gg_mdl

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

• Loss: 1.7053
• Cer: 26.9304

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: 4
• eval_batch_size: 4
• 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: 100000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
0.2863	0.8	1000	0.9834	34.5643
0.3725	1.6	2000	0.9299	36.7432
0.3335	2.4	3000	0.9437	32.7097
0.1498	3.2	4000	0.9722	26.1319
0.2081	4.0	5000	0.9881	31.9533
0.213	4.8	6000	1.0197	30.8832
0.094	5.6	7000	1.0486	29.5707
0.0637	6.4	8000	1.0741	26.9211
0.0518	7.2	9000	1.0964	30.4382
0.0512	8.0	10000	1.1179	26.4199
0.0288	8.8	11000	1.1420	27.2981
0.0274	9.6	12000	1.1617	26.9620
0.0255	10.4	13000	1.1779	27.3133
0.0215	11.2	14000	1.2062	25.8813
0.0128	12.0	15000	1.2138	25.8497
0.013	12.8	16000	1.2354	26.8496
0.0054	13.6	17000	1.2323	27.6025
0.0088	14.4	18000	1.2596	25.8228
0.0031	15.2	19000	1.2807	29.0122
0.0093	16.0	20000	1.2865	25.8907
0.0113	16.8	21000	1.2983	28.7241
0.0051	17.6	22000	1.3118	25.8685
0.0019	18.4	23000	1.3225	26.2256
0.0031	19.2	24000	1.3419	25.9586
0.0096	20.0	25000	1.3516	28.7066
0.0051	20.8	26000	1.3419	25.9937
0.0028	21.6	27000	1.3634	28.4256
0.0019	22.4	28000	1.3659	26.7876
0.0041	23.2	29000	1.3855	25.7631
0.005	24.0	30000	1.3848	27.5709
0.0043	24.8	31000	1.3801	27.5252
0.0046	25.6	32000	1.3974	26.5253
0.0017	26.4	33000	1.3992	26.9854
0.0017	27.2	34000	1.4133	26.5405
0.0007	28.0	35000	1.4214	27.7360
0.0009	28.8	36000	1.4275	28.0322
0.0018	29.6	37000	1.4315	26.6939
0.0012	30.4	38000	1.4424	26.2431
0.0007	31.2	39000	1.4498	26.0640
0.0007	32.0	40000	1.4652	27.6891
0.001	32.8	41000	1.4652	26.2478
0.0003	33.6	42000	1.4696	26.8297
0.0004	34.4	43000	1.4603	26.3309
0.0004	35.2	44000	1.4692	26.9234
0.0003	36.0	45000	1.4689	26.7981
0.001	36.8	46000	1.4907	26.5323
0.0015	37.6	47000	1.4897	26.7817
0.0002	38.4	48000	1.4874	26.9093
0.0003	39.2	49000	1.4884	26.8637
0.0009	40.0	50000	1.4854	26.9386
0.001	40.8	51000	1.4978	26.8449
0.0002	41.6	52000	1.5018	27.8132
0.0007	42.4	53000	1.5129	27.7219
0.0002	43.2	54000	1.5252	27.9010
0.0024	44.0	55000	1.5070	25.5617
0.0007	44.8	56000	1.5149	27.3964
0.0025	45.6	57000	1.5287	25.9973
0.0004	46.4	58000	1.5313	27.6294
0.0001	47.2	59000	1.5313	26.6799
0.0005	48.0	60000	1.5478	27.5381
0.0003	48.8	61000	1.5353	27.3402
0.0001	49.6	62000	1.5550	25.4680
0.0001	50.4	63000	1.5463	25.9656
0.0001	51.2	64000	1.5609	26.2935
0.0001	52.0	65000	1.5556	25.8509
0.0012	52.8	66000	1.5704	26.3110
0.0007	53.6	67000	1.5673	26.3087
0.0003	54.4	68000	1.5767	26.2396
0.0	55.2	69000	1.5727	26.2139
0.0001	56.0	70000	1.5723	27.4116
0.0001	56.8	71000	1.5863	26.9082
0.0004	57.6	72000	1.5943	26.4949
0.0006	58.4	73000	1.5944	26.6330
0.0001	59.2	74000	1.5860	28.3659
0.0	60.0	75000	1.5973	26.7759
0.0	60.8	76000	1.6017	27.2278
0.0	61.6	77000	1.6070	26.2619
0.0	62.4	78000	1.6092	27.0030
0.0	63.2	79000	1.6108	26.6576
0.0	64.0	80000	1.6146	25.9387
0.0	64.8	81000	1.6202	25.7291
0.0	65.6	82000	1.6215	27.0042
0.0	66.4	83000	1.6256	27.1915
0.0	67.2	84000	1.6330	26.7677
0.0	68.0	85000	1.6279	26.5803
0.0	68.8	86000	1.6343	26.8625
0.0	69.6	87000	1.6417	26.1296
0.0	70.4	88000	1.6505	26.5874
0.0	71.2	89000	1.6558	26.0640
0.0	72.0	90000	1.6602	25.9469
0.0	72.8	91000	1.6662	26.2338
0.0	73.6	92000	1.6719	26.1460
0.0	74.4	93000	1.6783	26.6576
0.0	75.2	94000	1.6836	26.3099
0.0	76.0	95000	1.6891	26.4984
0.0	76.8	96000	1.6946	26.4328
0.0	77.6	97000	1.6988	26.7056
0.0	78.4	98000	1.7023	26.6049
0.0	79.2	99000	1.7046	27.1821
0.0	80.0	100000	1.7053	26.9304

Framework versions

• Transformers 4.47.0.dev0
• Pytorch 2.4.1
• Datasets 3.1.0
• Tokenizers 0.20.1