namph204/whisper-base-vi

Whisper base Vi - Nam Phung

This model is a fine-tuned version of openai/whisper-base on the vlsp2020_vinai_100h dataset. It achieves the following results on the evaluation set:

• Loss: 0.3606
• Wer: 16.9148

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: 8
• seed: 42
• optimizer: Use OptimizerNames.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: 500
• training_steps: 15000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.7975	0.0886	250	0.7610	36.8155
0.6074	0.1772	500	0.6467	32.4870
0.5934	0.2658	750	0.5843	29.5521
0.5497	0.3544	1000	0.5450	26.5531
0.5559	0.4429	1250	0.5176	26.0146
0.4872	0.5315	1500	0.4967	25.8677
0.5001	0.6201	1750	0.4795	25.0705
0.4597	0.7087	2000	0.4644	24.5844
0.4507	0.7973	2250	0.4536	22.6308
0.4356	0.8859	2500	0.4412	22.1019
0.4589	0.9745	2750	0.4315	22.3294
0.3347	1.0631	3000	0.4250	21.2764
0.3318	1.1517	3250	0.4204	20.9716
0.3473	1.2403	3500	0.4134	20.9027
0.3358	1.3288	3750	0.4097	20.2717
0.3467	1.4174	4000	0.4034	20.3648
0.3325	1.5060	4250	0.3987	19.7828
0.3396	1.5946	4500	0.3938	20.0876
0.3429	1.6832	4750	0.3897	18.9360
0.3347	1.7718	5000	0.3852	19.5118
0.3318	1.8604	5250	0.3816	19.1070
0.3362	1.9490	5500	0.3765	19.3152
0.3083	2.0376	5750	0.3780	18.7174
0.2372	2.1262	6000	0.3779	18.7188
0.2534	2.2147	6250	0.3742	18.6181
0.271	2.3033	6500	0.3729	18.5588
0.2836	2.3919	6750	0.3718	18.3712
0.2648	2.4805	7000	0.3689	18.3843
0.2678	2.5691	7250	0.3665	17.6009
0.2714	2.6577	7500	0.3652	17.7202
0.2504	2.7463	7750	0.3640	17.9457
0.275	2.8349	8000	0.3631	17.7382
0.2538	2.9235	8250	0.3598	17.3451
0.1795	3.0120	8500	0.3612	17.2499
0.1879	3.1006	8750	0.3648	17.5003
0.1947	3.1892	9000	0.3627	17.2665
0.1968	3.2778	9250	0.3620	17.0700
0.1954	3.3664	9500	0.3621	17.1148
0.1921	3.4550	9750	0.3617	17.0251
0.2068	3.5436	10000	0.3601	17.2162
0.2115	3.6322	10250	0.3604	17.0293
0.2242	3.7208	10500	0.3591	16.8072
0.2015	3.8094	10750	0.3574	17.0858
0.2261	3.8979	11000	0.3573	16.7017
0.2129	3.9865	11250	0.3556	17.1631
0.1739	4.0751	11500	0.3603	16.8362
0.1532	4.1637	11750	0.3603	16.8603
0.1408	4.2523	12000	0.3613	16.8631
0.1743	4.3409	12250	0.3604	16.8196
0.1832	4.4295	12500	0.3613	16.9534
0.1688	4.5181	12750	0.3609	17.0279
0.1767	4.6067	13000	0.3595	17.1865
0.1589	4.6953	13250	0.3596	16.8824
0.1778	4.7838	13500	0.3591	16.8376
0.1806	4.8724	13750	0.3590	16.8714
0.1551	4.9610	14000	0.3591	16.8231
0.163	5.0496	14250	0.3598	16.9541
0.1365	5.1382	14500	0.3604	16.8079
0.1563	5.2268	14750	0.3606	16.9176
0.1429	5.3154	15000	0.3606	16.9148

Framework versions

• Transformers 4.51.3
• Pytorch 2.6.0+cu124
• Datasets 3.5.0
• Tokenizers 0.21.1