1-bpe20k-freeze_cnn-drop.1

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 1.0929
• Wer: 1.0076

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: 2e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• total_train_batch_size: 32
• total_eval_batch_size: 32
• optimizer: Adam with betas=(0.9,0.98) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• training_steps: 200000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
91.1741	0.05	2000	179.2776	1.0
56.7966	0.09	4000	111.7828	1.0
11.4825	0.14	6000	16.6212	1.0
7.5731	0.19	8000	8.3634	1.0
7.3291	0.23	10000	8.3149	1.0
7.1995	0.28	12000	8.1837	1.0
7.1092	0.33	14000	8.1877	1.0
6.9362	0.38	16000	7.8934	1.0
6.7999	0.42	18000	7.7947	1.0
6.7773	0.47	20000	7.8173	1.0
6.7489	0.52	22000	7.8197	1.0117
6.7139	0.56	24000	7.6664	1.0078
6.6024	0.61	26000	7.8397	1.0
6.3451	0.66	28000	7.2039	1.0044
5.7315	0.7	30000	6.3228	1.0039
4.8494	0.75	32000	5.3767	1.0058
4.4103	0.8	34000	4.5216	1.0073
3.8434	0.85	36000	3.9239	1.0125
3.3268	0.89	38000	3.4547	1.0074
2.9034	0.94	40000	3.0617	1.0079
2.5747	0.99	42000	2.7700	1.0108
2.4532	1.03	44000	2.5497	1.0113
2.3169	1.08	46000	2.3780	1.0105
2.2304	1.13	48000	2.2581	1.0075
2.0592	1.17	50000	2.1189	1.0051
2.023	1.22	52000	2.0325	1.0049
1.8011	1.27	54000	1.9284	1.0081
1.817	1.31	56000	1.8635	1.0091
1.6902	1.36	58000	1.7967	1.0076
1.6398	1.41	60000	1.7413	1.0081
1.8097	1.46	62000	1.7066	1.0054
1.5504	1.5	64000	1.6525	1.0074
1.5759	1.55	66000	1.6402	1.0098
1.6652	1.6	68000	1.5884	1.0055
1.5074	1.64	70000	1.5423	1.0066
1.414	1.69	72000	1.5397	1.0095
1.4792	1.74	74000	1.4822	1.0062
1.4648	1.78	76000	1.5023	1.0083
1.4326	1.83	78000	1.4627	1.0070
1.3691	1.88	80000	1.4435	1.0073
1.3624	1.92	82000	1.4290	1.0066
1.5254	1.97	84000	1.4090	1.0104
1.3362	2.02	86000	1.3790	1.0065
1.3586	2.07	88000	1.3730	1.0061
1.29	2.11	90000	1.3825	1.0099
1.387	2.16	92000	1.3499	1.0087
1.356	2.21	94000	1.3591	1.0102
1.3145	2.25	96000	1.3154	1.0084
1.2747	2.3	98000	1.3095	1.0080
1.3444	2.35	100000	1.2910	1.0087
1.2304	2.39	102000	1.2850	1.0074
1.3306	2.44	104000	1.2698	1.0085
1.2431	2.49	106000	1.2645	1.0087
1.1938	2.54	108000	1.2497	1.0061
1.2476	2.58	110000	1.2416	1.0077
1.1743	2.63	112000	1.2460	1.0064
1.2554	2.68	114000	1.2369	1.0076
1.1518	2.72	116000	1.2212	1.0076
1.1812	2.77	118000	1.1954	1.0054
1.2447	2.82	120000	1.2144	1.0082
1.2839	2.86	122000	1.1976	1.0076
1.117	2.91	124000	1.1809	1.0060
1.1151	2.96	126000	1.1930	1.0087
1.182	3.0	128000	1.1750	1.0079
1.214	3.05	130000	1.1687	1.0047
1.1701	3.1	132000	1.1651	1.0071
1.1268	3.15	134000	1.1635	1.0068
1.193	3.19	136000	1.1550	1.0075
1.1245	3.24	138000	1.1436	1.0068
1.136	3.29	140000	1.1378	1.0059
1.1318	3.33	142000	1.1526	1.0074
1.1129	3.38	144000	1.1452	1.0074
1.1175	3.43	146000	1.1491	1.0092
1.0812	3.47	148000	1.1300	1.0075
1.0735	3.52	150000	1.1442	1.0083
1.0626	3.57	152000	1.1298	1.0083
1.0952	3.61	154000	1.1326	1.0087
1.0756	3.66	156000	1.1341	1.0089
1.0695	3.71	158000	1.1287	1.0092
1.0672	3.76	160000	1.1379	1.0085
1.1802	3.8	162000	1.1235	1.0074
1.094	3.85	164000	1.1155	1.0068
1.0811	3.9	166000	1.1172	1.0078
0.9901	3.94	168000	1.0964	1.0064
1.0907	3.99	170000	1.1031	1.0084
1.0799	4.04	172000	1.0924	1.0071
1.0271	4.08	174000	1.1020	1.0086
1.0482	4.13	176000	1.1067	1.0084
1.1078	4.18	178000	1.1083	1.0088
1.1798	4.23	180000	1.0964	1.0064
1.0933	4.27	182000	1.1034	1.0080
1.0272	4.32	184000	1.1036	1.0075
1.1125	4.37	186000	1.1022	1.0084
1.033	4.41	188000	1.0906	1.0070
1.1048	4.46	190000	1.0923	1.0079
1.1565	4.51	192000	1.0976	1.0078
1.0698	4.55	194000	1.0950	1.0073
1.0735	4.6	196000	1.0920	1.0074
1.0137	4.65	198000	1.0899	1.0073
1.0669	4.69	200000	1.0929	1.0076

Framework versions

• Transformers 4.35.0
• Pytorch 2.1.0+cu118
• Datasets 2.14.6
• Tokenizers 0.14.1