hubert-large-korean-finetuned-korspeech-ser

This model is a fine-tuned version of team-lucid/hubert-large-korean on the None dataset. It achieves the following results on the evaluation set:

Loss: 1.0010
Macro F1: 0.6297
Accuracy: 0.6321
Weighted f1: 0.6297
Micro f1: 0.6321
Weighted recall: 0.6321
Micro recall: 0.6321
Macro recall: 0.6321
Weighted precision: 0.6354
Micro precision: 0.6321
Macro precision: 0.6354

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
gradient_accumulation_steps: 4
total_train_batch_size: 32
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
lr_scheduler_warmup_ratio: 0.1
lr_scheduler_warmup_steps: 100
num_epochs: 6

Training results

Training Loss	Epoch	Step	Validation Loss	Macro F1	Accuracy	Weighted f1	Micro f1	Weighted recall	Micro recall	Macro recall	Weighted precision	Micro precision	Macro precision
1.3419	0.28	100	1.2665	0.4034	0.4276	0.4034	0.4276	0.4276	0.4276	0.4276	0.4520	0.4276	0.4520
1.2356	0.57	200	1.2013	0.4506	0.4702	0.4506	0.4702	0.4702	0.4702	0.4702	0.4645	0.4702	0.4645
1.1905	0.85	300	1.2031	0.4480	0.4957	0.4480	0.4957	0.4957	0.4957	0.4957	0.5256	0.4957	0.5256
1.1335	1.13	400	1.1278	0.5083	0.5178	0.5083	0.5178	0.5178	0.5178	0.5178	0.5178	0.5178	0.5178
1.0794	1.42	500	1.0613	0.5507	0.5518	0.5507	0.5518	0.5518	0.5518	0.5518	0.5511	0.5518	0.5511
1.0535	1.7	600	1.0259	0.5517	0.5618	0.5517	0.5618	0.5618	0.5618	0.5618	0.5630	0.5618	0.5630
1.0396	1.99	700	1.0276	0.5514	0.5554	0.5514	0.5554	0.5554	0.5554	0.5554	0.5557	0.5554	0.5557
0.9841	2.27	800	1.0614	0.5440	0.5611	0.5440	0.5611	0.5611	0.5611	0.5611	0.5698	0.5611	0.5698
0.9527	2.55	900	1.0643	0.5405	0.5547	0.5405	0.5547	0.5547	0.5547	0.5547	0.5646	0.5547	0.5646
0.9616	2.84	1000	0.9971	0.5850	0.5895	0.5850	0.5895	0.5895	0.5895	0.5895	0.5924	0.5895	0.5924
0.9284	3.12	1100	1.0220	0.5626	0.5717	0.5626	0.5717	0.5717	0.5717	0.5717	0.5808	0.5717	0.5808
0.8744	3.4	1200	1.0368	0.5778	0.5831	0.5778	0.5831	0.5831	0.5831	0.5831	0.5884	0.5831	0.5884
0.8654	3.69	1300	1.0320	0.5925	0.5980	0.5925	0.5980	0.5980	0.5980	0.5980	0.6057	0.5980	0.6057
0.8716	3.97	1400	0.9965	0.5875	0.5930	0.5875	0.5930	0.5930	0.5930	0.5930	0.6001	0.5930	0.6001
0.7969	4.26	1500	0.9810	0.6102	0.6122	0.6102	0.6122	0.6122	0.6122	0.6122	0.6176	0.6122	0.6176
0.799	4.54	1600	0.9982	0.6075	0.6072	0.6075	0.6072	0.6072	0.6072	0.6072	0.6094	0.6072	0.6094
0.7785	4.82	1700	0.9812	0.6092	0.6136	0.6092	0.6136	0.6136	0.6136	0.6136	0.6136	0.6136	0.6136
0.7653	5.11	1800	0.9913	0.6124	0.6165	0.6124	0.6165	0.6165	0.6165	0.6165	0.6208	0.6165	0.6208
0.7192	5.39	1900	1.0057	0.6201	0.6207	0.6201	0.6207	0.6207	0.6207	0.6207	0.6244	0.6207	0.6244
0.7201	5.67	2000	1.0022	0.6207	0.6229	0.6207	0.6229	0.6229	0.6229	0.6229	0.6249	0.6229	0.6249
0.7351	5.96	2100	1.0010	0.6297	0.6321	0.6297	0.6321	0.6321	0.6321	0.6321	0.6354	0.6321	0.6354

Framework versions

Transformers 4.30.2
Pytorch 2.0.0
Datasets 2.1.0
Tokenizers 0.13.3

byc3230/hubert-large-korean-finetuned-korspeech-ser

hubert-large-korean-finetuned-korspeech-ser

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

Seed this model