Instructions to use Red-fuchs/ncm-modular with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use Red-fuchs/ncm-modular with MLX:
# Make sure mlx-lm is installed # pip install --upgrade mlx-lm # if on a CUDA device, also pip install mlx[cuda] # Generate text with mlx-lm from mlx_lm import load, generate model, tokenizer = load("Red-fuchs/ncm-modular") prompt = "Once upon a time in" text = generate(model, tokenizer, prompt=prompt, verbose=True) - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
- MLX LM
How to use Red-fuchs/ncm-modular with MLX LM:
Generate or start a chat session
# Install MLX LM uv tool install mlx-lm # Generate some text mlx_lm.generate --model "Red-fuchs/ncm-modular" --prompt "Once upon a time"
- Atomic Chat
Download PUBLISH.md from Red-fuchs/ncm-modular: direct link, hf CLI and curl.
- Browser
- Download file 2.87 kB
-
https://huggingface.co/Red-fuchs/ncm-modular/resolve/main/PUBLISH.md
- Command line
-
hf download hf://Red-fuchs/ncm-modular/PUBLISH.md
-
curl -L -o PUBLISH.md https://huggingface.co/Red-fuchs/ncm-modular/resolve/main/PUBLISH.md
Publish checklist (NOTHING HERE HAS BEEN RUN)
Nothing in this package uploads, creates a repo or pushes. Run these yourself, in order, after a last look.
Pre-flight (local, safe)
cd ~/Claude-Workspace/ncm-modular
git status --short && git log --format='%h %an <%ae> %s' # review; author must be neutral
grep -rnE "/Users/|@[a-z0-9-]+\.[a-z]{2,}|hf_[A-Za-z0-9]{10,}|ancap|Obsidian|iCloud" . --exclude-dir=.git --exclude=PUBLISH.md # must be empty
du -sh . ; git count-objects -vH # expect well under 5 MB
Optional: drop this file before publishing (git rm PUBLISH.md && git commit -m "Drop publish notes").
GitHub (public, pushes current branch)
gh auth status
gh repo create Red-fuchs/ncm-modular --public --source . --push \
--description "Modular Next-Concept Modeling on Apple Silicon (MLX): codec port, surrogate masking, block-seam experiment, honest logs"
gh repo edit Red-fuchs/ncm-modular --add-topic ncm --add-topic next-concept-modeling --add-topic mlx \
--add-topic apple-silicon --add-topic text-generation --add-topic diffusion-lm --add-topic vq-vae \
--add-topic modular-llm --add-topic local-llm --add-topic research-code
Suggested description: "Modular Next-Concept Modeling on Apple Silicon (MLX): codec port, surrogate masking, block-seam experiment, honest logs". Topics: ncm, next-concept-modeling, mlx, apple-silicon, text-generation, diffusion-lm, vq-vae, modular-llm, local-llm, research-code.
Hugging Face (model repo, no weights; README.md front matter is the model card)
hf auth whoami
hf repo create Red-fuchs/ncm-modular
hf upload Red-fuchs/ncm-modular . . --repo-type model \
--commit-message "Initial release: MLX codec port, masking, MASSIVE pilot, seam experiment" \
--exclude ".git/*" "PUBLISH.md" "**/__pycache__/*" "w/*" "*.safetensors" "*.pt" "*.npz"
hf repo create defaults to a model repo; add --private first if you want a look before it goes public. Check the exact flags with hf repo create --help and hf upload --help for your version.
Decisions already taken
- Name
ncm-modular; python foldersrc/ncm_modular. Credit to HazyResearch NCM and janmus1c/ncm-mlx is in README and NOTICE. - TinyStories gold excerpts in
results/seam/samples.mdkept, with attribution + CDLA-Sharing-1.0 header. - Organisation names and common fictional-looking names in test sentences kept; real person names replaced; outputs re-run.
- Gemma / JANG converter and benches stay excluded (LICENSES.md, UNCLEAR). Only unverified side-note numbers remain in RESULTS.md section 5.
Still open
- MASSIVE: confirm CC BY 4.0 on the dataset card you downloaded (only aggregate metrics are shipped).
- Courtesy note to the janmus1c / HazyResearch authors is your call (nothing was sent).
- The TinyStories codec round trip (62 %) has no stored log.