whisper-small-si-bank-v4

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: 0.8527
• Wer Ortho: 81.5217
• Wer: 68.4337

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
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• 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	0.8	2	2.3673	183.6957	250.3614
No log	1.4	4	2.3668	183.6957	250.3614
No log	2.0	6	2.3480	173.3696	254.6988
No log	2.8	8	2.2898	184.2391	254.2169
No log	3.4	10	2.1772	174.4565	255.6627
No log	4.0	12	2.0396	202.7174	240.0
No log	4.8	14	1.9599	225.0	236.6265
No log	5.4	16	1.8790	227.1739	220.9639
No log	6.0	18	1.7862	264.1304	243.3735
No log	6.8	20	1.7070	217.3913	232.5301
No log	7.4	22	1.6313	167.3913	227.7108
No log	8.0	24	1.5601	157.6087	230.1205
1.679	8.8	26	1.5093	162.5	246.7470
1.679	9.4	28	1.4718	178.8043	224.8193
1.679	10.0	30	1.4363	154.8913	204.8193
1.679	10.8	32	1.4004	148.3696	172.2892
1.679	11.4	34	1.3628	125.0	153.9759
1.679	12.0	36	1.3214	104.8913	114.4578
1.679	12.8	38	1.2853	117.3913	126.2651
1.679	13.4	40	1.2414	103.8043	128.1928
1.679	14.0	42	1.2084	93.4783	114.4578
1.679	14.8	44	1.1600	93.4783	107.2289
1.679	15.4	46	1.1256	92.3913	92.0482
1.679	16.0	48	1.0850	91.3043	90.8434
0.8861	16.8	50	1.0483	84.7826	85.0602
0.8861	17.4	52	1.0226	84.2391	84.3373
0.8861	18.0	54	0.9942	90.2174	80.0
0.8861	18.8	56	0.9647	82.6087	74.4578
0.8861	19.4	58	0.9485	92.9348	78.7952
0.8861	20.0	60	0.9217	80.4348	79.2771
0.8861	20.8	62	0.9001	78.2609	66.5060
0.8861	21.4	64	0.9018	78.8043	75.6627
0.8861	22.0	66	0.8823	86.9565	68.6747
0.8861	22.8	68	0.8551	76.0870	66.0241
0.8861	23.4	70	0.8548	77.7174	65.0602
0.8861	24.0	72	0.8346	77.1739	64.3373
0.8861	24.8	74	0.8190	80.4348	67.2289
0.294	25.4	76	0.8248	78.2609	62.4096
0.294	26.0	78	0.8349	77.1739	64.8193
0.294	26.8	80	0.8284	76.0870	64.3373
0.294	27.4	82	0.8440	77.7174	61.6867
0.294	28.0	84	0.8471	77.1739	62.4096
0.294	28.8	86	0.8378	76.0870	62.6506
0.294	29.4	88	0.8577	76.6304	60.2410
0.294	30.0	90	0.8401	74.4565	57.8313
0.294	30.8	92	0.8487	73.3696	60.0
0.294	31.4	94	0.8454	73.9130	55.6627
0.294	32.0	96	0.8615	73.9130	60.4819
0.294	32.8	98	0.8593	79.3478	64.3373
0.1107	33.4	100	0.8527	81.5217	68.4337

Framework versions

• Transformers 4.48.3
• Pytorch 2.5.1+cu124
• Datasets 3.3.2
• Tokenizers 0.21.0