whisper-multilang-asr-20260308

This model is a fine-tuned version of vinai/PhoWhisper-medium on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.2116
• Wer: 0.2695
• Bleu: 0.6869
• Precisions: [0.7840584675117049, 0.7116923444404634, 0.6562773813994242, 0.6078895463510848]
• Brevity Penalty: 1.0
• Length Ratio: 1.0724
• Translation Length: 8757
• Reference Length: 8166

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
• num_devices: 3
• gradient_accumulation_steps: 2
• total_train_batch_size: 96
• total_eval_batch_size: 24
• 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_ratio: 0.1
• num_epochs: 5
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Bleu	Precisions	Brevity Penalty	Length Ratio	Translation Length	Reference Length
0.3361	1.3200	5000	0.3421	0.4026	0.5694	[0.7045060658578857, 0.6052215189873418, 0.5293005671077504, 0.46584158415841587]	1.0	1.1316	9232	8158
0.2648	2.6399	10000	0.2565	0.3913	0.5899	[0.7018725015779508, 0.6193817145362859, 0.5558480201419089, 0.5010773282259995]	1.0	1.1695	9506	8128
0.2291	3.9599	15000	0.2116	0.2695	0.6869	[0.7840584675117049, 0.7116923444404634, 0.6562773813994242, 0.6078895463510848]	1.0	1.0724	8757	8166

Framework versions

• Transformers 4.57.6
• Pytorch 2.7.1+cu118
• Datasets 4.0.0
• Tokenizers 0.22.2