ai-light-dance_drums_ft_pretrain_wav2vec2-base-new-v6-2

This model is a fine-tuned version of gary109/ai-light-dance_drums_pretrain_wav2vec2-base-new on the GARY109/AI_LIGHT_DANCE - ONSET-IDMT-MDB-ENST3 dataset. It achieves the following results on the evaluation set:

• Loss: 0.7363
• Wer: 0.4070

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: 0.0004
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 30
• num_epochs: 100.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
17.273	0.99	35	2.9044	1.0
1.845	1.99	70	3.5094	1.0
1.7091	2.99	105	2.0063	1.0
1.5219	3.99	140	1.9601	0.9519
1.3926	4.99	175	2.0205	0.6782
1.2051	5.99	210	1.8955	0.5909
1.1599	6.99	245	1.6756	0.6057
1.0186	7.99	280	1.2840	0.5660
0.9545	8.99	315	1.6695	0.5407
0.987	9.99	350	1.2654	0.5449
0.9851	10.99	385	1.6630	0.5192
1.0328	11.99	420	1.6264	0.5243
0.8752	12.99	455	1.3431	0.5247
0.8746	13.99	490	1.2251	0.4951
0.8503	14.99	525	1.1992	0.4943
0.8431	15.99	560	1.1651	0.5247
0.7538	16.99	595	1.2192	0.5027
0.84	17.99	630	1.0983	0.4947
0.7836	18.99	665	1.1258	0.4762
0.7024	19.99	700	1.0138	0.4736
0.6788	20.99	735	1.0691	0.4977
0.757	21.99	770	1.1725	0.4720
0.7369	22.99	805	1.0670	0.4985
0.7461	23.99	840	1.0139	0.4854
0.706	24.99	875	1.3573	0.4623
0.6968	25.99	910	1.1377	0.4829
0.6703	26.99	945	1.0983	0.4829
0.7503	27.99	980	1.0445	0.4973
0.6806	28.99	1015	1.0953	0.4644
0.6947	29.99	1050	1.0280	0.4703
0.6513	30.99	1085	0.9437	0.4745
0.651	31.99	1120	1.3161	0.4656
0.7611	32.99	1155	1.1245	0.4770
0.7898	33.99	1190	1.2651	0.4648
0.6446	34.99	1225	1.0163	0.4736
0.6135	35.99	1260	0.9704	0.4551
0.7348	36.99	1295	1.0713	0.4610
0.6297	37.99	1330	1.0118	0.4547
0.5841	38.99	1365	1.0330	0.4585
0.5828	39.99	1400	0.9250	0.4589
0.5559	40.99	1435	0.8737	0.4429
0.6374	41.99	1470	0.8740	0.4471
0.7267	42.99	1505	0.8053	0.4441
0.6019	43.99	1540	0.9041	0.4395
0.5968	44.99	1575	0.8793	0.4323
0.5474	45.99	1610	0.9814	0.4407
0.5276	46.99	1645	0.8298	0.4344
0.8823	47.99	1680	0.8099	0.4175
0.5711	48.99	1715	0.8020	0.4159
0.6347	49.99	1750	0.9176	0.4353
0.5777	50.99	1785	0.9645	0.4171
0.5422	51.99	1820	0.8586	0.4209
0.5576	52.99	1855	0.8729	0.4344
0.5031	53.99	1890	0.8961	0.4420
0.5703	54.99	1925	0.9182	0.4264
0.4779	55.99	1960	0.8736	0.4399
0.5558	56.99	1995	0.8764	0.4235
0.5102	57.99	2030	0.8493	0.4235
0.6319	58.99	2065	0.8209	0.4268
0.6132	59.99	2100	0.8049	0.4251
0.5159	60.99	2135	0.8424	0.4171
0.5212	61.99	2170	0.9220	0.4188
0.4947	62.99	2205	0.8612	0.4167
0.5086	63.99	2240	0.8042	0.4121
0.5259	64.99	2275	0.8656	0.4036
0.4974	65.99	2310	0.7363	0.4070
0.459	66.99	2345	0.7454	0.4007
0.5064	67.99	2380	0.8203	0.4095
0.5419	68.99	2415	0.7699	0.4011
0.4637	69.99	2450	0.7702	0.4112
0.5033	70.99	2485	0.7649	0.4083
0.4585	71.99	2520	0.8038	0.4184
0.4804	72.99	2555	0.8288	0.4036
0.4907	73.99	2590	0.7794	0.4062
0.4846	74.99	2625	0.7792	0.4078
0.4733	75.99	2660	0.7900	0.4078
0.4772	76.99	2695	0.8084	0.4057
0.4785	77.99	2730	0.7884	0.4074
0.4888	78.99	2765	0.7988	0.4074
0.4506	79.99	2800	0.8100	0.4074
0.5364	80.99	2835	0.8187	0.4036
0.4787	81.99	2870	0.8040	0.4019
0.5052	82.99	2905	0.8412	0.4100
0.4705	83.99	2940	0.8162	0.4066
0.5137	84.99	2975	0.8558	0.4070
0.5117	85.99	3010	0.8376	0.4078
0.5671	86.99	3045	0.8111	0.4053
0.5333	87.99	3080	0.8297	0.4121
0.4534	88.99	3115	0.7829	0.4040
0.4703	89.99	3150	0.7961	0.4045
0.5478	90.99	3185	0.7836	0.4032
0.453	91.99	3220	0.8295	0.4078
0.4513	92.99	3255	0.8083	0.3998
0.4221	93.99	3290	0.7993	0.4032
0.4725	94.99	3325	0.7943	0.4019
0.4381	95.99	3360	0.8145	0.4057
0.6443	96.99	3395	0.8085	0.4011
0.4089	97.99	3430	0.7982	0.4028
0.4779	98.99	3465	0.7993	0.4003
0.4731	99.99	3500	0.8004	0.4028

Framework versions

• Transformers 4.25.0.dev0
• Pytorch 1.8.1+cu111
• Datasets 2.7.1.dev0
• Tokenizers 0.13.2