CommunityJev Demos & Experiments
KaLM-Jev Small
A Gradio playground for KaLM-Jev Small structured judgments: choice, score, noul, and mixed JSON questions return probability tables and full responses without generating answer text on shared ZeroGPU.
- Format
- Playground
- Published by
- Community
- Author
- Yuki131
- Last verified
- Oct 2026
Highlights
- Answers choice, score, noul, and mixed JSON questions, returning a probability table and the full structured response with no generated answer text.
- Backed by KaLM-Reranker-V1-Small-R2; the kalm_jev engine is pinned to GitHub commit d8b1385d and checkpoint revision d1c8fdf9.
- Public limits: 8 questions, 32 point-wise pairs, 32 KiB of request JSON, 512 state tokens, 1,024 document tokens, 2,048 decoder-template tokens.
- Overlong input is rejected rather than truncated; the README states probabilities are uncalibrated and confidence is not correctness.
- The API panel ships a Python gradio_client snippet for /evaluate_json, including a noul human-escalation example.
Watch out
Probabilities are uncalibrated: the README says choice and score confidence measures distribution concentration, not correctness, and that business thresholds belong in your own validation data. Weights are not stored in the Space — they are fetched from a pinned checkpoint revision — and hosted inputs are processed on Hugging Face infrastructure, so use the standalone engine for local processing.
Questions about this entry
Short answers drawn from the linked project or page.
Choice among candidates, score against ordered levels returning a zero-based expected index, noul for an independent condition on a 0-1 scale, and mixed JSON combining question types in one request.
No. Every run returns an easy-to-read table and the full structured response, and the card states that no answer text is generated.
No. The README says probabilities are uncalibrated and that choice and score confidence measures distribution concentration, not correctness, so thresholds should come from your own validation data.
At most 8 questions, 32 point-wise pairs, 32 KiB of request JSON, 512 state tokens, 1,024 document tokens, and 2,048 full decoder-template tokens; overlong input is rejected, never truncated.
Through the Use via API panel: a Python gradio_client example calls the /evaluate_json endpoint with a KaLM-Jev JSON request and returns the table, response, and timing.
KaLM-Reranker-V1-Small-R2 from checkpoint revision d1c8fdf9, with the unmodified engine from KaLM-Embedding/KaLM-Jev pinned at GitHub commit d8b1385d; weights are not stored in the Space.
No. The README says KaLM-Jev is an independent project not affiliated with TypeSafe, and it links the GitHub repository for the source and local quickstart.
More Jev resources like this
An interactive ZeroGPU demo of AutoTrust's JEV-9B: System 1 answers noul, choice, and score questions in one prefill pass with calibrated probabilities, while an Escalate tab hands low-confidence cases to the untouched Qwen3.5-9B base.
Jev Demos & ExperimentsPlayground#community#playground#demo
Community
A three-lane demo of pngwn's System One 4B scorer: one prefill plus a batched forward for every option branch, a naive per-option equivalence check, and a Qwen3.5-4B instruct baseline generating JSON.
Jev Demos & ExperimentsPlayground#community#playground#demo
Community
A Gradio demo of Maincode's 26B MATILDA-jev-v1 one-pass decision model: feed a text or JSON state, optionally an image, and a choice, noul, or score question to get calibrated probabilities over your options.
Jev Demos & ExperimentsPlayground#community#playground#demo
Community
A ZeroGPU demo of AutoTrust's JEV-27B-VL: System 1 returns calibrated noul, choice, and score probabilities from one prefill pass over text and images, with System 2 on the untouched Qwen3.8-27B for escalation.
Jev Demos & ExperimentsPlayground#community#playground#demo
Community
A Gradio playground for Jev-Omni: give a situation plus a question with options and get one probability per option from a single forward pass, with no generated text, from a Gemma 4 12B IT fine-tune.
Jev Demos & ExperimentsPlayground#community#playground#demo
Community
AutoTrust's live game-arena demo: watch JEV-27B play Mario, a 3D Rubik's Cube, and Tetris while a side panel reports action probabilities and per-decision inference time, alongside a decision lab and a ViZDoom arena.
Jev Demos & ExperimentsPlayground#community#playground#demo
Community
From the Jev guides
Original write-ups on what this entry is for.
A game tick is a state plus a legal action list. How Clash Royale, Minecraft, and the earlier Mario, Pokémon, StarCraft, and drone loops ask Jev one step at a time.
Jev Demos & Experiments#gaming#agents#control
Read article
From the Jev community
Posts from builders shipping with Jev right now.