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

Finetuned seamless-m4t-medium for darija speech translation

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

• Loss: 0.9487
• Bleu: 0.6520

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
11.4385	12.5	50	7.7186	0.0
7.0056	25.0	100	4.6583	0.0
4.7721	37.5	150	3.4895	0.0402
3.9618	50.0	200	2.7834	0.0324
3.2307	62.5	250	2.3169	0.0601
2.9341	75.0	300	2.1102	0.0969
2.5517	87.5	350	1.9636	0.1041
2.3681	100.0	400	1.8770	0.0888
2.1031	112.5	450	1.7589	0.1302
2.1191	125.0	500	1.6578	0.1765
1.9185	137.5	550	1.5659	0.1802
1.9021	150.0	600	1.4514	0.4482
2.0155	162.5	650	1.3543	0.3924
1.8151	175.0	700	1.3195	0.3651
1.7461	187.5	750	1.1878	0.4723
1.6512	200.0	800	1.1107	0.5124
1.6378	212.5	850	1.0550	0.5922
1.4851	225.0	900	0.9948	0.6548
1.5016	237.5	950	0.9596	0.6390
1.4868	250.0	1000	0.9487	0.6520

Framework versions

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