xbilek25/whisper-small-train-v2.2

basic_train_basic_test 1000 similar params: per_device_train_batch_size=32, # bylo 16 a pod tim 1 gradient_accumulation_steps=2, warmup_steps=300, max_steps=3000

This model is a fine-tuned version of openai/whisper-small on the xbilek25/xbilek25/train_set_1000_en_de_en dataset. It achieves the following results on the evaluation set:

• Loss: 0.2957
• Wer: 10.8108

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
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 3000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.0087	7.02	500	0.2377	11.3924
0.0024	15.02	1000	0.2643	11.8029
0.0006	23.02	1500	0.2832	10.8792
0.0004	31.02	2000	0.2901	10.6055
0.0003	39.01	2500	0.2941	10.7766
0.0003	47.01	3000	0.2957	10.8108

Framework versions

• Transformers 4.37.2
• Pytorch 2.2.1+cu121
• Datasets 2.19.0
• Tokenizers 0.15.2