Flint INT4: community serving-contract proposal for an additional local API path
The official Rec instructions remain the primary path for this INT4 pack. I propose a clearly separated community note for Rapid-MLX's local API integration.
There are existing discussions about packed weights and reasoning-budget behavior. I am not claiming those reports are fixed by switching servers. The concrete implementation references below cover Flint checkpoint/template identity, native prompt rendering, tool argument replay and streaming parity in Rapid's tests.
Before any executable deployment recipe, the exact publisher-owned pack should pass a real load, a bounded-thinking request, and a parsed tool/result cycle. The note should record the effective request budget and preserve your Rec reference instructions. This is a compatibility-documentation proposal with source contracts available for review, not a benchmark or a resolution of the existing bug reports.
Concrete catalog entry: flint-40b-a2.5b-4bit → RecursiveSI/flint-int4@3b93f9053b530488405ae9fca51d4900d6ce21f0. This is a publisher-owned artifact.
Source references (fixed Rapid commit): catalog, Flint contracts, native prompt tests, routing 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.