Skip to content

Inquiry About Pre-training Data Processing #8

@Makoto2303

Description

@Makoto2303

I’m very interested in understanding how your large language model processes raw pre-training data into its final training format. Specifically, I’d like to ask:

1,.Data Format Conversion:
Is the original text retained in its raw form (e.g., as plain text or structured fields), or is it transformed into other representations (e.g., tokenized IDs, binary encodings, or embeddings)?

2.Field Connection:
How are different fields or data segments connected during preprocessing? For instance, are delimiters (like commas or special tokens) used, or is another method employed?

3.Example Request:
If possible, could you share a simplified example of how a raw input (e.g., a structured climate record) is converted into the format used for training?

Thank you in advance for your time and expertise! I’d greatly appreciate any insights you can provide.

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