miosipof/whisper-small-ft-balbus-sep28k-v1.3

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

• Loss: 0.1212
• Accuracy: 0.8145

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: 32
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.5
• training_steps: 1000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
0.171	0.1253	50	0.1711	0.5641
0.1689	0.2506	100	0.1662	0.6058
0.1601	0.3759	150	0.1529	0.6846
0.135	0.5013	200	0.1231	0.7692
0.1173	0.6266	250	0.1155	0.7932
0.105	0.7519	300	0.1101	0.8032
0.1101	0.8772	350	0.1114	0.7996
0.1058	1.0025	400	0.1082	0.8076
0.0926	1.1278	450	0.1126	0.7969
0.0952	1.2531	500	0.1087	0.8088
0.0958	1.3784	550	0.1145	0.7999
0.0971	1.5038	600	0.1080	0.8156
0.0922	1.6291	650	0.1130	0.8121
0.095	1.7544	700	0.1100	0.8066
0.0962	1.8797	750	0.1090	0.8186
0.0925	2.0050	800	0.1052	0.8176
0.0681	2.1303	850	0.1171	0.8115
0.0624	2.2556	900	0.1307	0.8060
0.0651	2.3810	950	0.1223	0.8123
0.0598	2.5063	1000	0.1212	0.8145

Framework versions

• Transformers 4.45.2
• Pytorch 2.2.0
• Datasets 3.2.0
• Tokenizers 0.20.3