veryfront 0.1.964 → 0.1.965
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/cli/commands/eval/command.d.ts +1 -1
- package/esm/cli/commands/eval/command.d.ts.map +1 -1
- package/esm/cli/commands/eval/command.js +21 -5
- package/esm/deno.js +1 -1
- package/esm/src/utils/version-constant.d.ts +1 -1
- package/esm/src/utils/version-constant.js +1 -1
- package/package.json +1 -1
|
@@ -37,7 +37,7 @@ type EvalSummaryArtifact = {
|
|
|
37
37
|
baseline?: EvalReportComparison;
|
|
38
38
|
};
|
|
39
39
|
type EvalModelComparisonPolicy = Omit<EvalModelComparisonOptions, "baselineModel">;
|
|
40
|
-
export declare function createDefaultEvalReportDir(runId: string): string;
|
|
40
|
+
export declare function createDefaultEvalReportDir(runId: string, label?: string): string;
|
|
41
41
|
export declare function createEvalArtifactPaths(reportDir: string): EvalArtifactPaths;
|
|
42
42
|
export declare function createEvalModelArtifactPaths(reportDir: string, model: string): EvalModelArtifactPaths;
|
|
43
43
|
export declare function normalizeEvalCliId(id: string): string;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"command.d.ts","sourceRoot":"","sources":["../../../../src/cli/commands/eval/command.ts"],"names":[],"mappings":"AAOA,OAAO,KAAK,EAAS,aAAa,EAAE,MAAM,6BAA6B,CAAC;AACxE,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,8BAA8B,CAAC;AACpE,OAAO,KAAK,EACV,cAAc,EAGd,mBAAmB,EAEnB,0BAA0B,EAE1B,UAAU,EACV,oBAAoB,EAEpB,YAAY,EAEb,MAAM,4BAA4B,CAAC;AAmBpC,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,cAAc,CAAC;AAE7C,MAAM,WAAW,WAAY,SAAQ,QAAQ;CAAG;AAEhD,KAAK,cAAc,GAAG;IACpB,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,EAAE,MAAM,CAAC;IAChB,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,MAAM,CAAC;IACf,QAAQ,EAAE,MAAM,CAAC;IACjB,OAAO,EAAE,UAAU,CAAC,SAAS,CAAC,CAAC,SAAS,CAAC,CAAC;CAC3C,CAAC;AAEF,KAAK,iBAAiB,GAAG;IACvB,SAAS,EAAE,MAAM,CAAC;IAClB,OAAO,EAAE,MAAM,CAAC;IAChB,OAAO,EAAE,MAAM,CAAC;IAChB,cAAc,EAAE,MAAM,CAAC;CACxB,CAAC;AAEF,KAAK,sBAAsB,GAAG,iBAAiB,GAAG;IAChD,KAAK,EAAE,MAAM,CAAC;CACf,CAAC;AASF,KAAK,mBAAmB,GAAG;IACzB,IAAI,EAAE,cAAc,CAAC;IACrB,KAAK,EAAE,MAAM,CAAC;IACd,YAAY,EAAE,MAAM,CAAC;IACrB,UAAU,EAAE,UAAU,CAAC,YAAY,CAAC,CAAC;IACrC,MAAM,EAAE,MAAM,CAAC;IACf,SAAS,EAAE,MAAM,CAAC;IAClB,OAAO,EAAE,MAAM,CAAC;IAChB,OAAO,EAAE,UAAU,CAAC,SAAS,CAAC,CAAC;IAC/B,QAAQ,CAAC,EAAE,UAAU,CAAC,UAAU,CAAC,CAAC;IAClC,OAAO,CAAC,EAAE,UAAU,CAAC,SAAS,CAAC,CAAC;IAChC,QAAQ,CAAC,EAAE,oBAAoB,CAAC;CACjC,CAAC;AAEF,KAAK,yBAAyB,GAAG,IAAI,CAAC,0BAA0B,EAAE,eAAe,CAAC,CAAC;
|
|
1
|
+
