Jev directory
CommunityJev Models & Reimplementations
WorkMagic Jev Churn Distill Qwen3-8B v1
WorkMagic's self-distilled churn decision model for Slack customer conversations: a LoRA on Qwen/Qwen3-8B trained with no human labels that returns churn-likelihood option probabilities through the AnyJev Tacit interface.
- Reading time
- 2 min
- Published by
- Community
- Author
- WorkMagic
- Last verified
- Oct 2026
Used inJev for Sales & Marketing
- Languages
- Python
- Capabilities
- Open sourceSelf-hostedRuns locally
Highlights
- AnyJev one-forward-pass decision model distilled from the teacher Decider(level=L0) into a LoRA on Qwen/Qwen3-8B.
- Trained with no human labels: the loss is computed only on answer tokens in the AnyJev readout prompt format.
- tacit.decide(state, question, options) returns an answer plus probabilities; the documented usage passes adaptive=False.
- Built for Slack-channel churn decisions over low, medium and high options; front matter lists accuracy, F1 and expected calibration error.
- Repository declares Apache-2.0, English, version v1, with workmagicio/slack-churn listed as the dataset.
Quickstart
python
from anyjev import Tacit
tacit = Tacit.from_pretrained("workmagicio/workmagic-jev-churn-distill-qwen3-8b-v1", adaptive=False)
d = tacit.decide(
state="...slack conversation...",
question="...how likely is the customer to churn...",
options=["low", "medium", "high"],
)
print(d["answer"], d["probs"])Watch out
Card reviewed 2026-10-17; the card is written mostly in Chinese, reports no measured results, and lists accuracy, F1 and expected calibration error only as front-matter metrics, so its quality rests on the self-distillation setup. Because training used no human labels, the behaviour it learned comes from the L0 teacher decider rather than human gold.
Questions about this entry
Short answers drawn from the linked project or page.
A single-forward-pass decision model for Slack-channel customer churn from WorkMagic: it takes a Slack conversation and a churn question with options like low, medium and high, and returns the answer plus probabilities, distilled from the AnyJev teacher Decider(level=L0).
Through AnyJev self-distillation: the LoRA on Qwen/Qwen3-8B is trained in the AnyJev readout prompt format with loss computed only on the answer tokens, using the L0 teacher decider's answers instead of human-labelled data.
Load it with Tacit.from_pretrained("workmagicio/workmagic-jev-churn-distill-qwen3-8b-v1", adaptive=False) and call tacit.decide(state, question, options); the result carries an answer and probabilities. A revision="v1" tag can pin the version.
The base is Qwen/Qwen3-8B, and the repository declares Apache-2.0 with English as the only language and v1 as the version.
No. The card reports no measured results; its front matter lists accuracy, F1 and expected calibration error as declared metrics and names workmagicio/slack-churn as the dataset, so no accuracy figure can be checked from the repository.
The library_name recorded on the card. AnyJev serves the readout prompt format and the distill-and-serve workflow behind this checkpoint: a teacher decider produces the targets and the student answers with option probabilities in a single forward pass.
Like Jev, it answers a state plus a question with probabilities over supplied options instead of generated prose, but it is an independent WorkMagic distillation for one domain, churn, rather than a general-purpose System One model.
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
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-flowPythonA local /v1/systemone server that reads label probabilities from a GGUF chat model, with prompt repetition and a confidence shrink toward uniform.
Jev Models & Reimplementations#community#python#open-models
Community
Mushroom-Systems/lichenPython
Mushroom-Systems/lichenPythonA 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
A 24 GB-GPU alternative from the train-jev project: Qwen3.5-4B plus a rank-16 LoRA and 24-slot decision head scoring 0.647 PK accuracy, 40.4 Decision Index and 43 ms p50 latency, ahead of Kev-4B at the same size.
Jev Models & Reimplementations#community#open-models#local
Community
A small distilled sibling of Perplexity's pplx-decider-v1.2: a Qwen3.5-4B decider that reads a state plus typed choice, noul and score questions in one pass, keeping most of the 27B model's Decision Index.
Jev Models & Reimplementations#community#open-models#multimodal
Community
A Laya fine-tune for coding decisions: roughly 9k rows from SWE-Gym teach keep/drop file relevance, priority scoring and injection detection, reporting 0.833 accuracy on 1,150 held-out rows at 126 ms p50.
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.