Skip to content

[Feature]: Tokenizer path different from model name #302

@attafosu

Description

@attafosu

Motivation

For a remote server, the served model name may be different from its tokenizer name/path. The tokenizer is not loaded in such case, and thus tokens-level stats may be incomplete at the end of the run.

Proposed Solution

Add an optional tokenizer_path parameter to ModelParams and use when the model's name parameter isn't a valid path (either locally or on huggingface)

Alternatives Considered

No response

Additional Context

No response

Metadata

Metadata

Assignees

No one assigned

    Labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions