Skip to content

Lines 46-61 in infra_utils.py fails for Llama models #1

@r-gould

Description

@r-gould

Line 46 of lit/utils/infra_utils.py splits text with "[INST]" which is only present in the Mistral chat template (resulting in line 47 throwing an error for non-Mistral models), and line 49 similarly for "<|User|>" which is only present for DeepSeek models. Should lines 46-48 and lines 49-61 each be wrapped in if blocks?

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions