erjoy/whisper-tiny-gu-5k-steps

Whisper Tiny Gu - V1

This model is a fine-tuned version of openai/whisper-tiny on the ai4bharat/IndicVoices dataset. It achieves the following results on the evaluation set:

• Loss: 0.2995
• Wer: 58.8371

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: 5e-05
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• optimizer: Use OptimizerNames.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: 500
• training_steps: 5000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.2832	0.9042	500	0.3072	75.1796
0.1942	1.8083	1000	0.2364	67.4873
0.1301	2.7125	1500	0.2086	62.7551
0.0974	3.6166	2000	0.2098	60.8008
0.0688	4.5208	2500	0.2155	60.7146
0.0412	5.4250	3000	0.2278	59.9387
0.0254	6.3291	3500	0.2493	59.2107
0.0138	7.2333	4000	0.2765	58.8658
0.0061	8.1374	4500	0.2924	59.2394
0.0034	9.0416	5000	0.2995	58.8371

Framework versions

• Transformers 4.51.3
• Pytorch 2.6.0+cu124
• Datasets 3.5.1
• Tokenizers 0.21.0