Repository navigation
catalog: add Recursive Flint (MLX INT4 and GGUF UD-Q4_K_XL) - #215
Conversation
|
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info
📝 Walkthrough
Merge Risk: ⚪ Minimal · up to
|
sanchitmonga22
left a comment
There was a problem hiding this comment.
Limited current-diff review of 6d30c99: the MLX/GGUF catalog entries use commit-pinned artifact URLs and per-file checksums; both remain explicitly harness_compatible=false. The five kit archive checksums match the published cpp-desktop-v0.20.42 release asset digests. CI and release workflows consistently pin the Swift SDK checkout to fa3b7a46ce57c946cccdb31c48e388e1004a6238, matching the kit source. Version goldens and the generated binding header are updated consistently. No obvious blocking defect found in this scoped diff. Approval does not establish release readiness: current-head platform CI is still running and I have not independently validated Flint MLX or GGUF generation with this build. The author description still contains pre-kit caveats and should be reconciled with actual test results. This does not approve or validate Command Code integration or unspecified URL replacements.
|
You can run this model via this command cc: @sanchitmonga22 |
Adds Recursive Flint, a 40B-total / 2.5B-active MoE coding model (Apache-2.0), as two catalog rows that share the
flintmerge key.mlx-flint-int4RecursiveSI/flint-int4(11 files)flint-ud-q4_k_xlunsloth/flint-GGUFUD-Q4_K_XLBoth are mirrored into
runanywhere/decision-models-stable-fork(mlx/flint-int4/andgguf/flint/). Every URL is pinned to its upload commit, and every weight file carries its SHA-256. The MLX checksums come from the checkpoint's ownflint-int4.json, and I checked them against the uploaded files.The golden outputs for
models list --all(plain and--json) each gain the one mergedflintrow and change nothing else.MLX needs
mlx-swift-lm3.31.13, which is in runanywhere-sdks#1150. Built against that branch throughWALLY_SDK_SWIFT_PATH,wally models show mlx-flint-int4resolves all 11 files. I haven't run a generation through wally yet, so the MLX row is untested end to end. The GGUF row lists and downloads, but it can't load until the C++ kit's llama.cpp supports Flint's architecture. The new kit pins will land in this PR once that kit is released, so don't merge before then.ctest passes (6/6, including the golden corpus) and
cargo fmt --checkis clean.Summary by cubic
Adds Recursive Flint, a 40B-total / 2.5B-active MoE coding model (Apache-2.0), as two catalog rows sharing the
flintmerge key: MLX INT4 and GGUF UD-Q4_K_XL. Bumps thecpp-desktopkit to 0.20.42, which adds llama.cpp Flint support so the GGUF row can now load, and regenerates sys bindings and version goldens.The MLX row is untested end to end; no generation was run through wally.
Written for commit 6d30c99. Summary will update on new commits.
Summary by CodeRabbit
New Features
Chores