[Feature]: Enable BM25 support on RISC-V / riscv64

[ihb2032]

Problem / Motivation

When using zvec on Linux RISC-V / riscv64, the package itself can be installed successfully, but BM25 functionality is not usable.

The issue appears to be caused by the DashText dependency. DashText currently does not provide a Linux riscv64 wheel on PyPI, and I also could not find a source distribution that can be built locally on RISC-V.

As a result, users can install zvec on RISC-V, but cannot use BM25-related features.

Proposed Solution

Alternatives Considered

No response

Affected Area

No response

[Cuiyus]

Hi @ihb2032

The BM25 functionality in zvec relies on DashText as a third-party dependency. Unfortunately, DashText is not
open-sourced, so building it from source on RISC-V is not an option, and we don't control its platform support or
release schedule.

Since this is a limitation outside of our control, we'd suggest exploring alternative tokenizers on your side — if
you can find a RISC-V-compatible tokenizer that provides equivalent functionality, it may serve as a drop-in
replacement for your BM25 workflow.

If you find a viable alternative or a solution that requires changes on our end, feel free to share your findings
here and we can discuss further.