kpriyanshu256/whisper-large-v2-as-1000-32-1e-05-bn-multi

kpriyanshu256/whisper-large-v2-as-600-32-1e-05-bn-Assamese

This model is a fine-tuned version of kpriyanshu256/whisper-large-v2-as-600-32-1e-05-bn on the Common Voice 11.0 and the FLEURS datasets. It achieves the following results on the evaluation set:

• Loss: 0.2486
• Wer: 17.5600

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: 1e-05
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• 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: 50
• training_steps: 1000

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.1273	0.1	100	0.1737	20.8988
0.0811	0.2	200	0.1739	19.0038
0.0638	0.3	300	0.1823	18.4804
0.0404	1.05	400	0.1893	17.1810
0.0316	1.15	500	0.2067	17.0186
0.027	1.25	600	0.2081	17.7405
0.025	2.01	700	0.2213	17.7585
0.0213	2.11	800	0.2237	17.8488
0.0176	2.21	900	0.2390	16.7479
0.0184	2.31	1000	0.2486	17.5600

Framework versions

• Transformers 4.26.0.dev0
• Pytorch 1.13.0+cu117
• Datasets 2.7.1.dev0
• Tokenizers 0.13.2