manhvh2601/whisper_data_self_tiny

Whisper Vietnamese Vu Hoang Manh

This model is a fine-tuned version of openai/whisper-small on the Vietnamese ASR dataset. It achieves the following results on the evaluation set:

• Loss: 0.0549
• Wer: 8.9064

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: 0.0003
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 50
• training_steps: 5250
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.1693	1.0	105	0.1621	224.6900
0.1317	2.0	210	0.0764	26.2683
0.0868	3.0	315	0.1650	26.3247
0.0645	4.0	420	0.0592	43.7993
0.0578	5.0	525	0.0755	41.0372
0.0342	6.0	630	0.0640	29.2559
0.0325	7.0	735	0.0514	13.1342
0.0397	8.0	840	0.0587	37.8241
0.0307	9.0	945	0.0628	32.1308
0.0215	10.0	1050	0.0742	23.8444
0.0268	11.0	1155	0.0710	35.6257
0.009	12.0	1260	0.0539	11.9504
0.0084	13.0	1365	0.0705	16.2345
0.0158	14.0	1470	0.0489	20.6877
0.0122	15.0	1575	0.0582	10.3720
0.0139	16.0	1680	0.0620	17.1928
0.0121	17.0	1785	0.0734	16.0654
0.0055	18.0	1890	0.0719	32.6381
0.0044	19.0	1995	0.0614	83.5964
0.0083	20.0	2100	0.0930	22.3224
0.0081	21.0	2205	0.0872	14.2616
0.0112	22.0	2310	0.0417	17.4183
0.0076	23.0	2415	0.0698	13.6979
0.0043	24.0	2520	0.0428	9.1319
0.0005	25.0	2625	0.0466	9.5829
0.0002	26.0	2730	0.0597	11.2176
0.0	27.0	2835	0.0527	10.0338
0.0	28.0	2940	0.0521	9.4138
0.0	29.0	3045	0.0523	9.4138
0.0	30.0	3150	0.0526	9.3574
0.0	31.0	3255	0.0527	9.4138
0.0	32.0	3360	0.0529	9.3574
0.0	33.0	3465	0.0531	9.3010
0.0	34.0	3570	0.0533	9.2446
0.0	35.0	3675	0.0535	9.1883
0.0	36.0	3780	0.0537	9.1319
0.0	37.0	3885	0.0538	9.0755
0.0	38.0	3990	0.0540	9.0192
0.0	39.0	4095	0.0541	9.0192
0.0	40.0	4200	0.0542	9.0192
0.0	41.0	4305	0.0543	9.0192
0.0	42.0	4410	0.0544	9.0192
0.0	43.0	4515	0.0545	8.9064
0.0	44.0	4620	0.0546	8.9064
0.0	45.0	4725	0.0547	8.9064
0.0	46.0	4830	0.0548	8.9064
0.0	47.0	4935	0.0549	8.9064
0.0	48.0	5040	0.0549	8.9064
0.0	49.0	5145	0.0549	8.7937
0.0	50.0	5250	0.0549	8.9064

Framework versions

• Transformers 4.38.0.dev0
• Pytorch 2.1.0+cu121
• Datasets 2.16.1
• Tokenizers 0.15.1