VibeThinker 1.5B: proposal for bounded local-serving and reasoning-output documentation

#12
by Raullen - opened

Proposing a community Mac-serving note for VibeThinker-1.5B that makes output budgeting and reasoning completion explicit.

Rapid-MLX has a VibeThinker family route and reasoning handling, but your card recommends allowing a substantial output budget. A short successful load or arithmetic response is not sufficient evidence for that configuration. The example should record the actual budget, distinguish a complete final answer from truncation, and test streamed reasoning separation.

The 3B catalog entry is a separate checkpoint and should not inherit a 1.5B test result. Function/XML behavior also needs verification against the selected artifact rather than assuming a generic Qwen parser. The linked implementation and protocol tests provide a starting point for a narrowly scoped community deployment example.

Concrete catalog entry: vibethinker-1.5b-4bit β†’ mlx-community/VibeThinker-1.5B-mlx-4bit@2c092d732009cb3a2dd4871456c0f1c4e3c0ccec. This is a conversion/distribution artifact; it is not presented as an official upstream release.

Source references (fixed Rapid commit): catalog, family configuration, function/XML 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