base-model-with-warmup-fulldata-fairseq-V1

This model is a fine-tuned version of facebook/wav2vec2-base-960h on the None dataset. It achieves the following results on the evaluation set:

• Loss: 4.6583

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: 16
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.98) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 200
• num_epochs: 50

Training results

Training Loss	Epoch	Step	Validation Loss
15.6587	2.05	200	4.4816
3.8464	4.1	400	4.7516
3.8377	6.15	600	4.4104
3.915	8.21	800	4.6832
3.8094	10.26	1000	5.1651
3.8095	12.31	1200	4.7847
3.8427	14.36	1400	4.6613
3.7917	16.41	1600	4.7898
3.804	18.46	1800	4.7368
3.8492	20.51	2000	4.7656
3.7943	22.56	2200	4.8043
3.7982	24.62	2400	4.5903
3.7979	26.67	2600	4.5546
3.7938	28.72	2800	4.6392
3.7891	30.77	3000	4.8080
3.7936	32.82	3200	4.5811
3.7926	34.87	3400	4.5891
3.7922	36.92	3600	4.6251
3.79	38.97	3800	4.7409
3.791	41.03	4000	4.6539
3.7893	43.08	4200	4.6752
3.7882	45.13	4400	4.6751
3.7924	47.18	4600	4.6222
3.7991	49.23	4800	4.6583

Framework versions

• Transformers 4.28.0
• Pytorch 2.0.1
• Datasets 2.12.0
• Tokenizers 0.13.2