Skip to content

Model "pythainlp/thaitts-onnx" in this repository requires validation #10

@MendyPanelli232

Description

@MendyPanelli232

Hi, @wannaphong, I'd like to report that a potentially risky pretrained models are being used in this project, which may pose backdoor threats. Please check the following code example:

pythaitts/pretrained/lunarlist_onnx.py

class LunarlistONNX:
    def __init__(self) -> None:
        self.encoder = ort.InferenceSession(hf_hub_download(repo_id="pythainlp/thaitts-onnx",filename="tacotron2encoder-th.onnx"))
        self.decoder = ort.InferenceSession(hf_hub_download(repo_id="pythainlp/thaitts-onnx",filename="tacotron2decoder-th.onnx"))
        self.postnet = ort.InferenceSession(hf_hub_download(repo_id="pythainlp/thaitts-onnx",filename="tacotron2postnet-th.onnx"))
        self.hifi = ort.InferenceSession(hf_hub_download(repo_id="pythainlp/thaitts-onnx",filename="vocoder.onnx"))
    def tts(self, text: str):
        mel = inference(text, self.encoder, self.decoder, self.postnet)
        return self.hifi.run(None, {"spec": mel[0]})

Issue Description

As shown above, in the pythaitts/pretrained/lunarlist_onnx.py, the model "pythainlp/thaitts-onnx" is used as the default model parameter in the hf_hub_download method and download all onnx files. Finally the tacotron2decoder-th.onnx file runs via self.hifi.run.

At the same time, the model is flagged as risky on the HuggingFace platform. The encode.onnx files in these models are marked as risky and may trigger backdoor threats. For certain specific inputs, the backdoor in the models could be activated, effectively altering the model's behavior.

Image

Related Risk Reports:model risk report

Suggested Repair Methods

  1. Convert the model to safer safetensors format and re-upload
  2. Try to regenerate the model using the latest onnx library
  3. Visually inspect the model using OSS tools like Netron. If no issues are found, report the false threat to the scanning platform

As a popular machine learning libraries, every potential risk could be propagated and amplified. Could you please address the above issues?

Thanks for your help~

Best regards,
Mendy

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions