ft_0123_korean_1

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.7160
• Cer: 0.1832

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: 8
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
60.2779	0.09	100	120.0675	0.9556
45.2315	0.18	200	66.7612	1.0
30.7852	0.27	300	50.1041	1.0
22.2754	0.35	400	31.5537	1.0
13.4785	0.44	500	13.3387	1.0
6.4901	0.53	600	5.4424	1.0
4.8734	0.62	700	5.0449	1.0
4.7501	0.71	800	4.9923	1.0
4.7477	0.8	900	4.9037	1.0
4.7511	0.88	1000	4.9181	1.0
4.7076	0.97	1100	4.8488	1.0
4.6829	1.06	1200	4.7397	1.0
4.6768	1.15	1300	4.7180	1.0
4.6089	1.24	1400	4.6611	1.0
4.56	1.33	1500	4.6175	1.0
4.499	1.41	1600	4.5056	0.9999
4.38	1.5	1700	4.2237	0.9857
4.1179	1.59	1800	3.8646	0.9816
3.7534	1.68	1900	3.2313	0.6453
3.3149	1.77	2000	2.8237	0.5349
3.0137	1.86	2100	2.5717	0.5091
2.8094	1.94	2200	2.3614	0.4624
2.5519	2.03	2300	2.2160	0.4501
2.4768	2.12	2400	2.0897	0.4274
2.3625	2.21	2500	1.9856	0.4168
2.2876	2.3	2600	1.9065	0.4016
2.2289	2.39	2700	1.8137	0.3930
2.0996	2.47	2800	1.7548	0.3755
2.0683	2.56	2900	1.6713	0.3590
2.0371	2.65	3000	1.6082	0.3528
1.8858	2.74	3100	1.5596	0.3453
1.8692	2.83	3200	1.5088	0.3346
1.8373	2.92	3300	1.4700	0.3299
1.8318	3.0	3400	1.4286	0.3208
1.7482	3.09	3500	1.4024	0.3151
1.6887	3.18	3600	1.3581	0.3126
1.6574	3.27	3700	1.3161	0.2998
1.6119	3.36	3800	1.2720	0.2924
1.6103	3.45	3900	1.2531	0.2921
1.5362	3.53	4000	1.2326	0.2917
1.4972	3.62	4100	1.1904	0.2816
1.5005	3.71	4200	1.1757	0.2784
1.4586	3.8	4300	1.1463	0.2737
1.4483	3.89	4400	1.1246	0.2694
1.4354	3.98	4500	1.0976	0.2641
1.3648	4.06	4600	1.0730	0.2606
1.3194	4.15	4700	1.0460	0.2579
1.3316	4.24	4800	1.0362	0.2516
1.3138	4.33	4900	1.0166	0.2475
1.3217	4.42	5000	0.9917	0.2456
1.2914	4.51	5100	0.9835	0.2411
1.2364	4.59	5200	0.9647	0.2409
1.2034	4.68	5300	0.9621	0.2368
1.2028	4.77	5400	0.9255	0.2311
1.2354	4.86	5500	0.9119	0.2280
1.2295	4.95	5600	0.9113	0.2287
1.2007	5.04	5700	0.8934	0.2229
1.1637	5.12	5800	0.8867	0.2256
1.1221	5.21	5900	0.8787	0.2213
1.171	5.3	6000	0.8607	0.2176
1.1042	5.39	6100	0.8514	0.2171
1.063	5.48	6200	0.8510	0.2175
1.0965	5.57	6300	0.8355	0.2107
1.0611	5.65	6400	0.8298	0.2096
1.0697	5.74	6500	0.8149	0.2074
1.0342	5.83	6600	0.8043	0.2037
1.0586	5.92	6700	0.8060	0.2028
1.0553	6.01	6800	0.8017	0.2029
1.0369	6.1	6900	0.7906	0.2015
0.9646	6.18	7000	0.7870	0.1987
0.9747	6.27	7100	0.7836	0.1970
0.9933	6.36	7200	0.7708	0.1963
0.9793	6.45	7300	0.7740	0.1957
0.9642	6.54	7400	0.7618	0.1934
0.9936	6.63	7500	0.7554	0.1919
0.9466	6.71	7600	0.7438	0.1891
0.9597	6.8	7700	0.7437	0.1900
0.9374	6.89	7800	0.7415	0.1909
0.9719	6.98	7900	0.7352	0.1908
0.9067	7.07	8000	0.7358	0.1880
0.8998	7.16	8100	0.7329	0.1879
0.9271	7.24	8200	0.7262	0.1864
0.8951	7.33	8300	0.7217	0.1860
0.9136	7.42	8400	0.7239	0.1854
0.9446	7.51	8500	0.7214	0.1844
0.8978	7.6	8600	0.7220	0.1837
0.8923	7.69	8700	0.7174	0.1838
0.9406	7.77	8800	0.7187	0.1836
0.9242	7.86	8900	0.7159	0.1836
0.8994	7.95	9000	0.7160	0.1832

Framework versions

• Transformers 4.36.2
• Pytorch 2.1.2+cu118
• Datasets 2.16.1
• Tokenizers 0.15.0