hamsa-tiny-finetuned-qasr

This model is a fine-tuned version of openai/whisper-tiny on the nadsoft/QASR-Speech-Resource default dataset. It achieves the following results on the evaluation set:

• Loss: 0.3310
• Wer: 25.4515

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: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 150000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.643	0.1	2500	0.6272	51.4156
0.5445	0.2	5000	0.5443	40.7508
0.4944	0.3	7500	0.5005	38.5676
0.4722	0.4	10000	0.4747	39.1490
0.4659	0.5	12500	0.4541	35.6867
0.4261	0.6	15000	0.4383	36.0877
0.4166	0.7	17500	0.4257	31.8968
0.4051	0.8	20000	0.4160	32.5898
0.4107	0.9	22500	0.4070	32.9291
0.3753	1.0	25000	0.3996	30.2095
0.3755	1.1	27500	0.3943	32.4497
0.3749	1.2	30000	0.3893	31.3320
0.3697	1.3	32500	0.3856	30.2024
0.3574	1.4	35000	0.3802	27.4662
0.3583	1.5	37500	0.3774	28.9257
0.3619	1.6	40000	0.3731	28.9447
0.3414	1.7	42500	0.3702	27.6751
0.3465	1.8	45000	0.3667	27.2716
0.3489	1.9	47500	0.3640	25.7695
0.3173	2.0	50000	0.3623	26.2773
0.3227	2.11	52500	0.3608	25.5844
0.3236	2.21	55000	0.3592	26.8564
0.324	2.31	57500	0.3565	27.4639
0.3315	2.41	60000	0.3555	26.7187
0.3238	2.51	62500	0.3531	26.3343
0.3406	2.61	65000	0.3513	26.4031
0.3214	2.71	67500	0.3496	25.1999
0.3197	2.81	70000	0.3481	25.4657
0.3232	2.91	72500	0.3463	24.6684
0.3136	3.01	75000	0.3456	25.8668
0.3082	3.11	77500	0.3445	26.3248
0.3058	3.21	80000	0.3439	25.3874
0.3217	3.31	82500	0.3434	25.1857
0.3158	3.41	85000	0.3417	24.5521
0.3021	3.51	87500	0.3414	25.6295
0.2912	3.61	90000	0.3405	24.7941
0.281	3.71	92500	0.3402	24.5426
0.3017	3.81	95000	0.3391	25.1809
0.2986	3.91	97500	0.3387	25.1145
0.2996	4.01	100000	0.3377	24.6185
0.2734	4.11	102500	0.3374	24.7229
0.3088	4.21	105000	0.3373	24.2578
0.2794	4.31	107500	0.3361	25.6532
0.2988	4.41	110000	0.3357	25.7813
0.3085	4.51	112500	0.3352	24.8345
0.2888	4.61	115000	0.3346	24.5687
0.2923	4.71	117500	0.3342	25.0006
0.2782	4.81	120000	0.3336	25.7766
0.2948	4.91	122500	0.3334	25.2355
0.2791	5.01	125000	0.3329	25.6057
0.2988	5.11	127500	0.3333	25.6129
0.2933	5.21	130000	0.3330	25.7291
0.2801	5.31	132500	0.3321	25.7529
0.2885	5.41	135000	0.3325	25.7861
0.2953	5.51	137500	0.3319	25.0742
0.2677	5.61	140000	0.3319	25.2379
0.2833	5.71	142500	0.3315	25.5749
0.2923	5.81	145000	0.3313	25.6627
0.2602	5.91	147500	0.3311	25.4467
0.2757	6.01	150000	0.3310	25.4515

Framework versions

• Transformers 4.37.0.dev0
• Pytorch 2.1.2+cu121
• Datasets 2.16.2.dev0
• Tokenizers 0.15.0