Jev directory
CommunityJev Models & Reimplementations
JEV-CPU — huggingface.co
A CPU port of SemIf with a web UI: JEV-CPU loads Qwen3-0.6B in float32 and reads a decision from its option logits in about a second, across eight domains from support to incident severity, with no GPU.
- Reading time
- 2 min
- Published by
- Community
- Author
- Meanblock
- Last verified
- Oct 2026
This is one of several Jev listings for the same project. The canonical entry is JEV-CPU — leesk212.
- Languages
- Python
- Capabilities
- Open sourceRuns locally
Highlights
- CPU port of SemIf: loads Qwen/Qwen3-0.6B in float32 (~2.4 GB) on a laptop and needs no GPU.
- Reads the decision from option logits in about a second: one forward pass, no text generated, no decoding loop.
- One engine across eight domains, from support and content moderation to incident severity, email intent and loan/credit risk.
- Options are pinned to single tokens that round-trip, then only those slots are softmaxed; the winner is the argmax.
- MIT licensed; the project states it is independent and not affiliated with SemIf's author, TypeSafe or Jev.
Quickstart
python
logits = model(**inputs, use_cache=False).logits[:, -1, :] # (vocab,)
selected = logits[slot_ids] # e.g. logits at tokens A, B, C
probs = softmax(selected) # conditional distribution over the options
winner = options[argmax(probs)]Watch out
Page copy reviewed 2026-10-15 from the Hugging Face model card, which was truncated mid-section. The ~1 second latency and the CPU port design are the card's own description; the digest carries no independent benchmark, and the project is a thin adaptation of TheoLeeCJ/SemIf.
Questions about this entry
Short answers drawn from the linked project or page.
A CPU port of SemIf with a web UI, published by Meanblock: it runs the same decision engine on a laptop CPU with Qwen3-0.6B instead of the upstream CUDA GPU and 4B model, and reads a decision from option logits in about a second.
The request becomes a letter-choice prompt, each option is pinned to a single token that round-trips, and one forward pass takes the logits at the final position. Only the option-letter slots are softmaxed, and the winner is the argmax over the supplied options.
A laptop CPU and no GPU: it loads Qwen/Qwen3-0.6B in float32, about 2.4 GB, and reuses SemIf's original, unmodified scoring code. The upstream project targets a CUDA GPU holding a 4B BF16 model.
The live proof of concept runs one CPU engine across eight domains: support, content moderation, code-review triage, incident severity, email intent, compliance, loan/credit risk and support prioritization.
No. The card states JEV-CPU is an independent, thin CPU adaptation of TheoLeeCJ/SemIf, not affiliated with or endorsed by SemIf's author, TypeSafe or Jev. It is MIT licensed.
The card describes a decision read from Qwen3-0.6B's option logits in about one second on CPU, with no text generated. That is the project's own figure; the digest contains no independent benchmark.
The card has a section on input token limits, but the fetched text was truncated before its contents, so the exact budget is not in the digest. Check the model card for the current limit.
More Jev resources like this
HAL-X AI's 322M non-autoregressive decision model returns calibrated answers in one forward pass of about 10 ms, with native number, excerpt and citation outputs beyond Jev's choice, noul and score types. Apache-2.0, post-trained in 18 languages.
Jev Models & Reimplementations#community#catalog-import#madewithjev
Community
An open-source, Jev-compatible System One model: Dex answers choice, score and yes/no questions over text in a single forward pass from an Ettin-1B encoder, and speaks the same /v1/systemone shape as Jev.
Jev Models & Reimplementations#community#catalog-import#madewithjev
Community
canadaornaw/dexPython
canadaornaw/dexPythonGestalt Labs' Jeff is a LoRA adapter on Qwen3-4B-Instruct-2507 that answers Jev-style choice, noul and score questions with labels and probabilities instead of prose, runs locally, and ships a train-your-own guide; on 9,730 fact-checking examples it trails Jev 1.13.0 on accuracy (0.8183 vs 0.8283).
Jev Models & Reimplementations#community#catalog-import#madewithjev
Community
Zefan Cai's LoRA adapter and scalar decision head for Qwen3.5-9B score caller-supplied candidates for choice, yes/no and score answers without generating text, served over a local /v1/systemone endpoint.
Jev Models & Reimplementations#community#catalog-import#madewithjev
Community
cklxx fine-tuned Laya into the decision head of browser-use/jev-ultrafast: one bidirectional encoder pass per step picks the browser operation and target element, and the server answers Jev's /v1/systemone requests identically.
Jev Models & Reimplementations#community#catalog-import#madewithjev
Community
A local server that reads next-token probabilities from a fine-tuned Spark model and returns Choice, Score, and Noul answers with zero generated tokens.
Jev Models & Reimplementations#community#python#open-models
Community
Rizzo-AI-Academy/rizzo-flowPython
Rizzo-AI-Academy/rizzo-flowPythonFrom 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.