whisper-small-CV_Fleurs_AMMI_ALFFA-sw-1hrs-v3

This model is a fine-tuned version of openai/whisper-small on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.5122
• Wer: 0.5874
• Cer: 0.2269

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: 2
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• optimizer: Use OptimizerNames.ADAMW_HF with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.05
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
4.0507	1.0	72	2.1583	1.0154	0.4089
1.4298	2.0	144	1.4889	0.9208	0.3593
0.9281	3.0	216	1.3124	0.8730	0.3546
0.5578	4.0	288	1.2409	0.8739	0.3717
0.3041	5.0	360	1.2416	0.8152	0.3435
0.1581	6.0	432	1.2562	0.7712	0.3292
0.0914	7.0	504	1.3176	0.6153	0.2296
0.0494	8.0	576	1.2867	0.6931	0.2864
0.027	9.0	648	1.3295	0.6288	0.2477
0.0159	10.0	720	1.3462	0.6035	0.2336
0.0119	11.0	792	1.3529	0.6573	0.2691
0.0095	12.0	864	1.3698	0.6083	0.2441
0.0057	13.0	936	1.3639	0.6362	0.2629
0.0039	14.0	1008	1.3545	0.6142	0.2460
0.0033	15.0	1080	1.4090	0.5793	0.2234
0.0042	16.0	1152	1.3781	0.5798	0.2258
0.0018	17.0	1224	1.3992	0.6388	0.2664
0.0012	18.0	1296	1.4246	0.5849	0.2284
0.0017	19.0	1368	1.4143	0.5828	0.2340
0.0015	20.0	1440	1.4234	0.5896	0.2307
0.0007	21.0	1512	1.4385	0.5679	0.2176
0.0014	22.0	1584	1.4416	0.5867	0.2285
0.0004	23.0	1656	1.4561	0.5913	0.2300
0.0004	24.0	1728	1.4588	0.5882	0.2282
0.0003	25.0	1800	1.4638	0.5916	0.2313
0.0003	26.0	1872	1.4692	0.5945	0.2339
0.0003	27.0	1944	1.4739	0.5920	0.2312
0.0003	28.0	2016	1.4782	0.5905	0.2307
0.0002	29.0	2088	1.4828	0.5869	0.2276
0.0002	30.0	2160	1.4890	0.5869	0.2274
0.0002	31.0	2232	1.4932	0.5867	0.2273
0.0002	32.0	2304	1.4971	0.5869	0.2274
0.0002	33.0	2376	1.4999	0.5872	0.2275
0.0002	34.0	2448	1.5048	0.5913	0.2301
0.0002	35.0	2520	1.5080	0.5916	0.2323
0.0002	36.0	2592	1.5122	0.5874	0.2269

Framework versions

• Transformers 4.48.0
• Pytorch 2.5.1+cu124
• Datasets 3.2.0
• Tokenizers 0.21.0