Skip to content

[Model BadCase]: The model always confuses between Chinese and English Punctuations #55

@zimtsui

Description

@zimtsui

Basic Information - Models Used

MiniMax M2

Information about environment and deployment

Official API

Description

Steps to reproduce

The bug can be reproduced with the following steps:

  1. Input a long Chinese text containing Chinese double quote marks “” into the model.
  2. Ask the model to call a function with a substring containing quotes as an argument.

Expected behavior

The model should call the function with an argument which contains the same Chinese quotes as the original text.

Error logs

The actual behavior is that the model always calls the function with an argument which contains English quotes "".

Paste the related screenshots here

Original text:

Image

Function call:

Image

Possible reason

The model is trained in English without much Chinese data. The semantic vectors of Chinese and English quotation marks are too close.

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