hubert-large-ls960-ft-V2

This model is a fine-tuned version of facebook/hubert-large-ls960-ft on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.3760
• Wer: 0.0445
• Per: 0.0354

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: 2
• eval_batch_size: 2
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 20

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Per
5.7632	1.0	1637	0.8273	0.3876	0.3794
0.374	2.0	3274	0.4468	0.0635	0.0493
0.1657	3.0	4911	0.3662	0.0596	0.0460
0.1179	4.0	6548	0.3809	0.0518	0.0401
0.0922	5.0	8185	0.4233	0.0519	0.0386
0.0763	6.0	9822	0.3852	0.0517	0.0390
0.0649	7.0	11459	0.3056	0.0487	0.0377
0.056	8.0	13096	0.3266	0.0460	0.0353
0.0467	9.0	14733	0.3608	0.0458	0.0346
0.04	10.0	16370	0.3399	0.0461	0.0355
0.0356	11.0	18007	0.3549	0.0441	0.0341
0.0336	12.0	19644	0.3391	0.0430	0.0332
0.0338	13.0	21281	0.3968	0.0482	0.0383
0.0278	14.0	22918	0.3699	0.0468	0.0376
0.0224	15.0	24555	0.3680	0.0458	0.0365
0.0225	16.0	26192	0.3713	0.0477	0.0392
0.0189	17.0	27829	0.3681	0.0442	0.0350
0.0185	18.0	29466	0.3829	0.0444	0.0352
0.018	19.0	31103	0.3866	0.0450	0.0354
0.0169	20.0	32740	0.3760	0.0445	0.0354

Framework versions

• Transformers 4.35.2
• Pytorch 2.1.0+cu121
• Datasets 2.16.1
• Tokenizers 0.15.0