korean-aihub-learning-3

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: 2.2854
• Wer: 0.7921

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: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
No log	0.99	35	45.5713	1.0
No log	1.99	70	24.4376	1.0
35.4145	2.99	105	18.3030	1.0
35.4145	3.99	140	12.6702	1.0
35.4145	4.99	175	7.4939	1.0
11.687	5.99	210	4.9592	1.0
11.687	6.99	245	4.6777	1.0
11.687	7.99	280	4.6597	1.0
4.8003	8.99	315	4.6777	1.0
4.8003	9.99	350	4.7003	1.0
4.8003	10.99	385	4.6129	1.0
4.6383	11.99	420	4.6209	1.0
4.6383	12.99	455	4.6035	1.0
4.6383	13.99	490	4.6166	1.0
4.577	14.99	525	4.6026	1.0
4.577	15.99	560	4.5337	1.0
4.577	16.99	595	4.5284	1.0
4.5124	17.99	630	4.5710	1.0
4.5124	18.99	665	4.5223	1.0
4.3818	19.99	700	4.4472	1.0
4.3818	20.99	735	4.4272	0.9977
4.3818	21.99	770	4.4160	0.9977
4.2796	22.99	805	4.3741	0.9988
4.2796	23.99	840	4.3087	1.0
4.2796	24.99	875	4.2336	1.0
4.0489	25.99	910	4.1352	0.9988
4.0489	26.99	945	4.0669	1.0
4.0489	27.99	980	3.8551	0.9988
3.6122	28.99	1015	3.6699	0.9919
3.6122	29.99	1050	3.4580	0.9781
3.6122	30.99	1085	3.1899	0.9434
2.8886	31.99	1120	3.0746	0.9550
2.8886	32.99	1155	2.8143	0.9353
2.8886	33.99	1190	2.7004	0.9122
2.0277	34.99	1225	2.5284	0.9076
2.0277	35.99	1260	2.4677	0.8972
2.0277	36.99	1295	2.3426	0.8568
1.2486	37.99	1330	2.2456	0.8822
1.2486	38.99	1365	2.3250	0.9238
0.7572	39.99	1400	2.2832	0.8557
0.7572	40.99	1435	2.2671	0.8406
0.7572	41.99	1470	2.3070	0.8857
0.4768	42.99	1505	2.2138	0.8476
0.4768	43.99	1540	2.2034	0.8799
0.4768	44.99	1575	2.2215	0.8487
0.3362	45.99	1610	2.3416	0.8834
0.3362	46.99	1645	2.3452	0.8383
0.3362	47.99	1680	2.2449	0.8360
0.257	48.99	1715	2.2249	0.8199
0.257	49.99	1750	2.3455	0.8106
0.257	50.99	1785	2.2537	0.8233
0.2116	51.99	1820	2.2501	0.8025
0.2116	52.99	1855	2.3180	0.8649
0.2116	53.99	1890	2.1855	0.8106
0.1787	54.99	1925	2.2140	0.8014
0.1787	55.99	1960	2.3140	0.8453
0.1787	56.99	1995	2.2140	0.8025
0.1498	57.99	2030	2.3381	0.8314
0.1498	58.99	2065	2.2591	0.8256
0.1372	59.99	2100	2.2538	0.7979
0.1372	60.99	2135	2.2052	0.7933
0.1372	61.99	2170	2.2370	0.8233
0.129	62.99	2205	2.2331	0.7898
0.129	63.99	2240	2.3022	0.8002
0.129	64.99	2275	2.3514	0.7956
0.1075	65.99	2310	2.3303	0.8279
0.1075	66.99	2345	2.2747	0.8025
0.1075	67.99	2380	2.2899	0.8152
0.0979	68.99	2415	2.3299	0.8164
0.0979	69.99	2450	2.1819	0.7945
0.0979	70.99	2485	2.2141	0.8222
0.0973	71.99	2520	2.3683	0.8395
0.0973	72.99	2555	2.2235	0.8199
0.0973	73.99	2590	2.2474	0.8048
0.0814	74.99	2625	2.3116	0.7968
0.0814	75.99	2660	2.2494	0.7945
0.0814	76.99	2695	2.2441	0.7968
0.0745	77.99	2730	2.2489	0.7864
0.0745	78.99	2765	2.2568	0.7921
0.0741	79.99	2800	2.2598	0.7875
0.0741	80.99	2835	2.3131	0.8002
0.0741	81.99	2870	2.2719	0.7898
0.0662	82.99	2905	2.2901	0.7875
0.0662	83.99	2940	2.3092	0.7979
0.0662	84.99	2975	2.3361	0.8048
0.0556	85.99	3010	2.3308	0.8152
0.0556	86.99	3045	2.3106	0.8164
0.0556	87.99	3080	2.3363	0.8002
0.0504	88.99	3115	2.3588	0.7910
0.0504	89.99	3150	2.3528	0.7956
0.0504	90.99	3185	2.3201	0.7794
0.0496	91.99	3220	2.3386	0.7991
0.0496	92.99	3255	2.3423	0.7956
0.0496	93.99	3290	2.3312	0.7956
0.0468	94.99	3325	2.3362	0.7968
0.0468	95.99	3360	2.2962	0.7887
0.0468	96.99	3395	2.2864	0.7841
0.0475	97.99	3430	2.2870	0.7898
0.0475	98.99	3465	2.2866	0.7898
0.0411	99.99	3500	2.2854	0.7921

Framework versions

• Transformers 4.20.1
• Pytorch 1.12.0+cu113
• Datasets 2.3.2
• Tokenizers 0.12.1