shadcn.io is not affiliated with official shadcn/ui
Evaluation Results AI Block for Vue
A matrix of ten test cases by three models with Pass or Fail in words and glyphs, scores as solid tint weights, a category toggle, an Only where models disagree switch, a detail band for the cell you press, and a copy of every failure.
Installation
Overview
Compare how several models do on the same test cases, with this test bench for Vue with Ark UI. Evalmill, a model test bench, lays out ten cases (an id, a short prompt and a category) against Orvane 2, Orvane 2 Fast and Brellis Large. Each cell says Pass or Fail in a word and a glyph with its score, and the score sets how heavy a solid tint the cell has instead of a red or green. Model heads carry a tally such as 9 of 10 pass and a mean. A Toggle Group filters by category and a Switch keeps only the cases where models disagree. Pressing a cell shows the prompt, the expected answer and what the model said, and the Shine Button Copy failures copies every failing case. For eval pipelines, model picks and release checks.
FAQ
What stays in the page, and how do I make it real?
The ten cases, their scores and the wrong answers are typed in the file. Pass or Fail (a score of 70 or more passes), the tallies, the means and the disagreement filter are worked out from the scores. Nothing is run. To wire it for real, return the score and the model's answer for each case and model.
Why are the cells not red and green?
Colour alone is not enough, so the word and the glyph say Pass or Fail, and the shade only shows how high the score is: three solid weights of the foreground. Failing scores get the lightest one, so the order of the shades is the order of the scores.
What does Only where models disagree show?
The cases where at least one model passes and at least one fails. The tallies and means in the heads are worked out from the cases that are shown, so they follow the filter.
Does the page move when I filter?
No. The matrix sits in a fixed-height scroll region with a sticky head, so a filter changes the rows inside it and nothing else. If no case matches both filters, the region says so.