CommunityJev Models & Reimplementations
Interfaze Lev
Interfaze's 4B LoRA adapter over Qwen3.5-4B that speaks POST /v1/systemone: 0.807 held-out weighted accuracy and 0.061 ECE after calibration, with a 200,000-example training recipe documented.
- Published by
- Community
- Author
- Interfaze
- Last verified
- Oct 2026
Used inJev for Software & DevTools
- Languages
- Python
- Capabilities
- Open sourceSelf-hostedRuns locally
Highlights
- Speaks TypeSafe's /v1/systemone: lev serve --checkpoint interfaze-ai/lev is the whole deployment.
- One batched forward instead of prefill-and-fork cut engine compute from 169 ms to 69 ms on one H100.
- Serving MASSIVE's 60 intents through label tokens instead of the learned head lifted accuracy from 0.231 to 0.746.
- Held-out weighted accuracy 0.807 and ECE 0.061 after calibration, down from 0.180 before it.
Quickstart
bash
lev serve --checkpoint interfaze-ai/lev --port 8000
curl -s localhost:8000/v1/systemone -H 'Content-Type: application/json' -d '{
"state": "The invoice was paid twice.",
"questions": {"refund": {"type": "noul", "instructions": "Is a refund needed?"}}
}'Watch out
The numbers are self-reported: its 68.9% S1Bench macro trails the 0.761 Jev reference, and it is weakest on minimal-edit pairs and 5-level ratings. Fine-tuning hurt summary consistency (0.271 against Jev's 0.812), and a CUDA GPU is needed for real-time latency; CPU works but takes seconds per call.
Questions about this entry
Short answers drawn from the linked project or page.
Yes for the wire protocol: the server exposes POST /v1/systemone, so TypeSafe SDK clients work after a base-URL change. The card requires Python 3.12 or newer and says a CUDA GPU is needed for real-time use; on CPU a 4B backbone takes seconds per call.
Yes. All questions share one forward pass, so extra questions are nearly free, but the card says answers do not condition on each other — there is no joint distribution across questions, which matters if one answer should depend on another.
Run lev serve --checkpoint interfaze-ai/lev --port 8000, then POST /v1/systemone with a state and typed questions. That one command is the whole deployment; the card requires Python 3.12 or newer.
A CUDA GPU is needed for real-time latency, because one batched forward cut engine compute to 69 ms on an H100. On CPU the 4B backbone works but takes seconds per call.
Held-out weighted accuracy is 0.807 and ECE 0.061 after calibration, down from 0.180 before it. The numbers are self-reported, and its 68.9% S1Bench macro trails the 0.761 Jev reference.
Serving MASSIVE's 60 intents through label tokens instead of the learned head lifted accuracy from 0.231 to 0.746, which is why label-token serving is the documented path for wide option sets.
Yes. The card says fine-tuning hurt summary consistency, which it puts at 0.271 against Jev's 0.812, and it is weakest on minimal-edit pairs and 5-level ratings.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
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 NewsHacker News: A one-token logprob wrapper that also classifies webcam frames.
allanrbo151 pointsA single function Jev-like wrapper for LLMs, including vision modelsRead the thread on Hacker NewsMore Jev resources like this
58GitHub stars
An Apache-2.0 decision engine with a TypeSafe-compatible endpoint, document answers with source passages, and a ViZDoom demo that picks actions from structured game state.
Jev Models & Reimplementations#community#python#local
Community
deepanwadhwa/OpenDecisionPython
deepanwadhwa/OpenDecisionPythonThe Hugging Face hub for Laya's open System 1 checkpoints: the 421M English model at the repo root, a 322M multilingual build in a subfolder, and the typed-decisions fine-tune, all Apache-2.0.
Jev Models & Reimplementations#community#open-models#local
Community
Supersonic Labs' 144.3M-parameter decision model on mmBERT-small: choice, score, and noul answers from supplied options in one pass, 73.15% on 2,000 typed decisions, Apache-2.0.
Jev Models & Reimplementations#community#open-models#local
Community
An open distillation of Jev 1.13 onto Qwen3.8-27B: a 108.9M-parameter decision block answers typed questions in one prefill pass and adds a POST /v1/decide route beside a full System 2.
Jev Models & Reimplementations#community#open-models#local
Community
The vision sibling: the same text-trained System 1 head on Qwen3.8-27B's image encoder, taking images zero-shot, ranked first of 20 vision decision models on the community Jev Decision Index 0.3.
Jev Models & Reimplementations#community#open-models#local
Community
The 9B member of AutoTrust's Jev students: a 40.2M-parameter decision block on Qwen3.5-9B with 0.019 mean KL to Jev 1.13 and 70.7% HumanEval left untouched, small enough for one modest GPU.
Jev Models & Reimplementations#community#open-models#local
Community
From the Jev guides
Original write-ups on what this entry is for.
Interface, size, license, claims — in that order. A selection guide across the system-one catalog, from a 706K form scorer to 27B Jev students.
Jev Practices & Patterns#open-models#evaluation#benchmarks
Read article
Base decision checkpoints sit near chance on purpose. The typed-decisions loop, the corpora behind the students, and the temperature fit that turns probabilities into thresholds.
Jev Practices & Patterns#fine-tuning#calibration#open-models
Read article
Taking a decision checkpoint out of Python: what the GGUF and browser WebGPU exports preserve, what they change, and the parity evidence the model cards publish.
Jev Tools & CLIs#open-models#local#onnx
Read article
From the Jev community
Posts from builders shipping with Jev right now.