Jev directory
CommunityJev Models & Reimplementations
Kev-4B MLX FP16
An MLX float16 package of jaredpalmer/kev-4b for Apple silicon: the backbone with its adapter merged, a float32 pointer head, the tokenizer, and a package.json carrying per-file SHA-256 digests.
- Reading time
- 2 min
- Published by
- Community
- Author
- Verathos AI
- Last verified
- Oct 2026
Works forJev for Classification
Used inJev for Software & DevTools
- Platforms
- macOS
- Capabilities
- Open sourceSelf-hostedRuns locally
Highlights
- Repackages kev-4b at revision 6cfce5c2fa4b4bd64026336ab649c5ca78857d52 for the Verathos private lane app on Apple silicon.
- The backbone ships with the adapter merged in float16; the DeltaNet decay and gated-norm weights and the pointer head stay float32.
- package.json records every file's SHA-256 and size; the package digest is a590b5aec27e91b6732ac3227d0859e3c5817fd60b4dec3e6e3ad158e1752465.
- Apache-2.0, the same as kev-4b and its base model Qwen/Qwen3.5-4B-Base.
Watch out
Card reviewed 2026-10-17; the card is only a packaging note with no benchmarks, latency or accuracy figures, so quality has to be judged from kev-4b's own card, and the package is scoped to the Verathos private lane app.
Questions about this entry
Short answers drawn from the linked project or page.
An MLX package of jaredpalmer/kev-4b at a pinned revision, prepared for the Verathos private lane app on Apple silicon: merged backbone in float16, float32 pointer head, tokenizer and a digest manifest.
The backbone with the adapter merged is float16; the DeltaNet decay and gated-norm weights and the pointer head remain float32 for stability. No retraining or quantisation is involved.
package.json lists every file's SHA-256 and size, and its own SHA-256 digest is a590b5aec27e91b6732ac3227d0859e3c5817fd60b4dec3e6e3ad158e1752465, so you can check integrity end to end.
Revision 6cfce5c2fa4b4bd64026336ab649c5ca78857d52 of jaredpalmer/kev-4b, pinned explicitly in the card so the package cannot silently move with upstream updates.
Apache-2.0, matching kev-4b and its base model Qwen/Qwen3.5-4B-Base. The card states the licence directly; the package adds no extra restrictions of its own.
No. This is a packaging card only: it documents the files, formats and digests, and publishes no accuracy, latency or calibration numbers, so evaluation must come from kev-4b's own card.
It is an MLX package for Apple silicon, so it is consumed by MLX-based runtimes; the card frames the target as the Verathos private lane app rather than a general CLI or server.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
X: Standard One SH (StandardThinking)
LinkedIn: What multimodal decision models actually change, and what they do not.
LinkedInJevDirectory.org
Hacker News: Decision Index 0.2.1 scores open reproductions of Jev on a frozen suite.
jackbravo4 pointsDecision Index 0.2.1 – 70 open reproductions of TypeSafe Jev's Decision ModelRead the thread on Hacker NewsHacker News: Pointer to a DiffusionGemma plus vLLM tutorial for a Jev-shaped classifier.
1 pointHow to build a Jev-style classifier with DiffusionGemma and vLLMRead the thread on Hacker NewsMore Jev resources like this
Serves Shanghai AI Lab's Intern-Decision-0.8B vision decision model on Apple Silicon behind a /v1/systemone-shaped endpoint: screenshots plus typed questions in, calibrated Choice, Score and Noul probabilities out, for thresholding and escalation.
Jev Models & Reimplementations#community#apple-silicon#local
An 8-bit MLX conversion by SirSahOl of chaoliangUNSW/Jev-Style-0.8B-Decision-v3 built for Apple Silicon native GPU inference: about 990 MB of active VRAM, a 262,144-token context and an average of 8.25 bits per weight.
Jev Models & Reimplementations#community#open-models#local
Community
Ruiruiz30's Apple Silicon MLX conversion of Jev-Omni: 4-bit affine language weights on a Mac mini M4 16GB, about 7 GB peak Metal memory and roughly 963 ms text decisions, Apache-2.0.
Jev Models & Reimplementations#community#open-models#local
Community
An Apple Silicon MLX bf16 build of the v1 Jev-Style decision model: 77 ms decisions on an M1 Max with calibrated choice, bool and score probabilities, Apache-2.0 weights.
Jev Models & Reimplementations#community#open-models#local
Community
An independent Apple Core AI conversion of the JEV-9B text backbone with 8-bit weights and fp16 activations for on-device generation on Apple silicon; it excludes the System 1 decision head, calibration and vision.
Jev Models & Reimplementations#community#open-models#local
Community
A calibrated image-capable 27B decision model that fits an 8 GB GPU: the JEV LoRA-and-head recipe trained on Prism ML's ternary Bonsai 2, returning typed probabilities with no generated text.
Jev Models & Reimplementations#community#open-models#local
Community
From the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.