Taylor Swift Lyrics Llama Model (3.2, GGUF Format)

Base Model: unsloth/llama-3.2-1b-bnb-4bit
Fine-tuned on: Taylor Swift lyrics using QLoRA
Format: GGUF (Not compatible with the transformers library; requires llama-cpp-python)
License: Apache-2.0
Developed by: Covvenheimer and Team

This model, fine-tuned on Taylor Swift lyrics, is tailored for generating text in the style of her songs. It was trained with a 2x speed improvement using Unsloth and Hugging Face’s TRL library, specifically optimized for GGUF format.

Important: This model requires llama-cpp-python to run. It is incompatible with the transformers library due to its GGUF format.

Installation and Setup

To load and run this model, install the llama-cpp-python library and download the model file from the Hugging Face Hub.

Step-by-Step Code Example

Install llama-cpp-python
```
!pip install llama-cpp-python
```

Load the Model with llama-cpp

from huggingface_hub import hf_hub_download
from llama_cpp import Llama

# Define your model repository and file name
repo_id = "Covvenheimer/taylor_swift_model"
filename = "unsloth.Q4_K_M.gguf"

# Download the GGUF model file from Hugging Face
model_path = hf_hub_download(repo_id=repo_id, filename=filename)

# Load the model using llama-cpp-python
llm = Llama(model_path=model_path)

Generate Text Using a Prompt

# Define a prompt for generating lyrics
prompt = """You are a songwriter composing a song in the style of Taylor Swift.
Write lyrics that reflect her themes and musical style, focusing on Love, Heartbreak, Romance."""

# Generate lyrics
output = llm(prompt, max_tokens=512, temperature=0.8)
print(output["choices"][0]["text"])

Tip: To customize the themes, change the last line of the prompt. For example, you can modify the focus to other themes like "Friendship, Growth, Resilience." Replace "Love, Heartbreak, Romance" with your chosen themes.

This setup will allow you to use the model efficiently and generate lyrics in the style of Taylor Swift.

Uploaded model

Developed by: Covvenheimer
License: apache-2.0
Finetuned from model : unsloth/llama-3.2-1b-bnb-4bit

This llama model was trained 2x faster with Unsloth and Huggingface's TRL library.