torgo_xlsr_finetune_F04_keep_all

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.6928
• Wer: 0.2375

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
3.6072	0.56	1000	3.3393	1.0
2.6659	1.11	2000	2.2682	0.9615
1.1671	1.67	3000	1.5754	0.7076
0.8034	2.22	4000	1.6181	0.5738
0.681	2.78	5000	1.4303	0.4888
0.5537	3.33	6000	1.6473	0.4605
0.5213	3.89	7000	1.5112	0.4377
0.4445	4.44	8000	1.3818	0.4182
0.4396	5.0	9000	1.5070	0.4274
0.4179	5.55	10000	1.4717	0.3995
0.3641	6.11	11000	1.3974	0.3359
0.3264	6.66	12000	1.6107	0.3607
0.3252	7.22	13000	1.2008	0.3023
0.2894	7.77	14000	1.4290	0.3039
0.2959	8.33	15000	1.3412	0.3126
0.2778	8.88	16000	1.4307	0.3035
0.2495	9.44	17000	1.3922	0.3092
0.2704	9.99	18000	1.3564	0.2627
0.2307	10.55	19000	1.4333	0.2612
0.2211	11.1	20000	1.6846	0.2775
0.1995	11.66	21000	1.4738	0.2856
0.2208	12.22	22000	1.5382	0.2695
0.2087	12.77	23000	1.3165	0.2722
0.1769	13.33	24000	1.9005	0.2791
0.1883	13.88	25000	1.7298	0.2768
0.1835	14.44	26000	1.6170	0.2608
0.1829	14.99	27000	1.8436	0.2711
0.1563	15.55	28000	1.7982	0.2627
0.1474	16.1	29000	1.6996	0.2398
0.155	16.66	30000	1.6696	0.2482
0.1295	17.21	31000	1.8057	0.2429
0.1345	17.77	32000	1.8119	0.2474
0.1475	18.32	33000	1.8016	0.2505
0.1246	18.88	34000	1.7389	0.2425
0.1395	19.43	35000	1.7249	0.2421
0.1223	19.99	36000	1.6928	0.2375

Framework versions

• Transformers 4.26.1
• Pytorch 2.2.0
• Datasets 2.16.1
• Tokenizers 0.13.3