helsinki-opus-de-en-fine-tuned-wmt16-finetuned-src-to-trg

This model is a fine-tuned version of mariav/helsinki-opus-de-en-fine-tuned-wmt16 on the None dataset. It achieves the following results on the evaluation set:

Loss: 0.8597
Rouge1: 64.539
Rouge2: 32.7634
Rougel: 61.3523
Rougelsum: 61.3758
Gen Len: 23.9561
Bleu-1: 64.1391
Bleu-2: 45.1093
Bleu-3: 32.4697
Bleu-4: 24.2684
Meteor: 0.5436

Model description

This model is a fine-tuned version of mariav/helsinki-opus-de-en-fine-tuned-wmt16 on Phoenix Weather dataset (PHOENIX-2014-T).

Intended uses & limitations

The purpose is Neural Machine Translation from German text into German Sign Glosses, which could be used for avatar generation within the Sign Language Production task.

Training and evaluation data

Phoenix Weather dataset (PHOENIX-2014-T)

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 5e-05
train_batch_size: 5
eval_batch_size: 5
seed: 7575
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
num_epochs: 5

Training results

Training Loss	Epoch	Step	Validation Loss	Rouge1	Rouge2	Rougel	Rougelsum	Gen Len	Bleu-1	Bleu-2	Bleu-3	Bleu-4	Meteor
1.1513	1.0	1189	0.9604	61.8236	30.0156	58.9651	58.9484	22.8563	58.6480	40.5508	29.0090	21.2884	0.4961
0.9067	2.0	2378	0.8825	62.8824	30.8604	59.9543	59.9884	22.7564	60.5598	42.0443	29.9532	21.8711	0.5138
0.739	3.0	3567	0.8547	63.8251	31.6294	60.7141	60.7508	24.5219	62.6847	43.6395	31.1174	22.8704	0.5318
0.636	4.0	4756	0.8554	64.5308	32.6897	61.347	61.3929	22.7912	63.0309	44.4786	32.0956	23.8647	0.5369
0.5745	5.0	5945	0.8597	64.539	32.7634	61.3523	61.3758	23.9561	64.1391	45.1093	32.4697	24.2684	0.5436

Framework versions

Transformers 4.30.1
Pytorch 2.0.1+cu118
Datasets 2.12.0
Tokenizers 0.13.3