StefanJevtic63/whisper-large-v3-sr-lora

Whisper - Serbian Model

This model is a fine-tuned version of openai/whisper-large-v3 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.1417

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: 0.0009
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• 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: 400
• training_steps: 16000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss
0.1801	0.0705	500	0.1853
0.1588	0.1409	1000	0.1679
0.1448	0.2114	1500	0.1582
0.1746	0.2819	2000	0.2240
0.1407	0.3524	2500	0.1523
0.1411	0.4228	3000	0.1491
0.1327	0.4933	3500	0.1472
0.1389	0.5638	4000	0.1509
0.1331	0.6342	4500	0.1364
0.1273	0.7047	5000	0.1396
0.1209	0.7752	5500	0.1417

Framework versions

• PEFT 0.13.2
• Transformers 4.46.3
• Pytorch 2.5.1
• Datasets 3.0.0
• Tokenizers 0.20.3