User repository on GitHub
- Forks
- 36
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
CommunityJev Models & Reimplementations
Djev Run
Serves DiffusionGemma-Jev behind a TypeSafe-compatible API on Google Cloud Run with an NVIDIA RTX PRO 6000 GPU, built on mmastrac/djev. Cold start is about 47.5 s and the median server response 61 ms; it exposes /v1/systemone plus snake, dino and tetris demos.
- Stars
- 582
- Language
- HTML
- Published by
- Community
- Last verified
- Oct 2026
About this project
Serve DiffusionGemma-Jev (djev) on Cloud Run with an NVIDIA RTX PRO 6000 Blackwell GPU, built on mmastrac/djev. Follows Cloud Run GPU best practices. You have two options for deployment: Deploy the pre-built image with /dev/shm staging and all vllm serve flags baked in. Directly deploy the upstream Docker container running pure OpenAI completions. ghcr.io/taeold/djev-run:latest also includes support for the /v1/systemone endpoint: Cold start from zero instances: 47.5s (down from 4m 05s baseline)
Quoted from the project's README. Read the full README.
Works forJev for Classification
Used inJev for Software & DevTools
Highlights
- Serves DiffusionGemma-Jev on Cloud Run with an RTX PRO 6000 Blackwell GPU; cold start about 47.5 s
- Median server response 61 ms and 117 ms end-to-end at steps=1; about 100-123 requests/sec at concurrency 32
- Exposes a /v1/systemone endpoint returning choice probabilities and timing diagnostics
- Ships /snake, /dino and /tetris demos wired to the decision endpoint
- JevBench v1.3.0 (N=231): 81.4% accuracy, 73.4 composite score, 117 ms median
Watch out
No license file, so the code is all rights reserved by default. Requires a Google Cloud project with RTX PRO 6000 quota, a GCS bucket and an 18 GB model download, at roughly $3.19 per active GPU hour.
Questions about this entry
Short answers drawn from the linked project or page.
A Google Cloud project with RTX PRO 6000 quota in a supported region, a GCS bucket for the DiffusionGemma-Jev weights, and gcloud. The README provides a pre-built container and a raw vLLM nightly option.
Yes. The pre-built image includes a /v1/systemone endpoint that takes state and typed questions and returns choices, probabilities and confidence with timing diagnostics.
About 47.5 seconds from zero instances, down from a 4 minute 5 second baseline. That is achieved by staging weights with /dev/shm and baking the vllm serve flags into the image, with a startup probe whose failure threshold tolerates the load time.
About 100 to 123 requests per second at concurrency 32, a median of 61 ms server-side and 117 ms end-to-end at steps=1, samples=1, which becomes 121 ms with samples set to auto. The deployment asks for one RTX PRO 6000, 20 vCPU, 80 GiB RAM, no CPU throttling and no zonal GPU redundancy.
On the JevBench v1.3.0 public suite of 231 cases, djev-run at steps=1, samples=1 reaches 81.4% accuracy and a 73.4 composite score at 117 ms median, and with samples=auto reaches 81.8% and 73.5 at 121 ms. The hosted api.djev.dev reference posts 81.8% and 73.0 at 237 ms.
One NVIDIA RTX PRO 6000 at 20 vCPU and 80 GiB RAM is $3.19 per hour while active, and it scales to $0 when idle because the deployment uses min-instances=0. There is also an 18 GB model download to stage into the bucket, and the repository has no license file, so the code is all rights reserved by default.
Three, each wired to the decision endpoint: /snake, /dino and /tetris, adapted from laya-coreml, laya-vs-jev and jev-tetris respectively. They exist to show the served model driving a real interactive loop rather than to benchmark anything.
More Jev resources like this
107GitHub stars
Cross-platform desktop app and serving system for open Jev-like decision models. It runs laya, kev, Jev-Omni and others locally behind the Jev API at POST /v1/systemone, with a playground, versioned templates, decision history and a LoRA Train page.
Jev Models & Reimplementations#community#desktop-app#local-ai
Community
BudEcosystem/Bud-Decision-StudioHTML
BudEcosystem/Bud-Decision-StudioHTML6.9kGitHub stars
A native MLX runtime for Laya's open typed-decision weights: short decisions on Apple Silicon, with no generated tokens, PyTorch, or cloud API.
Jev Models & Reimplementations#community#python#open-models
Community
mizorewww/laya-mlxPython
mizorewww/laya-mlxPython4.8kGitHub stars
An open-model semantic-if engine positioned as a local Jev-style decision stack, separate from the original SemIf repo but sharing the same author.
Jev Models & Reimplementations#community#python#open-models
Community
TheoLeeCJ/SemIf-OpenJevPython
TheoLeeCJ/SemIf-OpenJevPython2.2kGitHub stars
Open recipe for a local Jev-style decision model: a Qwen3.5-9B LoRA that scores one allowed answer token per Choice, boolean or rubric Score question, with data pipeline, training config and eval harness. Reports 90.1% on its 324-example holdout against Jev's 93.2%.
Jev Models & Reimplementations#community#lora#qwen
Community
bespokelabsai/nimblePython
bespokelabsai/nimblePython1.6kGitHub stars
Laya typed decisions through Apple Core ML and the Neural Engine, with the author's speed and energy benchmarks for short decisions on an M3 Max.
Jev Models & Reimplementations#community#python#open-models
Communitymizorewww/laya-coremlPython
mizorewww/laya-coremlPython1.7kGitHub stars
Open-weight 4B and 9B models from TokenRhythm: NeoHorse-1 for tool use, coding, and instruction following, plus NeoHorse-Jev-4B for prefill-only Choice, Noul, and Score decisions.
Jev Models & Reimplementations#community#jevbest
Community
TokenRhythm/NeoHorsePython
TokenRhythm/NeoHorsePythonFrom the Jev guides
Original write-ups on what this entry is for.
Jev takes text only; Clef-omni adds image, audio, and video input and Liquid d1 adds images, each keeping the same typed-answer shape. Where that matters, where it does not, and the costs the vendors underplay.
Jev Models & Reimplementations#multimodal#open-models#decision-model
Read article
One model in September, a dozen by October. Clef, Liquid d1, Perplexity, and OpenAI all ship the same typed-answer shape now. How to read the rival benchmarks and pick on your own task.
Jev Models & Reimplementations#decision-model#benchmarks#pricing
Read article
Hundreds of repositories carry the system-one tag, but the ecosystem is a dozen ideas in different sizes. A tour of the families, the licenses, and what a download does not include.
Jev Models & Reimplementations#open-models#hugging-face#local
Read article
From the Jev community
Posts from builders shipping with Jev right now.