Kamshat/wav2vec2-base-issai-colab

Kammi

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the BilalS96/ISSAI_KSC2 dataset. It achieves the following results on the evaluation set:

• Loss: 0.6797
• Wer: 0.3822

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: 0.0003
• train_batch_size: 4
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
5.2923	0.4278	400	4.4181	1.0
3.3189	0.8556	800	3.6105	1.0
3.2115	1.2834	1200	3.3715	1.0
3.148	1.7112	1600	3.1163	1.0
3.0788	2.1390	2000	3.2185	1.0
2.9677	2.5668	2400	2.7724	1.0000
2.3283	2.9947	2800	1.7294	0.9985
1.6653	3.4225	3200	1.3565	0.9627
1.4308	3.8503	3600	1.1434	0.9235
1.2196	4.2781	4000	0.9823	0.8583
1.0644	4.7059	4400	0.8573	0.8191
0.9649	5.1337	4800	0.8064	0.7725
0.849	5.5615	5200	0.7391	0.7389
0.8208	5.9893	5600	0.7014	0.6868
0.6995	6.4171	6000	0.6765	0.6687
0.703	6.8449	6400	0.6347	0.6476
0.6136	7.2727	6800	0.6371	0.6226
0.5957	7.7005	7200	0.6068	0.6000
0.5616	8.1283	7600	0.5877	0.5774
0.5128	8.5561	8000	0.5878	0.5605
0.5093	8.9840	8400	0.5502	0.5469
0.4544	9.4118	8800	0.5823	0.5424
0.4622	9.8396	9200	0.5546	0.5219
0.424	10.2674	9600	0.5910	0.5247
0.4041	10.6952	10000	0.5735	0.5130
0.3956	11.1230	10400	0.5673	0.5005
0.3694	11.5508	10800	0.5336	0.4940
0.3675	11.9786	11200	0.5304	0.4886
0.338	12.4064	11600	0.6132	0.4859
0.3355	12.8342	12000	0.6146	0.4872
0.3251	13.2620	12400	0.5979	0.4753
0.309	13.6898	12800	0.5721	0.4657
0.3065	14.1176	13200	0.5849	0.4598
0.2824	14.5455	13600	0.5872	0.4644
0.2875	14.9733	14000	0.5864	0.4540
0.2663	15.4011	14400	0.5885	0.4513
0.2711	15.8289	14800	0.6090	0.4553
0.2566	16.2567	15200	0.6312	0.4532
0.2524	16.6845	15600	0.6248	0.4450
0.2528	17.1123	16000	0.6329	0.4390
0.2381	17.5401	16400	0.6040	0.4370
0.2336	17.9679	16800	0.5855	0.4327
0.2184	18.3957	17200	0.6107	0.4327
0.2253	18.8235	17600	0.6087	0.4316
0.2169	19.2513	18000	0.6169	0.4261
0.2142	19.6791	18400	0.6025	0.4321
0.2125	20.1070	18800	0.6478	0.4261
0.1994	20.5348	19200	0.6504	0.4238
0.2025	20.9626	19600	0.6580	0.4229
0.1954	21.3904	20000	0.6401	0.4170
0.1939	21.8182	20400	0.6443	0.4119
0.1865	22.2460	20800	0.6588	0.4140
0.1847	22.6738	21200	0.6463	0.4087
0.185	23.1016	21600	0.6490	0.4058
0.1796	23.5294	22000	0.6653	0.4070
0.1745	23.9572	22400	0.6452	0.4042
0.173	24.3850	22800	0.6895	0.4018
0.1653	24.8128	23200	0.6482	0.4017
0.165	25.2406	23600	0.6620	0.3962
0.1622	25.6684	24000	0.6702	0.3971
0.1565	26.0963	24400	0.6899	0.3985
0.1563	26.5241	24800	0.7042	0.3932
0.1555	26.9519	25200	0.7017	0.3931
0.1548	27.3797	25600	0.6751	0.3895
0.1543	27.8075	26000	0.6831	0.3895
0.1464	28.2353	26400	0.6765	0.3842
0.1475	28.6631	26800	0.6842	0.3858
0.144	29.0909	27200	0.6904	0.3851
0.1461	29.5187	27600	0.6821	0.3834
0.1417	29.9465	28000	0.6797	0.3822

Framework versions

• Transformers 4.41.1
• Pytorch 2.3.0+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1