8000 Tags · infinigence/ApxInf · GitHub
[go: up one dir, main page]

Skip to content

Tags

Tags

v0.2

Toggle v0.2's commit message

Verified

This commit was created on GitHub.com and signed with GitHub’s verified signature.
refactor(mlx): let Metal families own model_variant admission (#133)

The common loader kept a hard-coded list of Qwen3/MiniCPM5 names that may
pass model_variant on Metal. The family loaders already parse and reject
their variants, and a Metal request can only reach a registered metal_
factory, so the list only forced a central edit for every new Metal family.

- Metal requests defer variant admission to the selected family. The
  non-Metal list is unchanged. An unregistered Metal family with a variant
  still fails with UnsupportedDevice before device or checkpoint access.
- metal_qwen3 and metal_minicpm5 move their existing option checks into a
  device-independent admit(), tested for the default, alias, invalid
  variant and DSpark asset rules without a GPU.
- Update mlx-backend.md and model-layer-architecture.md to match.

Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

v0.1

Toggle v0.1's commit message

Verified

This commit was created on GitHub.com and signed with GitHub’s verified signature.
docs: update readme (#67)

0