whisper-mediumFT-mixat-tri-arabic

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: 0.9811
• Wer: 43.1794

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: 2
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 16
• 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
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.84	0.9970	209	0.5599	53.9028
0.4543	1.9970	418	0.5139	47.1650
0.298	2.9970	627	0.5121	44.6383
0.1765	3.9970	836	0.5391	44.3977
0.1002	4.9970	1045	0.5863	45.6008
0.0583	5.9970	1254	0.6148	45.5106
0.0345	6.9970	1463	0.6591	44.1570
0.0268	7.9970	1672	0.6931	43.4351
0.0194	8.9970	1881	0.7197	43.2245
0.0148	9.9970	2090	0.7403	43.1644
0.0134	10.9970	2299	0.7233	42.7282
0.0102	11.9970	2508	0.7669	43.2997
0.0096	12.9970	2717	0.7598	44.3074
0.0083	13.9970	2926	0.7952	42.7583
0.0074	14.9970	3135	0.8091	43.7058
0.0075	15.9970	3344	0.7787	42.9237
0.0073	16.9970	3553	0.8204	42.9538
0.0054	17.9970	3762	0.8049	43.5554
0.0057	18.9970	3971	0.8266	41.9913
0.0057	19.9970	4180	0.8285	44.2773
0.0054	20.9970	4389	0.8425	42.2169
0.0036	21.9970	4598	0.8491	42.5478
0.0037	22.9970	4807	0.8568	44.1420
0.0038	23.9970	5016	0.8606	42.2620
0.0037	24.9970	5225	0.8653	42.4274
0.0042	25.9970	5434	0.8764	42.5778
0.0042	26.9970	5643	0.8698	41.5250
0.0044	27.9970	5852	0.8742	42.7733
0.0038	28.9970	6061	0.8882	43.4050
0.0031	29.9970	6270	0.8963	42.6230
0.0029	30.9970	6479	0.9003	43.3449
0.0027	31.9970	6688	0.8980	42.8937
0.0019	32.9970	6897	0.9029	42.3974
0.0025	33.9970	7106	0.9040	42.4726
0.0033	34.9970	7315	0.9252	43.2997
0.003	35.9970	7524	0.9288	42.9689
0.0028	36.9970	7733	0.9205	42.3071
0.0018	37.9970	7942	0.9147	43.0741
0.0023	38.9970	8151	0.9307	42.8335
0.0018	39.9970	8360	0.9242	43.7509
0.0026	40.9970	8569	0.9251	43.3749
0.0031	41.9970	8778	0.9158	44.0066
0.0017	42.9970	8987	0.9399	42.7733
0.0011	43.9970	9196	0.9310	43.0290
0.0011	44.9970	9405	0.9344	43.0892
0.0009	45.9970	9614	0.9569	43.8412
0.0019	46.9970	9823	0.9301	46.0220
0.0018	47.9970	10032	0.9519	43.9314
0.0017	48.9970	10241	0.9396	42.8485
0.0021	49.9970	10450	0.9303	45.0594
0.0024	50.9970	10659	0.9468	43.9615
0.0012	51.9970	10868	0.9559	43.2396
0.0008	52.9970	11077	0.9808	43.6457
0.0005	53.9970	11286	0.9560	42.9538
0.0005	54.9970	11495	0.9500	42.3673
0.0009	55.9970	11704	0.9405	42.5628
0.0012	56.9970	11913	0.9575	42.6230
0.0009	57.9970	12122	0.9700	43.4501
0.001	58.9970	12331	0.9778	43.5404
0.0014	59.9970	12540	0.9562	42.9388
0.0007	60.9970	12749	0.9609	43.4351
0.001	61.4914	12852	0.9811	43.1794

Framework versions

• Transformers 4.50.3
• Pytorch 2.4.1
• Datasets 2.16.1
• Tokenizers 0.21.1