my_medium_wspr

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

• Loss: 0.0939
• Wer: 11.9372

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: 1
• eval_batch_size: 1
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 4
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 3
• training_steps: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
2.6756	0.0020	3	2.8119	123.2984
2.7362	0.0041	6	2.1522	102.6702
1.0677	0.0061	9	1.2918	79.4764
1.1569	0.0081	12	0.8556	59.3717
1.0248	0.0102	15	0.7711	230.4712
0.5058	0.0122	18	0.8485	264.3979
0.4122	0.0143	21	0.7497	244.7644
0.4989	0.0163	24	0.7179	53.7173
0.7726	0.0183	27	0.8902	60.0524
0.793	0.0204	30	0.7948	52.2513
0.7562	0.0224	33	0.5553	41.5183
1.0738	0.0244	36	0.3708	36.7016
0.2565	0.0265	39	0.4904	40.5236
0.4824	0.0285	42	0.4613	44.7644
0.4291	0.0305	45	0.4332	56.0733
0.1382	0.0326	48	0.4353	47.6440
0.2113	0.0346	51	0.4286	37.7487
0.2521	0.0367	54	0.4604	38.4293
0.9556	0.0387	57	0.4387	43.4031
0.5493	0.0407	60	0.2983	30.8901
0.1505	0.0428	63	0.2391	26.5969
0.0664	0.0448	66	0.2358	27.6963
0.2826	0.0468	69	0.2045	27.9058
0.2103	0.0489	72	0.1674	20.1571
0.3009	0.0509	75	0.1725	17.9581
0.2875	0.0530	78	0.1275	18.1675
0.0258	0.0550	81	0.1209	16.6492
0.0935	0.0570	84	0.1545	19.0052
0.26	0.0591	87	0.1516	18.7435
0.0844	0.0611	90	0.1500	19.7906
0.0386	0.0631	93	0.1495	20.2618
0.403	0.0652	96	0.1036	12.9319
0.0091	0.0672	99	0.0939	11.9372

Framework versions

• Transformers 4.42.3
• Pytorch 2.3.1+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1