Jev directory
CommunityJev Models & Reimplementations
Turnike-JEV-4B Merged
The berkay123001/turnike-jev-4b LoRA adapter merged into Qwen/Qwen3.5-4B in bf16 so it can be served without PEFT: a bilingual Turkish and English decision model whose parent card reports 0.774 JevBench macro accuracy.
- Reading time
- 2 min
- Published by
- Community
- Author
- berkay123001
- Last verified
- Oct 2026
- Languages
- Python
- Capabilities
- Open sourceSelf-hostedRuns locally
Highlights
- bf16 merge of the turnike-jev-4b LoRA adapter into Qwen/Qwen3.5-4B, the same merge karar.py performs at load time.
- Use it without PEFT: Karar(None, taban="berkay123001/turnike-jev-4b-merged") answers typed questions with option probabilities.
- Apache-2.0 repository; the base Qwen/Qwen3.5-4B is Apache-2.0, Copyright 2026 Alibaba Cloud.
- The card defers evaluation, training data, limitations and declarations to the turnike-jev-4b adapter card.
Quickstart
python
from karar import Karar # karar.py from berkay123001/turnike-jev-4b
k = Karar(None, taban="berkay123001/turnike-jev-4b-merged")
k.sor("Kulaklık bozuk geldi, paramı geri istiyorum.", {"Yazar iade istiyor mu?": {"type": "noul"}})Watch out
Card reviewed 2026-10-17; this card publishes no evaluation of its own and defers to the turnike-jev-4b adapter card, where the 0.774 JevBench macro accuracy, the Turkish results and the latency numbers are self-reported by the author rather than independently reproduced.
Questions about this entry
Short answers drawn from the linked project or page.
The LoRA adapter berkay123001/turnike-jev-4b merged into Qwen/Qwen3.5-4B in bf16, the same merge karar.py performs at load time. It is published so the 4B Turkish and English decision model can be loaded without PEFT.
The decisions are the same; this repository stores the already-merged bf16 weights, while the adapter repository ships a roughly 130 MB LoRA (r=16) that karar.py merges while loading. Choose this one to skip PEFT at load time.
It takes a text and closed questions and returns option probabilities in a single forward pass with no generated text, following the same typed-question shape as the rest of the Turnike family and Jev's calibrated-probability contract.
Get karar.py from berkay123001/turnike-jev-4b, then call k = Karar(None, taban="berkay123001/turnike-jev-4b-merged") and k.sor(text, questions). Passing taban points the loader at the merged repository instead of the adapter.
Apache-2.0 for the repository, with the base model Qwen/Qwen3.5-4B also Apache-2.0, Copyright 2026 Alibaba Cloud, per the card.
The card forwards you to berkay123001/turnike-jev-4b for the model card, evaluation, training data, limitations and declarations, and states there that the training data is not released, that the model is not distilled from a commercial decision API, and that AI assistance was used.
Qwen/Qwen3.5-4B: the card lists both Qwen/Qwen3.5-4B and berkay123001/turnike-jev-4b as base models with a merge relation, matching the bf16 merge karar.py reproduces at load time from the adapter.
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
The berkay123001/turnike-jev-9b LoRA adapter merged into Qwen/Qwen3.5-9B in bf16, so the bilingual Turkish and English decision model can be served without PEFT through the karar.py Karar interface.
Jev Models & Reimplementations#community#open-models#local
Community
A 9B bilingual Turkish and English Jev-class decision model from Berkay Hasret: a 173 MB LoRA adapter on Qwen/Qwen3.5-9B that answers up to 16 typed questions about one text in a single forward pass and reports 0.783 macro accuracy on full JevBench.
Jev Models & Reimplementations#community#open-models#local
Community
A bilingual Turkish and English 4B Jev-class decision model from Berkay Hasret: a roughly 130 MB LoRA adapter on Qwen/Qwen3.5-4B that answers up to 16 typed questions per text in one pass and reports 0.774 JevBench macro accuracy.
Jev Models & Reimplementations#community#open-models#local
Community
A local yes/no decision model that speaks Jev's wire format on a laptop CPU, refusing Choice and Score until those question types ship.
Jev Models & Reimplementations#community#python#open-models
Community
feder-cr/jevC++
feder-cr/jevC++A Python server in front of llama-server that reads one-token label probabilities and exposes them as POST /v1/systemone.
Jev Models & Reimplementations#community#python#open-models
Community
khimaros/verdictPython
khimaros/verdictPythonA LoRA plus 24-slot decision head over Qwen3.8-27B from Ming Xu's train-jev project: it serves TypeSafe's System One API with 0.851 Kev transfer-v4 accuracy and ECE 0.053, and needs an 80 GB GPU.
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.