taska-wav2vec-300m-max22-WF-epoch-16-batch-8-whisper-2

This model is a fine-tuned version of whisper-large-v2 on the uriel/audio_data_kaggle_train_taska dataset. It achieves the following results on the evaluation set:

• Loss: 0.8168

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: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine_with_restarts
• lr_scheduler_warmup_steps: 100
• num_epochs: 6
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss
2.4269	0.0888	250	2.2703
1.8282	0.1777	500	1.7029
1.4952	0.2665	750	1.4932
1.4543	0.3553	1000	1.3942
1.319	0.4442	1250	1.3234
1.2642	0.5330	1500	1.2685
1.1427	0.6218	1750	1.2236
1.1739	0.7107	2000	1.1846
1.2141	0.7995	2250	1.1495
1.1487	0.8883	2500	1.1158
1.1129	0.9772	2750	1.0875
1.0279	1.0657	3000	1.0655
1.051	1.1546	3250	1.0451
1.0939	1.2434	3500	1.0276
0.9935	1.3322	3750	1.0116
0.9597	1.4211	4000	0.9981
0.9801	1.5099	4250	0.9833
1.0072	1.5987	4500	0.9723
1.0046	1.6876	4750	0.9610
0.9349	1.7764	5000	0.9499
0.954	1.8652	5250	0.9423
0.9476	1.9541	5500	0.9328
0.9019	2.0426	5750	0.9245
0.9377	2.1315	6000	0.9162
0.8679	2.2203	6250	0.9093
0.9139	2.3091	6500	0.9030
0.8849	2.3980	6750	0.8964
0.8529	2.4868	7000	0.8908
0.8874	2.5756	7250	0.8852
0.9127	2.6645	7500	0.8795
0.8952	2.7533	7750	0.8748
0.8817	2.8421	8000	0.8703
0.8542	2.9310	8250	0.8652
0.8651	3.0195	8500	0.8618
0.9227	3.1084	8750	0.8576
0.8792	3.1972	9000	0.8540
0.8417	3.2860	9250	0.8507
0.8692	3.3749	9500	0.8476
0.8254	3.4637	9750	0.8443
0.8894	3.5525	10000	0.8420
0.8395	3.6414	10250	0.8396
0.8509	3.7302	10500	0.8372
0.8282	3.8190	10750	0.8347
0.8377	3.9079	11000	0.8333
0.8156	3.9967	11250	0.8313
0.829	4.0853	11500	0.8289
0.8247	4.1741	11750	0.8278
0.8182	4.2629	12000	0.8268
0.8489	4.3518	12250	0.8252
0.8392	4.4406	12500	0.8240
0.8272	4.5294	12750	0.8229
0.8044	4.6183	13000	0.8217
0.8042	4.7071	13250	0.8211
0.7989	4.7959	13500	0.8199
0.8282	4.8848	13750	0.8193
0.8137	4.9736	14000	0.8188
0.8156	5.0622	14250	0.8185
0.8356	5.1510	14500	0.8181
0.8436	5.2399	14750	0.8178
0.8414	5.3287	15000	0.8174
0.7825	5.4175	15250	0.8174
0.8374	5.5064	15500	0.8168
0.8261	5.5952	15750	0.8169
0.8161	5.6840	16000	0.8169
0.7703	5.7729	16250	0.8168
0.8152	5.8617	16500	0.8168
0.8479	5.9505	16750	0.8168

Framework versions

• PEFT 0.15.2.dev0
• Transformers 4.53.0.dev0
• Pytorch 2.7.1+cu126
• Datasets 3.6.0
• Tokenizers 0.21.1