vinnj/Ling-3.0-flash-TypeScript-Edition-GGUF

Ling-3.0-flash-TS128-Q4_K_M (TypeScript Edition)

47/48 on six TypeScript tasks, at about 20 tokens/s on a 16 GB Mac. gpt-5-mini: 46/48.

Ling-3.0-flash-TS128-Q4_K_M.gguf (20,395,487,680 bytes, sha256: e9ae4335a36ed49e8262add1099459a807692edb18c6252ba855f0923c52f602) is a 20.4 GB TypeScript-focused expert cut of InclusionAI's inclusionAI/Ling-3.0-flash (cut from 512 to 128 experts per MoE layer across 40 MoE layers, 8 active per token), paired with the Kraits 0.6.17 plugin for Apple silicon Macs with 16 GB or more on macOS 15 or later.

Measured results (same prompts for both models)

Measured by the release tester, Mac mini M4 16 GB, Fri 6 Nov 2026, temperature 0, launcher as shipped:

• Speed: writing 19.6 to 20.7 tokens/s (mean 20.4), reading 37 to 45 tokens/s.

Task	Ling TS128 + Kraits 0.6.17	gpt-5-mini
Express route	7	8
Types and generics	8	8
Async error handling	8	8
Jest test suite	8	6 (average of 3 runs: 8, 8, 2)
JS to strict TypeScript	8	8
React component	8	8
Total	47/48	46/48

The Jest task uses a reworded prompt written by the release tester and unseen by the model's tuning. Ling's answers are identical on every run at temperature 0. gpt-5-mini's answers vary between runs, so its Jest score is the average of 3 runs; its other five scores are single runs from 2 Nov.

Files

File	Where to get it	Bytes	SHA-256
Ling-3.0-flash-TS128-Q4_K_M.gguf	This Hugging Face repository	20,395,487,680	e9ae4335a36ed49e8262add1099459a807692edb18c6252ba855f0923c52f602
Ling-3.0-flash-TS128-Q4_K_M.kraits-pins.txt (for Kraits 0.6.17)	Inside kraits-v0.6.17-macos-arm64.tar.gz (release-0617/Ling-3.0-flash-TS128-Q4_K_M.kraits-pins.txt)	16,192	b430f9b004cc6ea5da2f84aba52ab90e7be5c378a4985886d9ba67515b7f5a23

(Note: the Ling-3.0-flash-TS128-Q4_K_M.kraits-pins.txt file hosted directly in this Hugging Face repo with sha256: 7a5c3f33... is for Kraits 0.6.16. For Kraits 0.6.17, always copy release-0617/Ling-3.0-flash-TS128-Q4_K_M.kraits-pins.txt out of the 0.6.17 release tarball as shown below.)

How to run on a 16 GB Mac (Kraits 0.6.17)

Download the Kraits 0.6.17 release bundle and the model file, copy the 0.6.17 pin list out of the unpacked tarball so it sits beside the model file, and run:

curl -LO https://github.com/1picassoai/kraits/releases/download/v0.6.17/kraits-v0.6.17-macos-arm64.tar.gz
curl -LO https://huggingface.co/vinnj/Ling-3.0-flash-TypeScript-Edition-GGUF/resolve/main/Ling-3.0-flash-TS128-Q4_K_M.gguf
tar -xzf kraits-v0.6.17-macos-arm64.tar.gz
cp ./release-0617/Ling-3.0-flash-TS128-Q4_K_M.kraits-pins.txt .
./release-0617/kraits-server.sh ./Ling-3.0-flash-TS128-Q4_K_M.gguf 8080

Verify the model file and pin list hashes:

shasum -a 256 Ling-3.0-flash-TS128-Q4_K_M.gguf Ling-3.0-flash-TS128-Q4_K_M.kraits-pins.txt
# e9ae4335a36ed49e8262add1099459a807692edb18c6252ba855f0923c52f602  Ling-3.0-flash-TS128-Q4_K_M.gguf
# b430f9b004cc6ea5da2f84aba52ab90e7be5c378a4985886d9ba67515b7f5a23  Ling-3.0-flash-TS128-Q4_K_M.kraits-pins.txt

Credits and licence

• Base model: inclusionAI/Ling-3.0-flash by InclusionAI.
• Engine plugin: Kraits 0.6.17 (Apache 2.0).