lmejias/whisper-small-atc

WhisperATCSmall

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

• Loss: 8.2932
• Wer Ortho: 76.8786
• Wer: 77.1429

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: 0.0001
• train_batch_size: 32
• eval_batch_size: 32
• 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: 500
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
No log	10.0	10	2.4187	111.5607	105.7143
No log	20.0	20	2.4427	71.0983	71.4286
1.3246	30.0	30	2.9333	82.6590	82.8571
1.3246	40.0	40	2.3413	79.7688	79.4286
0.0534	50.0	50	2.7293	194.7977	193.1429
0.0534	60.0	60	3.9019	89.0173	89.1429
0.0534	70.0	70	5.5183	79.7688	80.0
0.0119	80.0	80	6.4220	323.6994	321.7143
0.0119	90.0	90	5.2189	654.9133	646.8571
0.0614	100.0	100	7.4941	435.2601	430.2857
0.0614	110.0	110	7.0980	82.0809	82.2857
0.0614	120.0	120	6.3404	204.0462	202.8571
0.0	130.0	130	7.3077	98.8439	98.8571
0.0	140.0	140	7.1859	83.2370	83.4286
0.0	150.0	150	7.1508	85.5491	85.7143
0.0	160.0	160	6.9526	79.1908	79.4286
0.0	170.0	170	6.8956	84.9711	85.1429
0.0	180.0	180	6.8715	79.7688	80.0
0.0	190.0	190	7.1711	83.2370	83.4286
0.0	200.0	200	7.2134	83.2370	83.4286
0.0	210.0	210	7.2450	84.9711	85.1429
0.0	220.0	220	7.2228	83.2370	83.4286
0.0	230.0	230	7.2169	80.3468	80.5714
0.0	240.0	240	7.3043	78.0347	78.2857
0.0	250.0	250	7.4497	80.9249	81.1429
0.0	260.0	260	7.6275	79.7688	80.0
0.0	270.0	270	7.4590	80.3468	80.5714
0.0	280.0	280	7.3765	80.3468	80.5714
0.0	290.0	290	7.1675	76.8786	77.1429
0.0	300.0	300	7.1708	78.6127	78.8571
0.0	310.0	310	7.2078	79.7688	80.0
0.0	320.0	320	7.1781	78.6127	78.8571
0.0	330.0	330	7.3729	82.6590	82.8571
0.0	340.0	340	7.9443	83.8150	84.0
0.0	350.0	350	7.7305	83.8150	84.0
0.0	360.0	360	7.7460	84.3931	84.5714
0.0	370.0	370	7.4392	86.1272	86.2857
0.0	380.0	380	7.4408	81.5029	81.7143
0.0	390.0	390	7.2765	84.3931	84.5714
0.0	400.0	400	7.7561	82.0809	82.2857
0.0	410.0	410	7.1968	82.0809	82.2857
0.0	420.0	420	7.1473	80.9249	81.1429
0.0	430.0	430	7.2199	81.5029	81.7143
0.0	440.0	440	7.3039	79.1908	79.4286
0.0	450.0	450	7.3708	75.1445	75.4286
0.0	460.0	460	8.0785	79.1908	79.4286
0.0	470.0	470	8.3612	74.5665	74.8571
0.0	480.0	480	8.4102	77.4566	77.7143
0.0	490.0	490	8.3026	74.5665	74.8571
0.0	500.0	500	8.2932	76.8786	77.1429

Framework versions

• Transformers 4.51.3
• Pytorch 2.9.1+cu128
• Datasets 4.4.1
• Tokenizers 0.21.0