This model is a fine-tuned version of LeoLM/leo-hessianai-7b on a dataset of 1048 schlager song lyrics. Schlager songs form a genre of German music and, therefore, the input should be in German as well to ensure best results.

Model description

The model takes a verse (in German) and uses it to generate the text to a schlager song.

Intended uses & limitations

This model is not intended for commercial use of any kind as the dataset used for fine-tuning contains propietary information.

Training and evaluation data

The training and evaluation data was extracted from Spotify via an API.

Training procedure

The model was fine-tuned using a Google colab notebook with a V100 GPU and High RAM. For details, please see the Github repository (https://github.com/NiclasFenton-Wiegleb/schlager-lyrics-bot/tree/main)

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 0.0005
train_batch_size: 4
eval_batch_size: 8
seed: 42
gradient_accumulation_steps: 2
total_train_batch_size: 8
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: constant
lr_scheduler_warmup_ratio: 0.03
num_epochs: 10

Framework versions

Transformers 4.34.0
Pytorch 2.1.0+cu118
Datasets 2.13.0
Tokenizers 0.14.1

niclasfw/schlager-bot-004

schlager-bot-004