Marialab/finetuned-whisper-large-v3-5000-step

Finetuned Whisper large-v3 for darija speech translation

This model is a fine-tuned version of openai/whisper-large-v3 on the Darija-C dataset. It achieves the following results on the evaluation set:

• Loss: 0.0000
• Bleu: 0.7440

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: 1
• eval_batch_size: 1
• 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
3.0209	0.8333	50	1.5160	0.0385
1.4379	1.6667	100	0.7678	0.1508
0.8536	2.5	150	0.3769	0.2512
0.4203	3.3333	200	0.2265	0.4909
0.3585	4.1667	250	0.0945	0.5903
0.1791	5.0	300	0.0498	0.6521
0.0894	5.8333	350	0.0388	0.6798
0.0827	6.6667	400	0.0185	0.7029
0.0119	7.5	450	0.0266	0.7213
0.0255	8.3333	500	0.0232	0.7021
0.0296	9.1667	550	0.0002	0.7440
0.0082	10.0	600	0.0001	0.7440
0.0001	10.8333	650	0.0001	0.7440
0.0001	11.6667	700	0.0000	0.7440
0.0	12.5	750	0.0000	0.7440
0.0	13.3333	800	0.0000	0.7440
0.0	14.1667	850	0.0000	0.7440
0.0	15.0	900	0.0000	0.7440
0.0	15.8333	950	0.0000	0.7440
0.0	16.6667	1000	0.0000	0.7440

Framework versions

• Transformers 4.47.0
• Pytorch 2.5.1+cu121
• Datasets 2.19.2
• Tokenizers 0.21.0