gca-americas/discriminative-models-workshop
Organization repository on GitHub
Getting started with Discriminative (Jev/DiffusionGemma) models: workshop workbench
- Forks
- 0
- Last pushed
- Updated Oct 2026
Repo stats from the GitHub API, cached Oct 2026.
Jev directory
CommunityJev Guides & Articles
Getting started with Discriminative (Jev/DiffusionGemma) models
A one-hour, six-step workshop built around a fighting game that teaches TypeSafe's Discriminative (Jev) decision model and puts it next to Gemini in a Google ADK workflow. It runs in Cloud Shell or anywhere signed into gcloud, with hosted Jev, a local DiffusionGemma VM, or a rehearsal stand-in.
- Stars
- 3
- Language
- JavaScript
- Reading time
- 3 min
- Published by
- Community
- Author
- gca-americas
- Last verified
- Oct 2026
- Languages
- JavaScriptPython
- Capabilities
- Open sourceRequires API key
Highlights
- Six steps, about sixty minutes; each page opens with concepts and ends with an exercise, and every check asks the machine rather than the student.
- The Discriminative model (jev-1.13, alias jev-latest) returns typed answers in roughly 70 to 500 ms for $0.042 per million input tokens, with output free.
- Students fight the ogre by hand, then hand reflexes to the model, then watch an ADK workflow where the model decides every tick and Gemini reads spell cards.
- Three model choices: hosted Jev with a key, DiffusionGemma on a 1x NVIDIA L4 VM at about $0.71 per hour, or a rehearsal stand-in with no model.
- Setup runs in Cloud Shell or any machine signed into gcloud; Gemini runs on Vertex AI with the project's credentials and no API key.
Quickstart
bash
./setup_project.sh
./setup_codelab.shWatch out
README and GitHub metadata reviewed 2026-10-17; the repository lists no license, and setup depends on Google Cloud (a project with billing and, for DiffusionGemma, L4 quota), so the full workshop cannot run without a GCP account.
Questions about this entry
Short answers drawn from the linked project or page.
A one-hour workshop on TypeSafe AI's Discriminative model, the System One decision model behind Jev, and on putting it next to Gemini in a Google ADK workflow. It is built around a fighting game with six steps, from playing manually to watching a workflow win.
The model answers typed Choice, Score and Noul questions about the fight state and returns calibrated probabilities in roughly 70 to 500 ms. In the last step an ADK graph uses it for one decision per tick while Gemini reads the spell card image and sings the spell.
In Cloud Shell or anywhere gcloud is signed in, run ./setup_project.sh to create a project with billing and ./setup_codelab.sh to install everything and start the workbench on port 4900. On a laptop, sign in with gcloud auth login and gcloud auth application-default login first.
Jev is billed per token at fractions of a cent per call. DiffusionGemma runs on a g2-standard-4 VM with one L4 for about $0.71 per hour while it runs, or about $10 per month when paused, and rehearsal mode costs nothing. Gemini runs on your own Google credentials.
Only if you choose the hosted Jev model. Step 2 verifies the key with one real request and does not save a refused key. If Jev and DiffusionGemma both fail, or nobody answers, the workshop falls back to rehearsal mode with a stand-in that speaks the real HTTP contract.
agents/arena/agent.py is an ADK graph with a fast loop where the Discriminative model decides every tick and a parallel branch where Gemini reads the spell card PNG and casts the spell. The arena judges the spell, and a bard writes the tale after the fight.
It is aimed at developers learning the Discriminative model. The format follows the Cloud 101 workbench: concepts first, the exercise last, and checks that ask the machine. GitHub user gca-americas publishes it, and the README shows no benchmarks beyond the workshop's own checks.
More Jev resources like this
Python sandbox and tutorial companion that turns Jev Choice and Noul judgments about issue category and reproduction into code-applied queue proposals, with confidence gates, a review fallback, synthetic offline fixtures and labeled replay. Live mode is opt-in.
Jev Guides & Articles#community#issue-triage#guardrails
Community
peakevergreen/jevidencePython
peakevergreen/jevidencePythonAn interactive site by Sarah Drasner that explains System One thinking — fast, automatic judgments versus deliberate checking — and what that distinction means for software design, closing with the Jev × WebMCP Chrome extension as a case study.
Jev Guides & Articles#community#github#javascript
Community
sdras/system-one-explainerJavaScript
sdras/system-one-explainerJavaScriptDemo pack from the French-language Jev meetup of 8 October 2026 on System One models: a Chrome ad-blocker, an unclutter browser fork, a decision-routing kit comparing a local engine with Jev via OpenRouter, and n8n email triage.
Jev Demos & Experiments#community#github#demos
Community
royam0820/Jev-MeetupJupyter Notebook
royam0820/Jev-MeetupJupyter NotebookAn unofficial TypeSafe Jev integration for Claude Code with hooks that suggest skills, classify task type, pick subagent model tiers and gate risky shell commands. It also ships jev-vision for yes/no, pick-one and bounding-box questions about images.
Jev Integrations & Adapters#community#github#javascript
A free Claude Code skill that puts Jev in front of the expensive model: thirteen commands sort, check, score, gate and verify with Jev, so Claude reads only the unsure slice and writes the output. Needs python3 and an OpenRouter key.
Jev Agents & MCP#community#python#claude-code
Community
charlesdove977/claude-x-jevPython
charlesdove977/claude-x-jevPythonThree open-weights 4B decision models for English (Deem-4B), Turkish (Karar-4B) and German (Wähler-4B) that choose between options and return a probability for each. The server implements the Jev API, POST /v1/systemone, so existing Jev clients work by changing the endpoint.
Jev Models & Reimplementations#community#github#python
Community
mertkayacs/jevaltPython
mertkayacs/jevaltPythonFrom 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.