torgo_xlsr_finetune_M01

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 1.3034
• Wer: 0.2292

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: 4
• eval_batch_size: 4
• 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: 1000
• num_epochs: 20

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.4693	0.85	1000	3.2808	1.0
1.4742	1.7	2000	1.3840	0.7581
0.7802	2.55	3000	1.2332	0.5535
0.5771	3.4	4000	1.3305	0.4423
0.4685	4.25	5000	1.2289	0.4032
0.4235	5.1	6000	1.3615	0.3540
0.3593	5.95	7000	1.1796	0.3311
0.3319	6.8	8000	1.2863	0.3336
0.298	7.65	9000	1.2067	0.3022
0.2729	8.5	10000	1.5681	0.3090
0.24	9.35	11000	1.3628	0.3022
0.2104	10.2	12000	1.6944	0.3022
0.2285	11.05	13000	1.6160	0.2997
0.2027	11.89	14000	1.6614	0.3081
0.2013	12.74	15000	1.3976	0.2683
0.1945	13.59	16000	1.0957	0.2317
0.1644	14.44	17000	1.4140	0.2699
0.163	15.29	18000	1.2615	0.2436
0.1414	16.14	19000	1.4278	0.2640
0.1476	16.99	20000	1.3421	0.2360
0.1415	17.84	21000	1.3527	0.2402
0.1217	18.69	22000	1.3593	0.2377
0.1353	19.54	23000	1.3034	0.2292

Framework versions

• Transformers 4.26.1
• Pytorch 2.2.1
• Datasets 2.18.0
• Tokenizers 0.13.3