Skip to content

Missing FP8 quantized MiniMax-H3 models referenced in configs #1445

Description

@nanwanj

Hi,

While going through the source code and config folder, we noticed that several H3 configs (e.g., configs/minimax_h3/fp8) reference a MiniMax-H3_quantized/fp8 model. However, we couldn't find this model in your Hugging Face repo.

Additionally, the model conversion/quantization tool (tools/convert) doesn't appear to support H3 models, so we're unable to generate the quantized weights ourselves.

Could you let us know where the FP8 quantized H3 models can be downloaded, or whether there are plans to release them (or add H3 support to the conversion tool)?

Thanks!

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    Status
    Todo

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions