wav2vec-large-zh

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 14.5731
• Wer: 1.0

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: 5e-06
• train_batch_size: 4
• eval_batch_size: 2
• 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
• training_steps: 4000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
121.3961	0.1039	100	124.9288	1.0
123.4081	0.2078	200	123.6992	1.0
107.9594	0.3117	300	93.1991	1.0
88.687	0.4156	400	74.7901	1.0
76.5985	0.5195	500	65.0490	1.0
67.3284	0.6234	600	59.1068	1.0
60.559	0.7273	700	54.5696	1.0
57.6745	0.8312	800	50.8430	1.0
52.8858	0.9351	900	47.6484	1.0
47.363	1.0390	1000	44.8001	1.0
47.1688	1.1429	1100	42.1867	1.0
44.9003	1.2468	1200	39.8359	1.0
41.8644	1.3506	1300	37.6657	1.0
39.7981	1.4545	1400	35.5738	1.0
36.9159	1.5584	1500	33.6546	1.0
34.9118	1.6623	1600	31.8388	1.0
33.5765	1.7662	1700	30.1401	1.0
31.0713	1.8701	1800	28.5309	1.0
30.3748	1.9740	1900	27.0503	1.0
29.2488	2.0779	2000	25.6837	1.0
26.2383	2.1818	2100	24.4220	1.0
25.098	2.2857	2200	23.2505	1.0
25.1944	2.3896	2300	22.1770	1.0
24.8906	2.4935	2400	21.2028	1.0
22.4045	2.5974	2500	20.3080	1.0
20.9461	2.7013	2600	19.4800	1.0
20.5321	2.8052	2700	18.7341	1.0
19.8295	2.9091	2800	18.0642	1.0
19.3534	3.0130	2900	17.4707	1.0
18.5557	3.1169	3000	16.9370	1.0
18.1417	3.2208	3100	16.4655	1.0
17.497	3.3247	3200	16.0514	1.0
16.9942	3.4286	3300	15.6962	1.0
17.0943	3.5325	3400	15.3935	1.0
16.4386	3.6364	3500	15.1432	1.0
15.6441	3.7403	3600	14.9392	1.0
16.0257	3.8442	3700	14.7813	1.0
15.725	3.9481	3800	14.6680	1.0
16.0983	4.0519	3900	14.5990	1.0
15.3737	4.1558	4000	14.5731	1.0

Framework versions

• Transformers 4.44.2
• Pytorch 2.4.1+cu121
• Datasets 3.1.0
• Tokenizers 0.19.1