wav2vec2-17

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

• Loss: 3.1355
• Wer: 1.0

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: 0.0003
• train_batch_size: 32
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 256
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 50
• num_epochs: 30

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
7.5865	1.38	25	3.4717	1.0
2.9762	2.77	50	3.1483	1.0
2.9265	4.16	75	3.1946	1.0
2.8813	5.55	100	3.0504	1.0
2.887	6.93	125	3.1358	1.0
2.9124	8.33	150	3.1653	1.0
2.8854	9.71	175	3.1243	1.0
2.91	11.11	200	3.0879	1.0
2.8868	12.49	225	3.1658	1.0
2.8827	13.88	250	3.1236	1.0
2.911	15.27	275	3.1206	1.0
2.8829	16.66	300	3.1171	1.0
2.9105	18.05	325	3.1127	1.0
2.8845	19.44	350	3.1377	1.0
2.8803	20.82	375	3.1157	1.0
2.9102	22.22	400	3.1265	1.0
2.8803	23.6	425	3.1493	1.0
2.8837	24.99	450	3.1085	1.0
2.9106	26.38	475	3.1099	1.0
2.8787	27.77	500	3.1352	1.0
2.9132	29.16	525	3.1355	1.0

Framework versions

• Transformers 4.19.2
• Pytorch 1.11.0+cu113
• Datasets 2.2.2
• Tokenizers 0.12.1