SemIf compares two ways to ask a local model to choose

SemIf runs a local decision experiment in the browser. One path reads a model’s scores for a supplied set of options; the other asks the same model to generate those probabilities as JSON. It times both paths on your device, making the extra work of text generation visible without requiring a hosted decision-model service.

SemIf

Why it matters

The demo lets developers examine decision latency without confusing structured output with reliable confidence.

More context

An undated discovery, formerly called OpenJev and explicitly independent of TypeSafe. Try a small labeled example after loading a model; the default download is 1.56 GB, with smaller and larger options. The page says inputs remain local.

Keep in mind

An experiment, not a Jev replacement or equivalence claim. Option scores are conditional and uncalibrated; quantization and hardware affect results. Browser measurements were not run by 1min.

Original read