ymoslem/qwen-audio-en-zh-ft-decoder-32-24layers-4bs-1e5lr-cosine-3epoch-acl6060kd-new

Qwen2-Audio-7B EN-ZH Speech Translation

This model is a fine-tuned version of ymoslem/qwen-audio-en-zh-ft-decoder-32-24layers on the ymoslem/ACL-6060 and the fixie-ai/covost2 datasets. It achieves the following results on the evaluation set:

• Loss: 1.3418

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: 4
• eval_batch_size: 4
• 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: cosine
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss
1.7532	0.3690	100	1.5221
0.5721	0.7380	200	1.4187
0.492	1.1070	300	1.4419
0.3965	1.4760	400	1.3967
0.2836	1.8450	500	1.3484
0.2053	2.2140	600	1.3410
0.3168	2.5830	700	1.3532
0.1005	2.9520	800	1.3418

Framework versions

• Transformers 4.51.3
• Pytorch 2.4.1+cu124
• Datasets 3.5.0
• Tokenizers 0.21.1