Jev glossary
Core ConceptsOne forward passParallel scoring
Non-autoregressive decode
Scoring every possible answer in a single forward pass instead of generating tokens one at a time.
- Category
- Core Concepts
- Also known as
- One forward pass, Parallel scoring
- Reading time
- 1 min
- Related terms
- 4
- Directory entries
- 0
- Docs
- —
- Added
- 2026-10-11
Definition
An autoregressive model emits one token per step, each conditioned on the last. A decision model bounds its output, so it can score the whole answer space at once — which is why Jev's latency is measured in milliseconds and its output tokens are free.
The same property runs through the open ecosystem: von, Julia, and other non-autoregressive checkpoints advertise sub-100 ms local decisions, and Jev's parallel sampler is the mechanism TypeSafe cites for its speed.
Tagscore-conceptsmodels
Related terms
Definitions that connect to this one.
Logit readoutReading the next-token probability of each declared label, then renormalising, so no answer text is generated.PatternsGeneration limitationJev is not trained to generate text; forcing it by chaining choices is slow and unreliable, so extraction becomes a Choice instead.Evaluation & TrainingDistillationTraining a smaller model to reproduce a larger model's output distribution, including its probabilities rather than labels alone.Evaluation & TrainingMachine Native IntelligenceTypeSafe's name for AI with software-like properties: structure, reliability, observability, testability, speed, consistency, and low cost.Core Concepts
From the Jev directory
Directory entries that go deeper on this term.
No linked entries yet — browse the full directory instead.
From the Jev AI model community
Posts from builders shipping with Jev right now.