How do I install the Mlx Convert skill?

Run npx skillmds@latest add mabaeyens/mlx-convert in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Mlx Convert safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Mlx Convert?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Mlx Convert free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Mlx Convert?

mabaeyens (@mabaeyens) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Mlx Convert by mabaeyens — Agent Skill

Name: Mlx Convert
Author: mabaeyens

Mlx Convert

Convert a HuggingFace model to MLX and publish it to mlx-community, end to end (download, pick the right tool, convert preserving all modalities, verify, write a real card, upload, update tracking docs, clean up locally) via a single script -- no manual steps to remember. Invoke with /mlx-convert <hf-repo> [q-bits] [q-group-size].

mabaeyens 468bf2a 2.8 KB Updated Sep 17, 2026

File contents

This skill is a thin wrapper. All the actual logic lives in mlx-conversions/scripts/pipeline.py -- run it, don't re-derive the steps by hand:

cd ~/Projects/mlx-conversions
source ~/.zprofile   # HF_TOKEN
uv run python scripts/pipeline.py --hf-path <source-repo> --q-bits 4 --q-group-size 64

For an unquantized (bf16 passthrough) build of a vision-language model:

uv run python scripts/pipeline.py --hf-path <source-repo> --no-quantize

What the script does (so you know what NOT to redo manually)

Fetches the source's config.json, decides mlx_lm (text-only) vs mlx_vlm (vision-language) automatically -- checks for vision_config/audio_config or a *ForConditionalGeneration architecture. Never assume text-only.
Checks mlx-community doesn't already have this exact repo/quant first -- exits early if so.
Converts, preserving whatever modalities the source has (never silently drops vision/audio components -- this was a real bug found 2026-07-18, see mlx-conversions/specs/model-verification.md).
Uploads to the target repo (defaults to mlx-community/<name>-<bits>bit).
Runs scripts/verify.py's structural (tensor-prefix diff) and functional (real generation, image+text too for VLMs) checks.
Writes a real model card: description pulled from the source, honest verification-status note, quantization/provenance notes, discoverability tags.
Updates BACKLOG.md (checks off the source if it was a tracked candidate) and appends a CHANGELOG.md entry.
Deletes the local copy once verified. Keeps it (for debugging) only if verification failed.

When NOT to use this

A one-off fix to something already published (e.g. just re-uploading a corrected README) -- use the specific script for that (scripts/write_model_cards.py, scripts/fix_processor_metadata.py), not the full pipeline.
Publishing multiple precision variants of the same family in one go -- scripts/convert_all_precisions.py already encodes that loop; pipeline.py handles one (source, target) conversion at a time.
Text-only models with --no-quantize -- not supported yet (scripts/convert.py always quantizes); the script will refuse and tell you so rather than silently ignoring the flag.

After running

Report what got published (or why it was skipped/failed) concisely. Don't re-explain the steps above -- the user already knows what the script does once it's been run a few times; just report the outcome.

---name: mlx-convertdescription: Convert a HuggingFace model to MLX and publish it to mlx-community, end to end (download, pick the right tool, convert preserving all modalities, verify, write a real card, upload, update tracking docs, clean up locally) via a single script -- no manual steps to remember. Invoke with /mlx-convert <hf-repo> [q-bits] [q-group-size].---This skill is a thin wrapper. All the actual logic lives in`mlx-conversions/scripts/pipeline.py` -- run it, don't re-derive the steps byhand:```bashcd ~/Projects/mlx-conversionssource ~/.zprofile # HF_TOKENuv run python scripts/pipeline.py --hf-path <source-repo> --q-bits 4 --q-group-size 64```For an unquantized (bf16 passthrough) build of a vision-language model:```bashuv run python scripts/pipeline.py --hf-path <source-repo> --no-quantize```## What the script does (so you know what NOT to redo manually)1. Fetches the source's `config.json`, decides `mlx_lm` (text-only) vs `mlx_vlm` (vision-language) automatically -- checks for `vision_config`/`audio_config` or a `*ForConditionalGeneration` architecture. Never assume text-only.2. Checks `mlx-community` doesn't already have this exact repo/quant first -- exits early if so.3. Converts, preserving whatever modalities the source has (never silently drops vision/audio components -- this was a real bug found 2026-07-18, see `mlx-conversions/specs/model-verification.md`).4. Uploads to the target repo (defaults to `mlx-community/<name>-<bits>bit`).5. Runs `scripts/verify.py`'s structural (tensor-prefix diff) and functional (real generation, image+text too for VLMs) checks.6. Writes a real model card: description pulled from the source, honest verification-status note, quantization/provenance notes, discoverability tags.7. Updates `BACKLOG.md` (checks off the source if it was a tracked candidate) and appends a `CHANGELOG.md` entry.8. Deletes the local copy once verified. Keeps it (for debugging) only if verification failed.## When NOT to use this- A one-off fix to something already published (e.g. just re-uploading a corrected README) -- use the specific script for that (`scripts/write_model_cards.py`, `scripts/fix_processor_metadata.py`), not the full pipeline.- Publishing multiple precision variants of the *same* family in one go -- `scripts/convert_all_precisions.py` already encodes that loop; `pipeline.py` handles one (source, target) conversion at a time.- Text-only models with `--no-quantize` -- not supported yet (`scripts/convert.py` always quantizes); the script will refuse and tell you so rather than silently ignoring the flag.## After runningReport what got published (or why it was skipped/failed) concisely. Don'tre-explain the steps above -- the user already knows what the script doesonce it's been run a few times; just report the outcome.

Frequently asked questions