lmajnaric/paligemma-architecture-styles
<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->
paligemma-architecture-styles
This model is a fine-tuned version of google/paligemma2-3b-pt-448 on the None dataset.
Model description
More information needed
Intended uses & limitations
More information needed
Training and evaluation data
More information needed
Training procedure
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 1e-05
- trainbatchsize: 1
- evalbatchsize: 1
- seed: 42
- gradientaccumulationsteps: 8
- totaltrainbatch_size: 8
- optimizer: Use OptimizerNames.ADAMWHF with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments
- lrschedulertype: cosine
- lrschedulerwarmup_steps: 20
- num_epochs: 3
Training results
TrainOutput(globalstep=261, trainingloss=1.761135561912681, metrics={'trainruntime': 1063.4627, 'trainsamplespersecond': 1.975, 'trainstepspersecond': 0.245, 'totalflos': 3.156513684279552e+16, 'train_loss': 1.761135561912681, 'epoch': 2.9714285714285715})
Evals on base vs fine-tune
Base model:
Evaluation complete - Accuracy: 0.2400 (240/1000)
Performance by style: Ancient Egyptian architecture: 0.09 (5/57) Art Deco architecture: 0.23 (17/75) Art Nouveau architecture: 0.01 (1/73) Baroque architecture: 0.26 (15/58) Bauhaus architecture: 0.00 (0/58) Brutalism: 0.00 (0/38) Byzantine architecture: 0.34 (17/50) Chicago school architecture: 0.06 (3/51) Colonial architecture: 0.30 (27/89) Deconstructivism: 0.00 (0/38) Gothic architecture: 0.98 (59/60) Greek Revival architecture: 0.45 (26/58) International style: 0.00 (0/66) Neoclassicism: 0.14 (18/125) Postmodern architecture: 0.94 (47/50) Romanesque architecture: 0.09 (5/54) Base model results saved to paligemma448archfinetunestyles/basemodelfoldereval20250316183525.csv
=== EVALUATION RESULTS COMPARISON === Fine-tuned model accuracy: 0.8440 Base model accuracy: 0.2400 Improvement: 0.6040
The checkpoint-176 performs better than the latest checkpoint by .02, even though the training loss is lower on the latest checkpoint.
Framework versions
- Transformers 4.50.0.dev0
- Pytorch 2.6.0+cu124
- Datasets 3.4.0
- Tokenizers 0.21.0
