hindi_wav2vec2_optimized_2

This model is a fine-tuned version of Harveenchadha/hindi_base_wav2vec2 on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.6089
• Wer: 0.1899

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: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 10
• num_epochs: 100

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.2406	5.56	25	0.5544	0.3418
0.0725	11.11	50	0.4668	0.2426
0.0437	16.67	75	0.6185	0.2405
0.0344	22.22	100	0.5553	0.2257
0.0311	27.78	125	0.5775	0.2300
0.0287	33.33	150	0.7120	0.2110
0.0242	38.89	175	0.7068	0.2152
0.0175	44.44	200	0.5265	0.2384
0.0128	50.0	225	0.5554	0.2300
0.0114	55.56	250	0.6330	0.2300
0.0105	61.11	275	0.6020	0.2194
0.0052	66.67	300	0.6073	0.2152
0.0041	72.22	325	0.5865	0.1941
0.0074	77.78	350	0.6138	0.1983
0.0033	83.33	375	0.6106	0.1920
0.0074	88.89	400	0.6089	0.1899

Framework versions

• Transformers 4.34.0
• Pytorch 2.0.1+cu118
• Datasets 2.14.5
• Tokenizers 0.14.1