Skip to content

使用paraformer_streaming下finetune脚本微调online模型后,直接使用微调后权重文件可以输出中文,但是导出为onnx文件后输出为乱码 #1590

@yepengxj

Description

@yepengxj

Notice: In order to resolve issues more efficiently, please raise issue following the template.
(注意:为了更加高效率解决您遇到的问题,请按照模板提问,补充细节)

❓ Questions and Help

What is your question?

使用paraformer_streaming下finetune脚本微调online模型后,直接使用微调后权重文件可以输出中文,但是导出为onnx文件后输出为乱码

What's your environment?

  • OS (e.g., Linux):Linux

  • FunASR Version (e.g., 1.0.0):1.0.18

  • ModelScope Version (e.g., 1.11.0):1.13.2

  • PyTorch Version (e.g., 2.0.0):2.0.1

  • How you installed funasr (pip, source): source pip install -e ./

  • Python version:

  • GPU (e.g., V100M32)P40

  • CUDA/cuDNN version (e.g., cuda11.7): 11.7

  • Any other relevant information:

  • onnx 1.16.0

  • onnxruntime 1.17.0

使用微调后权重文件输出:
examples/industrial_data_pretraining/paraformer_streaming# python3 ./demo.py
Notice: ffmpeg is not installed. torchaudio is used to load audio
If you want to use ffmpeg backend to load audio, please install it by:
sudo apt install ffmpeg # ubuntu
# brew install ffmpeg # mac
Notice: If you want to use the speaker diarization, please pip install hdbscan
Notice: If you want to use whisper, please pip install -U openai-whisper
ckpt: /root/app/examples/industrial_data_pretraining/paraformer_streaming/outputs/model.pb
rtf_avg: 9.888: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1/1 [00:01<00:00, 1.78s/it]
[{'key': 'rand_key_2yW4Acq9GFz6Y', 'text': '塔克欢迎迎家来体验达摩院推出的语音识别模型'}]
rtf_avg: 0.284: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1/1 [00:00<00:00, 5.82it/s]
[{'key': 'rand_key_1t9EwL56nGisi', 'text': '阿'}]
rtf_avg: 0.296: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1/1 [00:00<00:00, 5.59it/s]
[{'key': 'rand_key_WgNZq6ITZM5jt', 'text': '克'}]
rtf_avg: 0.365: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1/1 [00:00<00:00, 4.55it/s]
[{'key': 'rand_key_gUe52RvEJgwBu', 'text': '欢迎迎'}]
rtf_avg: 0.416: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1/1 [00:00<00:00, 3.99it/s]
[{'key': 'rand_key_NO6n9JEC3HqdZ', 'text': '家来'}]
rtf_avg: 0.748: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1/1 [00:00<00:00, 2.22it/s]
[{'key': 'rand_key_6J6afU1zT0YQO', 'text': '体验达'}]
rtf_avg: 0.308: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1/1 [00:00<00:00, 5.39it/s]
[{'key': 'rand_key_aNF03vpUuT3em', 'text': '摩院推'}]
rtf_avg: 0.307: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1/1 [00:00<00:00, 5.39it/s]
[{'key': 'rand_key_6KopZ9jZICffu', 'text': '出的语'}]
rtf_avg: 0.314: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1/1 [00:00<00:00, 5.28it/s]
[{'key': 'rand_key_4G7FgtJsThJv0', 'text': '音音别'}]
rtf_avg: 0.306: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1/1 [00:00<00:00, 5.42it/s]
[{'key': 'rand_key_7In9ZMJLsCfMZ', 'text': '模型'}]
rtf_avg: 0.855: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1/1 [00:00<00:00, 6.46it/s]
[{'key': 'rand_key_yuKpslm0lcNQq', 'text': ''}]
使用onnx格式输出:
runtime/python/onnxruntime# python3 ./demo_paraformer_online.py
[{'preds': ('咽滢螵生', ['咽', '滢', '螵', '生'])}]
[{'preds': ('thanks gh窠幕糕', ['thanks', 'gh', '窠', '幕', '糕'])}]
[{'preds': ('醇疾oms硌镟', ['醇', '疾', 'oms', '硌', '镟'])}]
[{'preds': ('导澄生镟隐', ['导', '澄', '生', '镟', '隐'])}]
[{'preds': ('篝阡谤糕摞', ['篝', '阡', '谤', '糕', '摞'])}]
[{'preds': ('若说少糕军螈', ['若', '说', '少', '糕', '军', '螈'])}]
[{'preds': ('佯龋足龋橙', ['佯', '龋', '足', '龋', '橙'])}]
[{'preds': ('sts逃螈披蜱', ['sts', '逃', '螈', '披', '蜱'])}]
[{'preds': ('楷选囹还丐', ['楷', '选', '囹', '还', '丐'])}]
[{'preds': ('渤载瓿藕', ['渤', '载', '瓿', '藕'])}]
咽滢螵生thanks gh窠幕糕醇疾oms硌镟导澄生镟隐篝阡谤糕摞若说少糕军螈佯龋足龋橙sts逃螈披蜱楷选囹还丐渤载瓿藕

Metadata

Metadata

Assignees

No one assigned

    Labels

    questionFurther information is requested

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions