IoanRazvan/marian_mt_rom_ro

MarianMT Romani-Romanian NMT

This model is a fine-tuned version of Helsinki-NLP/opus-mt-en-ro on the IoanRazvan/ro-romani dataset. It achieves the following results on the evaluation set:

• Loss: 0.3277
• Bleu: 0.6077

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
• train_batch_size: 64
• eval_batch_size: 32
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 128
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100
• training_steps: 6000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Bleu
0.5152	5.19	1000	0.4979	0.4753
0.4051	10.39	2000	0.3960	0.5528
0.351	15.58	3000	0.3577	0.5847
0.3222	20.78	4000	0.3394	0.5983
0.3108	25.97	5000	0.3305	0.6056
0.3005	31.17	6000	0.3277	0.6077

Framework versions

• Transformers 4.38.2
• Pytorch 2.2.1+cu121
• Datasets 2.18.0
• Tokenizers 0.15.2