razhan/whisper-base-glk

whisper-base-hac-telegram

This model is a fine-tuned version of openai/whisper-base on the razhan/DOLMA-speech gilaki dataset. It achieves the following results on the evaluation set:

• Loss: 2.6806
• Wer: 1.0472
• Cer: 0.5468

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: 256
• eval_batch_size: 128
• seed: 42
• distributed_type: multi-GPU
• num_devices: 2
• total_train_batch_size: 512
• total_eval_batch_size: 256
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100
• num_epochs: 5.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
No log	1.0	6	3.5224	1.1311	0.5560
2.4889	2.0	12	3.4807	1.0566	0.5018
2.4889	3.0	18	3.2108	1.0561	0.4986
2.3707	4.0	24	2.9445	1.0583	0.5155
2.0528	5.0	30	2.6806	1.0472	0.5468

Framework versions

• Transformers 4.49.0.dev0
• Pytorch 2.6.0+cu124
• Datasets 3.2.0
• Tokenizers 0.21.0