| --- |
| datasets: |
| - Geremia23/LatinChant |
| language: |
| - la |
| metrics: |
| - accuracy |
| - bleu |
| pipeline_tag: translation |
| --- |
| |
| Trained using [Eole](https://eole-nlp.github.io/eole/) for 90 minutes on an [NVIDIA Quadro RTX 4000](https://www.nvidia.com/en-gb/products/workstations/quadro/rtx-4000/) ([CC](https://developer.nvidia.com/cuda/gpus#:~:text=QUADRO%20RTX%204000) 7.5, 8 GB GDDR6 VRAM, 160W max power) GPU with `cuda_13.3.r13.3/compiler.38244171_0`, `torch 2.13.0a0`. |
|
|
| |Training|| |Validation|| |
| |---|---| -- |---|---| |
| |embeddings| 24936448| |validation BLEU| 64.90217613134918 |
| |encoder| 75524096| |Train perplexity| 4.23818 |
| |decoder| 100702208| |Train accuracy| 99.5824 |
| |generator| 24352| |Sentences processed| 644290 |
| |other| 0| |{'corpus_1'| {'count': 644290, 'index': 69}} |
| |* number of parameters| 201187104| |Average bsz| 681/1015/20 |
| |Trainable parameters | {'torch.float32': 201187104}||Validation perplexity| 53.242 |
| |Non trainable parameters | {}| |Validation accuracy| 60.1233 |
| | * src vocab size | 24352| |
| | * tgt vocab size | 24352| |
| |
