Hy3 preview: community Mac serving proposal with suffixed reasoning-tag checks

#12
by Raullen - opened

Proposing a community Apple Silicon deployment note for Hy3 preview, focused on its model-specific reasoning and tool protocol.

Rapid's Hy3 reasoning tests include suffixed opening/closing tags, tags split across streaming chunks, and incomplete-prefix finalization. Those are relevant to keeping internal reasoning out of the visible client response. The final example should pair those parser contracts with a real chat/tool-result round trip on the pinned 4-bit artifact and your recommended sampling settings.

The Tencent Hy Community model license should remain explicit; a runtime's license does not replace it. Hardware sizing, full-context operation and upstream benchmark reproduction would stay out of scope until measured. The source below is a concrete starting point for reviewing whether this belongs in community deployment resources.

Concrete catalog entry: hy3-preview-4bit → mlx-community/Hy3-preview-4bit@8e4d56f18efd912b8c7581a8ccfa8b2a79ba3469. This is a conversion/distribution artifact; it is not presented as an official upstream release.

Source references (fixed Rapid commit): catalog, Hy3 protocol tests, chat-template tests.

Verification status: this proposal is based on reading the catalog, implementation and the linked documentation/tests. No fresh model inference or test-suite run was performed for this submission; source tests are not an end-to-end qualification. It requests a community documentation scope, not an official endorsement or a performance listing.

Disclosure: I maintain Rapid-MLX. Prepared with AI assistance under my direction. Anonymous usage-metadata telemetry is enabled by default; DO_NOT_TRACK=1 disables it. The original model and conversion terms continue to apply.

Sign up or log in to comment