Skip to content

Gibberish output when using qwen with memory set to low. #4346

@lnux236

Description

@lnux236

My device is poco x3 gt. 8gb ram, dimensity 1100, mali g77 mc9. Mnn engine is very fast. The problem is when I use qwen models and feed them with images they produces gibberish. The fix is to change the memory in config.json from low to normal. The downside the processing and inference speed became very slow.

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