🧛 Llama-3.2-3B-Dracula Fine-tuned Model

Fine-tuned Llama 3.2 3B model on character dialogues from Bram Stoker's Dracula novel.

Model Details

• Base Model: Llama 3.2 3B
• Fine-tuning: LoRA on character dialogues
• Quantization: Q4_K_M (GGUF format)
• Size: 1.9 GB
• Use Case: Character-based conversational AI
• Characters: Dracula, Mina Harker, Van Helsing, Jonathan Harker, Lucy Westenra, Dr. Seward

Usage

With llama-cpp-python

from llama_cpp import Llama

# Load model
llm = Llama(
    model_path="llama-3.2-3b-dracula-q4_k_m.gguf",
    n_ctx=4096,
    n_gpu_layers=-1  # Use GPU if available
)

# Generate response
response = llm(
    "Tell me about your castle.",
    max_tokens=400,
    temperature=0.6,
    stop=["\n\n", "Human:", "User:"]
)

print(response['choices'][0]['text'])

Download from Hub

from huggingface_hub import hf_hub_download

model_path = hf_hub_download(
    repo_id="Priyanks27/llama-3.2-3b-dracula",
    filename="llama-3.2-3b-dracula-q4_k_m.gguf"
)

Training Details

• Dataset: Character dialogues extracted from Dracula novel
• Method: LoRA fine-tuning
• Base: Llama 3.2 3B Instruct
• Quantization: Q4_K_M for optimal size/quality balance

Performance

• Faithfulness: 5/5 (zero hallucination on novel content)
• Character Consistency: Distinct personalities maintained
• Response Quality: High-quality Victorian-era language
• Speed: ~10 tokens/second on CPU

Use in HuggingFace Space

This model is used by the Dracula Character Chat Space.

License

Apache 2.0

Citation

@misc{dracula-chatbot-2025,
  author = {Priyanks27},
  title = {Llama-3.2-3B-Dracula Fine-tuned Model},
  year = {2025},
  publisher = {HuggingFace},
  howpublished = {\url{https://huggingface.co/Priyanks27/llama-3.2-3b-dracula}}
}

Acknowledgments

• Base model: Meta Llama 3.2
• Novel: Bram Stoker's Dracula (1897)
• Framework: llama.cpp, llama-cpp-python