veryfront 0.1.954 → 0.1.955
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/esm/deno.js
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"model-comparison.d.ts","sourceRoot":"","sources":["../../../src/src/eval/model-comparison.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAGV,mBAAmB,EACnB,0BAA0B,EAE1B,UAAU,EACX,MAAM,YAAY,CAAC;
|
|
1
|
+
{"version":3,"file":"model-comparison.d.ts","sourceRoot":"","sources":["../../../src/src/eval/model-comparison.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAGV,mBAAmB,EACnB,0BAA0B,EAE1B,UAAU,EACX,MAAM,YAAY,CAAC;AA+NpB,oFAAoF;AACpF,wBAAgB,uBAAuB,CACrC,OAAO,EAAE,UAAU,EAAE,EACrB,OAAO,EAAE,0BAA0B,GAClC,mBAAmB,CAqBrB;AAUD,gFAAgF;AAChF,wBAAgB,iCAAiC,CAAC,UAAU,EAAE,mBAAmB,GAAG,MAAM,CAgCzF"}
|
|
@@ -34,6 +34,11 @@ function meetsImprovement(value, threshold) {
|
|
|
34
34
|
function formatPct(value) {
|
|
35
35
|
return `${Math.round(value * 100)}%`;
|
|
36
36
|
}
|
|
37
|
+
function hasBlockingRegression(comparison) {
|
|
38
|
+
return comparison.passRateDelta < 0 || comparison.failedDelta > 0 ||
|
|
39
|
+
comparison.newFailedExamples.length > 0 ||
|
|
40
|
+
comparison.metricDeltas.some((metric) => metric.severity !== "soft" && metric.regressed);
|
|
41
|
+
}
|
|
37
42
|
function modelSummary(report, baselineModel) {
|
|
38
43
|
const model = reportModel(report);
|
|
39
44
|
const groundedness = groundednessScore(report);
|
|
@@ -64,7 +69,7 @@ function createCandidateDecision(input) {
|
|
|
64
69
|
const candidateSummary = modelSummary(input.candidate, baselineModel);
|
|
65
70
|
const baselineComparison = compareEvalReports(input.candidate, input.baseline);
|
|
66
71
|
const reasons = [];
|
|
67
|
-
if (input.candidate.summary.failed > 0 || baselineComparison
|
|
72
|
+
if (input.candidate.summary.failed > 0 || hasBlockingRegression(baselineComparison) ||
|
|
68
73
|
baselineComparison.newFailedExamples.length > 0) {
|
|
69
74
|
reasons.push("candidate has quality regressions");
|
|
70
75
|
if (baselineComparison.newFailedExamples.length > 0) {
|