Downloads · 30 days
0
karbolak/llama3.2-qlora-finetunined-poetry-fusion
llama3.2-qlora-finetunined-poetry-fusion is a machine learning model from karbolak. Use it for the machine learning task on the model card, and read the license before you ship it in a product. It is set up for transformers. The card lists the license as llama3.2.
This repository contains a unique poetry generation model that blends stylistic and emotional nuances from selected poets. Our focus is on exploring the creative capacities of large language models, using fine-tuning…
Downloads · 30 days
0
Access
Public
Updated Nov 1, 2024
Repo size
27.3 MB
Likes
0
Public
Click a slice to open those files.
.safetensors27.3 MB · 97%
From the Hugging Face model README
This repository contains a unique poetry generation model that blends stylistic and emotional nuances from selected poets. Our focus is on exploring the creative capacities of large language models, using fine-tuning techniques to emulate and merge the poetic styles of celebrated authors.
The Poetry Fusion model is a transformer-based language model fine-tuned to produce poetry that captures elements from various renowned poets. The model aims to blend linguistic, thematic, and emotional features, creating novel verses that resonate with the stylistic depth of human poetry.
This model was fine-tuned to capture and fuse the stylistic nuances of poets such as:
These poets were chosen for their distinctive styles, themes, and emotional depth, providing a rich spectrum of characteristics for the model to emulate and blend.
The model is suited for tasks involving poetry generation, with particular relevance for creative applications in AI, literature studies, or projects within the humanities.
This model is not designed for fact-based or research-intensive tasks as it prioritizes stylistic and emotional resonance over factual accuracy.
This model is influenced by the poetic styles and linguistic nuances specific to a Western, English-speaking tradition, as represented in its training data. Users should be mindful of potential stylistic biases and limitations in cultural diversity.
For broader applicability, further fine-tuning with culturally diverse poetic datasets may be necessary. We recommend awareness of inherent biases in creative projects and contextually adjusting outputs as needed.
You may find all necessary information about the model deployment in the Jupyter Notebook "Poetry_Fusion_using_Llama_3.2.ipynb".
You may find all necessary information about the model training in the Jupyter Notebook "Poetry_Fusion_using_Llama_3.2.ipynb".
The model was fine-tuned on a subset of a larger poetry dataset "LLM_Dataset.csv" which may be found in the repository files. It includes 524 English poems, with 227 poems specifically from our selected poets. This curated set allowed the model to focus on the unique attributes of each poet’s work while providing additional data for improved generalization.
To achieve high-quality output, the model was fine-tuned using the following parameters:
alpha=16, dropout=0.1, r=64The model’s effectiveness was assessed through both human and automated evaluations. Human evaluators examined poetic characteristics and emotional depth, using a framework based on stylistic features specific to each poet in our selection. Additionally, GPT-4 cross-validated stylistic adherence.
Evaluation focused on creativity, coherence, thematic depth, and stylistic consistency to gauge the model’s capacity to blend distinct poetic styles effectively.
This project is licensed under the Meta's Llama 3.2 Community License.
If you reference this work, please use the following citation:
BibTeX:
@article{poetry_fusion_project,
author = {Mladenova, Natalie and Kozikowska, Karolina and Karbowski, Kajetan},
title = {Poetry Fusion - The Generation of New Poems Based on a Selection of Authors},
year = {2024},
institution = {University of Groningen},
}
For questions or further information, please reach out to the authors: