katyayego/w2v2-xlsr-eng-zh-cs

wav2vec2-large-xlsr-demo

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on ASCEND (a Mandarin-English codeswitching dataset). It achieves the following results on the evaluation set:

• Loss: 1.6751
• Wer: 0.7846

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: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 1000
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
10.5108	0.5701	500	15.4314	1.0
5.61	1.1403	1000	11.3143	1.0
5.546	1.7104	1500	9.6388	1.0
5.1105	2.2805	2000	5.9376	1.0
4.9007	2.8506	2500	5.7582	1.0
4.5876	3.4208	3000	6.7190	1.0
4.3145	3.9909	3500	4.6527	1.0
3.5332	4.5610	4000	3.3622	1.0011
2.9071	5.1311	4500	2.7400	1.0176
2.5077	5.7013	5000	2.8219	0.9460
2.4145	6.2714	5500	2.4336	0.9570
2.2432	6.8415	6000	2.1500	0.9169
2.1376	7.4116	6500	2.1445	0.8930
2.0841	7.9818	7000	2.1312	0.8864
1.8288	8.5519	7500	1.9040	0.8728
1.6863	9.1220	8000	1.8913	0.8434
1.7453	9.6921	8500	2.1214	0.8507
1.6896	10.2623	9000	1.8329	0.8548
1.6063	10.8324	9500	1.8248	0.8386
1.3838	11.4025	10000	1.7811	0.8379
1.5255	11.9726	10500	2.3148	0.8390
1.4269	12.5428	11000	2.1530	0.8184
1.3452	13.1129	11500	1.7208	0.8221
1.35	13.6830	12000	1.8269	0.8290
1.3656	14.2531	12500	1.6902	0.8313
1.2036	14.8233	13000	2.0816	0.8206
1.2144	15.3934	13500	1.7623	0.8103
1.1648	15.9635	14000	1.7197	0.8154
1.1341	16.5336	14500	1.7560	0.8110
1.0716	17.1038	15000	1.7750	0.8099
1.1187	17.6739	15500	1.7946	0.8180
1.0633	18.2440	16000	1.7877	0.7996
1.0069	18.8141	16500	1.8482	0.8243
0.9703	19.3843	17000	1.6073	0.7960
1.0122	19.9544	17500	1.7191	0.8099
0.9993	20.5245	18000	1.7208	0.7956
0.9861	21.0946	18500	1.6628	0.7949
0.9621	21.6648	19000	1.7685	0.7930
0.8936	22.2349	19500	1.7232	0.8026
0.888	22.8050	20000	1.7204	0.8015
0.9027	23.3751	20500	1.7844	0.7923
0.8808	23.9453	21000	1.7159	0.7945
0.8652	24.5154	21500	1.6887	0.7934
0.7545	25.0855	22000	1.6633	0.7937
0.7664	25.6556	22500	1.6745	0.7919
0.7518	26.2258	23000	1.7122	0.7930
0.8475	26.7959	23500	1.6901	0.7868
0.7527	27.3660	24000	1.6937	0.7835
0.7531	27.9361	24500	1.6835	0.7820
0.7686	28.5063	25000	1.6734	0.7901
0.7525	29.0764	25500	1.6766	0.7868
0.7765	29.6465	26000	1.6751	0.7846

Framework versions

• Transformers 4.41.2
• Pytorch 2.3.0
• Datasets 2.19.2
• Tokenizers 0.19.1