CommunityJev Models & Reimplementations
EmbeddingGemma 2 Typed Decisions
An EmbeddingGemma 2 fine-tune that answers choice, score, and noul questions from a state and a plain-data schema, with a 0.5M-parameter schema-conditioned scorer over 768-d embeddings.
- Published by
- Community
- Author
- mrm8488
- Last verified
- Oct 2026
Used inJev for Software & DevTools
- Languages
- Python
- Capabilities
- Open sourceSelf-hostedRuns locally
Highlights
- The head scores the text of each option against the state and question, so options can be renamed or added without retraining.
- Each answer is a TypedAnswer carrying kind, label, typed value, confidence, per-option probabilities, and expected_level for score questions.
- The state is rendered as readable key: value lines instead of raw JSON, which the card says embedding models read much better.
- The scorer is a gated correction on a zero-shot embedding-similarity prior: logit = prior + sigmoid(gate) * (s * cos(field, option) + residual).
- Further fine-tuning uses LocalLLaMA/typed-decisions rows of {workflow, state, questions, gold} through model.collate.
Watch out
The head was trained on English schemas only, so the card expects softer probabilities in other languages. Never run it in float16: EmbeddingGemma 2 overflows and returns NaNs without raising an error; use bfloat16 on Ampere or newer, or float32. Answers to questions far from the training data are less reliable.
Questions about this entry
Short answers drawn from the linked project or page.
Three: choice takes a dict of option keys to descriptions and returns a string, score takes an ordered list of level descriptions and returns a probability-weighted fractional level, and noul is a yes/no question that returns a bool with a probability.
bfloat16 on Ampere or newer, or float32. The card warns never to use float16, because EmbeddingGemma 2 overflows and returns NaNs without raising an error.
Yes. The head scores the text of each option against the state and question, so the schema is plain data: you can rename options, change descriptions, or add new questions. Answers to questions far from the training data are less reliable.
The encoder is multilingual, and the card shows a Spanish example, but the head was trained on English schemas only, so expect softer probabilities in other languages.
Load LocalLLaMA/typed-decisions with the workflow 'all', pass model.collate as the Trainer's data collator, and train with remove_unused_columns=False. One learning rate fine-tunes encoder and head together; the two-rate recipe used originally needs a custom optimizer.
A TypedAnswer per question: kind (choice, score, or noul), the winning option's label and typed value, confidence, the full probability dict, and expected_level for score questions. answer.to_dict() gives JSON.
model.encode(texts) returns normalized sentence embeddings exactly as in training, model.collate(rows) turns {workflow, state, questions, gold} rows into a training batch, and a forward pass returns GemmaDecisionOutput with loss, logits, probabilities, and text_embeds.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
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 NewsHacker News: A one-token logprob wrapper that also classifies webcam frames.
allanrbo151 pointsA single function Jev-like wrapper for LLMs, including vision modelsRead the thread on Hacker NewsRelated terms
Glossary definitions related to this entry.
More Jev resources like this
AutoTrust's combined System 1 and System 2 bundle on Gemma-4-26B-A4B-it: Decision Index 0.2.1 of 58.05, a 24-slot head with noul, choice and score, and the base chat model left bit-identical.
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
alibiserikbay's Apache-2.0 Jev alternative on Qwen3.5-4B with a merged rank-16 LoRA: one forward pass returns a probability per option, 0.784 on JevBench's public hard tier at about 13 ms per decision.
Jev Models & Reimplementations#community#open-models#local
Community
The 9B JevK5: the same v0.3 data on Qwen3.5-9B, retrained for 1.5 epochs by a pre-registered sweep, reaching index proxy 0.794 and a 0.859 hand-written hard set, Apache-2.0.
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
From the Jev guides
Original write-ups on what this entry is for.
Training recipes, looped architectures, and language ports: what the open side of the typed-decision class shipped this month, and how to pick by runtime and calibration instead of the leaderboard.
Jev Models & Reimplementations#open-models#training#local
Read article
Taking a decision checkpoint out of Python: what the GGUF and browser WebGPU exports preserve, what they change, and the parity evidence the model cards publish.
Jev Tools & CLIs#open-models#local#onnx
Read article
Interface, size, license, claims — in that order. A selection guide across the system-one catalog, from a 706K form scorer to 27B Jev students.
Jev Practices & Patterns#open-models#evaluation#benchmarks
Read article
From the Jev community
Posts from builders shipping with Jev right now.