zuazo/whisper-medium-eu-cv16_1

Whisper Medium Basque

This model is a fine-tuned version of openai/whisper-medium on the mozilla-foundation/common_voice_16_1 eu dataset. It achieves the following results on the evaluation set:

• Loss: 0.4303
• Wer: 9.2006

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: 4
• total_train_batch_size: 256
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 40000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.0055	10.03	1000	0.2463	11.8425
0.003	20.05	2000	0.2638	11.3178
0.0018	30.08	3000	0.2837	10.9583
0.0009	40.1	4000	0.2768	10.4414
0.0008	50.13	5000	0.2880	10.1776
0.0012	60.15	6000	0.2903	10.0526
0.0002	70.18	7000	0.2909	9.8357
0.0013	80.2	8000	0.2766	9.9392
0.0001	90.23	9000	0.3110	9.3003
0.0	100.25	10000	0.3278	9.3315
0.0	110.28	11000	0.3393	9.3081
0.0	120.3	12000	0.3508	9.2993
0.0	130.33	13000	0.3617	9.3218
0.0	140.35	14000	0.3732	9.3354
0.0	150.38	15000	0.3849	9.3735
0.0	160.4	16000	0.3073	9.3335
0.0	170.43	17000	0.3320	9.3569
0.0	180.45	18000	0.3453	9.3022
0.0	190.48	19000	0.3561	9.3071
0.0	200.5	20000	0.3660	9.2983
0.0	210.53	21000	0.3755	9.2876
0.0	220.55	22000	0.3847	9.4976
0.0	230.58	23000	0.3940	9.5054
0.0	240.6	24000	0.4021	9.4703
0.0	250.63	25000	0.4126	9.4537
0.0	260.65	26000	0.3174	9.2758
0.0	270.68	27000	0.3444	9.2622
0.0	280.7	28000	0.3588	9.2084
0.0	290.73	29000	0.3698	9.3472
0.0	300.75	30000	0.3786	9.3423
0.0	310.78	31000	0.3868	9.3169
0.0	320.8	32000	0.3948	9.3286
0.0	330.83	33000	0.4018	9.3335
0.0	340.85	34000	0.4081	9.3286
0.0	350.88	35000	0.4138	9.3364
0.0	360.9	36000	0.4191	9.3432
0.0	370.93	37000	0.4234	9.3315
0.0	380.95	38000	0.4270	9.3403
0.0	390.98	39000	0.4294	9.2153
0.0	401.0	40000	0.4303	9.2006

Framework versions

• Transformers 4.37.2
• Pytorch 2.2.0+cu121
• Datasets 2.16.1
• Tokenizers 0.15.1