused by the use of this model.