sit-justin/whisper-medium-seq-p2-moe-sox

Whisper Small Chinese MOE Response

This model is a fine-tuned version of sit-justin/whisper-medium-p2-moe on the MOE Response Chinese dataset. It achieves the following results on the evaluation set:

• Loss: 0.0360
• Cer: 2.4470

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: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 250
• training_steps: 2000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
0.1079	0.0852	200	0.1211	10.1923
0.1371	0.1704	400	0.1223	8.1479
0.1069	0.2556	600	0.1130	9.1791
0.1131	0.3409	800	0.1012	8.6100
0.0918	0.4261	1000	0.0844	6.0797
0.0846	0.5113	1200	0.0708	4.6004
0.0606	0.5965	1400	0.0594	4.1761
0.0545	0.6817	1600	0.0488	5.9528
0.044	0.7669	1800	0.0398	2.6056
0.0376	0.8522	2000	0.0360	2.4470

Framework versions

• Transformers 4.42.3
• Pytorch 2.3.1+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1