whisper-small-si

This model is a fine-tuned version of openai/whisper-small on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.8059
• Wer Ortho: 79.8913
• Wer: 68.1928

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: 16
• 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: constant_with_warmup
• lr_scheduler_warmup_steps: 50
• training_steps: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
No log	0.4	2	2.3675	183.6957	250.3614
No log	0.8	4	2.3670	183.6957	250.3614
No log	1.2	6	2.3486	173.3696	254.6988
No log	1.6	8	2.2915	184.2391	254.2169
No log	2.0	10	2.1795	179.8913	255.1807
No log	2.4	12	2.0471	203.2609	239.5181
No log	2.8	14	1.9688	225.0	236.8675
No log	3.2	16	1.8879	228.8043	220.7229
No log	3.6	18	1.8019	259.7826	249.1566
No log	4.0	20	1.7219	242.3913	239.0361
No log	4.4	22	1.6463	178.2609	230.8434
No log	4.8	24	1.5752	156.5217	229.3976
2.0076	5.2	26	1.5220	163.5870	232.5301
2.0076	5.6	28	1.4893	173.9130	241.4458
2.0076	6.0	30	1.4547	188.0435	234.2169
2.0076	6.4	32	1.4254	132.0652	175.1807
2.0076	6.8	34	1.3839	138.5870	156.8675
2.0076	7.2	36	1.3475	117.3913	128.6747
2.0076	7.6	38	1.3150	132.0652	161.6867
2.0076	8.0	40	1.2866	132.0652	168.4337
2.0076	8.4	42	1.2552	119.5652	127.4699
2.0076	8.8	44	1.2086	107.0652	147.2289
2.0076	9.2	46	1.1754	98.3696	112.7711
2.0076	9.6	48	1.1426	92.3913	102.1687
1.1338	10.0	50	1.1153	88.5870	102.6506
1.1338	10.4	52	1.0775	85.3261	79.7590
1.1338	10.8	54	1.0330	84.7826	96.3855
1.1338	11.2	56	1.0069	93.4783	101.2048
1.1338	11.6	58	0.9822	93.4783	75.1807
1.1338	12.0	60	0.9659	82.0652	81.6867
1.1338	12.4	62	0.9490	84.7826	87.7108
1.1338	12.8	64	0.9142	78.8043	73.4940
1.1338	13.2	66	0.8838	88.0435	72.0482
1.1338	13.6	68	0.8767	80.9783	66.2651
1.1338	14.0	70	0.8648	79.3478	66.5060
1.1338	14.4	72	0.8439	80.4348	66.2651
1.1338	14.8	74	0.8417	78.8043	64.8193
0.4647	15.2	76	0.8321	78.8043	71.8072
0.4647	15.6	78	0.8000	73.3696	59.0361
0.4647	16.0	80	0.7904	73.3696	57.1084
0.4647	16.4	82	0.8084	76.0870	69.3976
0.4647	16.8	84	0.7965	77.1739	60.9639
0.4647	17.2	86	0.8039	80.4348	70.8434
0.4647	17.6	88	0.7852	76.6304	65.3012
0.4647	18.0	90	0.7896	78.8043	64.8193
0.4647	18.4	92	0.7829	110.8696	76.1446
0.4647	18.8	94	0.7860	76.6304	60.7229
0.4647	19.2	96	0.8047	74.4565	58.7952
0.4647	19.6	98	0.7939	71.7391	57.3494
0.1678	20.0	100	0.8059	79.8913	68.1928

Framework versions

• Transformers 4.48.2
• Pytorch 2.5.1+cu124
• Datasets 3.2.0
• Tokenizers 0.21.0