Jev directory
CommunityJev Models & Reimplementations
Humor-Decider-2B
A LoRA fine-tune of Strands Decider 2B that rates jokes on seven comedy traits, answering each as a typed yes/no question with a probability; an unofficial fan project that cannot write jokes and is not endorsed by Conan O'Brien.
- Reading time
- 2 min
- Published by
- Community
- Author
- anandamri
- Last verified
- Oct 2026
- Languages
- Shell
- Capabilities
- Open sourceSelf-hostedRuns locally
Highlights
- LoRA fine-tune of Strands Decider 2B scoring seven comedy traits, from self_deprecating to silly_button.
- Trained with Strands' v21b recipe: about 120k general rows plus 24,366 joke rows (8,122 unique, repeated three times).
- Self-reported trait probes: mean AUROC 0.998 and Brier 0.027, against 0.946 and 0.156 for stock Strands Decider v21.
- Training ran 4,476 optimizer steps on one RTX PRO 6000 Blackwell for about 1 h 45 min; no JevBench run was made.
- Serves through strands-decider's /v1/systemone endpoint; only the seven yes/no questions were trained.
Quickstart
bash
pip install strands-decider
strands-decider serve <this folder> --port 8000
curl -s localhost:8000/v1/systemone -H 'content-type: application/json' -d '{"state": "My cooking is so bad the smoke alarm sings along.", "questions": {"self_deprecating": {"type": "noul", "instructions": "Does the speaker make fun of themselves?", "criteria": {"true": "the speaker mocks their own looks, height, age, career, or competence", "false": "the speaker is not the target of the humor"}}}}'Watch out
Card reviewed 2026-10-17; the 0.998 mean AUROC and 0.027 Brier are self-reported from single runs, and the card itself warns that the trait probes share a generator and prompt templates with the training jokes, so they are in-distribution and cannot prove transfer. No JevBench run was made, the eight-joke cross-generator spot check is too small, and the project is unofficial and unaffiliated with Conan O'Brien.
Questions about this entry
Short answers drawn from the linked project or page.
A LoRA fine-tune of Strands Decider 2B that rates a joke on seven comedy traits. Each trait is a typed yes/no question answered with a probability; the model cannot write jokes and was built as an unofficial fan project.
It is a decision model in the same shape: you send a state (the joke text) plus typed questions and receive probabilities, not generated prose. It serves the /v1/systemone endpoint through strands-decider, and only the seven trained yes/no questions are specialised.
self_deprecating, absurd_escalation, mock_outrage, undercuts_itself, silly_tangent, overcommits and silly_button; silly_button, for example, asks whether the joke ends with a separate goofy closing line after its main punchline.
Install the strands-decider package, serve the repository folder on a port, and send a state with questions to the /v1/systemone endpoint. The weights are Apache-2.0 and run locally from the published adapter.
Mean AUROC 0.998 and mean Brier 0.027 on 700 probe jokes, against 0.946 and 0.156 for stock Strands Decider v21. The card warns these are in-distribution: the probes share a generator and prompt templates with the training jokes, so they cannot prove transfer.
Strands' v21b recipe on Qwen3.5-2B-Base: about 120k general public-dataset rows plus 24,366 joke rows (8,122 unique, repeated three times) generated by Gemma-4-31B-IT and re-checked; no comedian's jokes or transcripts were used.
Only the self_deprecating trait was checked externally, zero-shot on the Humour Styles dataset of 1,463 texts with 265 self-deprecating labels. Most labels are website tags and only 200 were human-annotated at fair agreement (Fleiss' kappa about 0.27 to 0.28).
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 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
chaoliangUNSW's v1 Jev-Style decision model in GGUF from 1.3 GB: calibrated choice, bool and score answers on Qwen3.5-2B-Base, 82.3% on five held-out task families with ECE 0.017.
Jev Models & Reimplementations#community#open-models#local
Community
An Apple Silicon MLX bf16 build of the v1 Jev-Style decision model: 77 ms decisions on an M1 Max with calibrated choice, bool and score probabilities, Apache-2.0 weights.
Jev Models & Reimplementations#community#open-models#local
Community
A LoRA adapter for Qwen3-VL-4B-Instruct that scores supplied visual answer candidates instead of writing prose: macro accuracy over four choice benchmarks rises from 0.706 to 0.761, Apache-2.0.
Jev Models & Reimplementations#community#open-models#local
Community
Open recipe for calibrated decision models: an LLM teacher writes training data from a one-sentence task description, a LoRA plus decision head on Qwen3-4B-Base answers Choice, Score and Noul questions with calibrated probabilities, evaluated next to Jev on identical items.
Jev Models & Reimplementations#community#calibration#training
Community
scienthoon/lucePython
scienthoon/lucePythonFrom 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.