Marialab/finetuned-seamless-m4T-large-1000-step

Finetuned seamless-m4t-large for darija speech translation

This model is a fine-tuned version of facebook/seamless-m4t-large on the Darija-C dataset. It achieves the following results on the evaluation set:

• Loss: 1.5449
• Bleu: 0.1562

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: 16
• eval_batch_size: 4
• seed: 42
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100
• training_steps: 1000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Bleu
10.1923	12.5	50	7.8012	0.0
7.0388	25.0	100	4.9095	0.0023
4.1866	37.5	150	3.0521	0.0276
3.1409	50.0	200	2.4410	0.0653
2.581	62.5	250	2.1347	0.0795
2.2369	75.0	300	2.0097	0.0765
2.171	87.5	350	1.9385	0.0637
2.0371	100.0	400	1.8743	0.0907
1.9672	112.5	450	1.8273	0.0943
1.9463	125.0	500	1.7789	0.1028
1.8821	137.5	550	1.7346	0.1137
1.8424	150.0	600	1.7088	0.1208
1.836	162.5	650	1.6618	0.1470
1.757	175.0	700	1.6311	0.1347
1.7532	187.5	750	1.6137	0.1561
1.8539	200.0	800	1.5977	0.1184
1.7176	212.5	850	1.5758	0.1205
1.6944	225.0	900	1.5604	0.1396
1.7136	237.5	950	1.5521	0.1562
1.7354	250.0	1000	1.5449	0.1562

Framework versions

• Transformers 4.47.1
• Pytorch 2.5.1+cu121
• Datasets 3.2.0
• Tokenizers 0.21.0