Jev directory
CommunityJev Models & Reimplementations
SocialJev 0.6B Preview
A 0.6B classification foundation model from the SocialJev project on Qwen3-0.6B: define a task and its answer options, and it classifies social-science documents through a typed interface that returns a choice plus probabilities.
- Reading time
- 2 min
- Published by
- Community
- Author
- SocialJev
- Last verified
- Oct 2026
- Languages
- Python
- Capabilities
- Self-hostedRuns locally
Highlights
- A family of classification foundation models for social science: define a task and answer options, then classify documents through typed queries.
- Results return a choice and probabilities in nested lists that follow document order, then each document's question order.
- Two interfaces: AutoModel with trust_remote_code=True and revision v0.1a, or the social-jev package with vLLM integrations.
- Built on Qwen/Qwen3-0.6B; the preview card publishes no benchmark numbers and states no licence for the weights.
Quickstart
python
from transformers import AutoModel
model = AutoModel.from_pretrained(
"social-jev/social-jev-0.6B-preview",
trust_remote_code=True,
)
query = model.Query(text="Which policy area does this document concern?", options=["Transport", "Education", "Healthcare"])
results = model.query([model.Request(state="The council expanded the bus network.", queries=[query])])
print(results[0][0].model_dump())Watch out
Card reviewed 2026-10-17; the preview card publishes no evaluation numbers or licence, so quality and terms cannot be checked, and loading uses trust_remote_code=True, which executes the repository's custom model code on your machine.
Questions about this entry
Short answers drawn from the linked project or page.
A classification foundation model family for social science. You define a task and its answer options as a typed query, and the model classifies documents, returning a choice plus probabilities for each query in one forward pass over the state.
It shares the typed-question shape: supply a state (the document) plus answer options, and get back labels with probabilities instead of generated text. The loaded model exposes Request, Query, Example and Result types directly.
Install torch, transformers and pydantic, then call AutoModel.from_pretrained with trust_remote_code=True, using revision v0.1a to select the tagged version. No SocialJev package installation is required for this interface.
It provides a shared typed interface and inference integrations, including offline vLLM and asynchronous managed-server support through social-jev[vllm]. Install it with uv pip install social-jev, and call model.stop() when you are done.
query() returns list[list[Result]]: the outer list follows document order and each inner list follows that document's question order, with each result carrying a choice and probabilities. A single request returns the same nested structure.
This preview is built on Qwen/Qwen3-0.6B and published under the social-jev org; the card presents SocialJev as a family, so other sizes may follow. The repository ships custom code, so trust_remote_code=True is required.
The card states no licence and no benchmark results, so neither usage terms nor quality can be verified from the repository. Treat the preview as a research release and check the project website or GitHub repository for updates before relying on it.
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
Fastino's 287M multilingual classification model in the GLiNER2.5 family: pass any label set at call time and get labels back in a single forward pass, with no prompt template and no generated tokens. It averages 56.7% exact-match on the Fast Decisions suite.
Jev Models & Reimplementations#community#open-models#local
Community
Fastino's 1B English member of the GLiNER2.5 classification family: pass any label set at call time for intent, routing, sentiment or multi-label tags and get labels back in a single forward pass with no generated tokens. It averages 59.6% exact-match on Fast Decisions.
Jev Models & Reimplementations#community#open-models#local
Community
Fastino's 340M English classification model and the headline member of the GLiNER2.5 family: pass any label set at call time and get intent, routing, sentiment or multi-label tags back in one forward pass, at 60.2% exact-match average on the Fast Decisions suite.
Jev Models & Reimplementations#community#open-models#local
Community
akhilaaa3's merged multimodal decision classifier on Gemma 4 12B IT: probabilities for text, image, audio or video options, 87.57% on DecisionBench Medium and MMAU 63.10%, Apache-2.0.
Jev Models & Reimplementations#community#open-models#local
Community
Mithril's open, Jev-shaped typed-decision model on DeBERTa-v3-large: choice up to 255 options, score and noul from one forward pass, 0.854 in-domain with ECE 0.022, trained on public gold labels only.
Jev Models & Reimplementations#community#open-models#local
Community
TokenRhythm's 4B prefill-only decision model built on NeoHorse-1-4B: given a state and application-defined questions it returns Choice, Noul, and Score probabilities without generating text, reporting a 77.70 six-group aggregate under Apache-2.0.
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.