{"version":3,"file":"command.d.ts","sourceRoot":"","sources":["../../../../src/cli/commands/eval/command.ts"],"names":[],"mappings":"AAOA,OAAO,KAAK,EAAS,aAAa,EAAE,MAAM,6BAA6B,CAAC;AACxE,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,8BAA8B,CAAC;AACpE,OAAO,KAAK,EACV,cAAc,EAGd,mBAAmB,EAEnB,0BAA0B,EAE1B,UAAU,EACV,oBAAoB,EAEpB,YAAY,EAEb,MAAM,4BAA4B,CAAC;AAmBpC,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,cAAc,CAAC;AAE7C,MAAM,WAAW,WAAY,SAAQ,QAAQ;CAAG;AAEhD,KAAK,cAAc,GAAG;IACpB,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,EAAE,MAAM,CAAC;IAChB,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,MAAM,CAAC;IACf,QAAQ,EAAE,MAAM,CAAC;IACjB,OAAO,EAAE,UAAU,CAAC,SAAS,CAAC,CAAC,SAAS,CAAC,CAAC;CAC3C,CAAC;AAEF,KAAK,iBAAiB,GAAG;IACvB,SAAS,EAAE,MAAM,CAAC;IAClB,OAAO,EAAE,MAAM,CAAC;IAChB,OAAO,EAAE,MAAM,CAAC;IAChB,cAAc,EAAE,MAAM,CAAC;CACxB,CAAC;AAEF,KAAK,sBAAsB,GAAG,iBAAiB,GAAG;IAChD,KAAK,EAAE,MAAM,CAAC;CACf,CAAC;AASF,KAAK,mBAAmB,GAAG;IACzB,IAAI,EAAE,cAAc,CAAC;IACrB,KAAK,EAAE,MAAM,CAAC;IACd,YAAY,EAAE,MAAM,CAAC;IACrB,UAAU,EAAE,UAAU,CAAC,YAAY,CAAC,CAAC;IACrC,MAAM,EAAE,MAAM,CAAC;IACf,SAAS,EAAE,MAAM,CAAC;IAClB,OAAO,EAAE,MAAM,CAAC;IAChB,OAAO,EAAE,UAAU,CAAC,SAAS,CAAC,CAAC;IAC/B,QAAQ,CAAC,EAAE,UAAU,CAAC,UAAU,CAAC,CAAC;IAClC,OAAO,CAAC,EAAE,UAAU,CAAC,SAAS,CAAC,CAAC;IAChC,QAAQ,CAAC,EAAE,oBAAoB,CAAC;CACjC,CAAC;AAEF,KAAK,yBAAyB,GAAG,IAAI,CAAC,0BAA0B,EAAE,eAAe,CAAC,CAAC;AA6CnF,wBAAgB,0BAA0B,CAAC,KAAK,EAAE,MAAM,EAAE,KAAK,CAAC,EAAE,MAAM,GAAG,MAAM,CAIhF;AAED,wBAAgB,uBAAuB,CAAC,SAAS,EAAE,MAAM,GAAG,iBAAiB,CAO5E;AAMD,wBAAgB,4BAA4B,CAC1C,SAAS,EAAE,MAAM,EACjB,KAAK,EAAE,MAAM,GACZ,sBAAsB,CASxB;AAiDD,wBAAgB,kBAAkB,CAAC,EAAE,EAAE,MAAM,GAAG,MAAM,CAErD;AAQD,wBAAgB,gBAAgB,CAAC,KAAK,EAAE,cAAc,EAAE,EAAE,EAAE,EAAE,MAAM,GAAG,cAAc,GAAG,SAAS,CAKhG;AAED,wBAAgB,0BAA0B,CAAC,KAAK,EAAE,OAAO,GAAG,MAAM,CASjE;AAED,wBAAgB,qBAAqB,CAAC,MAAM,EAAE,UAAU,GAAG,cAAc,CAWxE;AAED,wBAAgB,qBAAqB,CACnC,MAAM,EAAE,UAAU,EAClB,QAAQ,CAAC,EAAE,oBAAoB,GAC9B,mBAAmB,CAcrB;AAkGD,wBAAsB,6BAA6B,CACjD,UAAU,EAAE,MAAM,EAClB,UAAU,CAAC,EAAE,MAAM,GAClB,OAAO,CAAC,yBAAyB,CAAC,CA0CpC;AAED,wBAAsB,uCAAuC,CAC3D,UAAU,EAAE,MAAM,EAClB,OAAO,EAAE,WAAW,GACnB,OAAO,CAAC,iCAAiC,GAAG,SAAS,CAAC,CAKxD;AAED,wBAAgB,iCAAiC,CAC/C,OAAO,EAAE,UAAU,EAAE,EACrB,aAAa,EAAE,MAAM,EACrB,MAAM,GAAE,yBAA8B,GACrC,mBAAmB,CAErB;AAED,wBAAgB,iCAAiC,CAAC,OAAO,EAAE,UAAU,EAAE,GAAG,CAAC,GAAG,CAAC,CAE9E;AAED,wBAAgB,kBAAkB,CAAC,MAAM,EAAE,UAAU,GAAG,MAAM,CAG7D;AA6DD,uEAAuE;AACvE,wBAAgB,wBAAwB,CACtC,MAAM,EAAE,UAAU,EAClB,QAAQ,CAAC,EAAE,oBAAoB,GAC9B,MAAM,CAkFR;AAED,wBAAgB,cAAc,CAAC,MAAM,EAAE,UAAU,GAAG,MAAM,CA0CzD;AAQD,wBAAsB,kBAAkB,CACtC,MAAM,EAAE,UAAU,EAClB,KAAK,EAAE,iBAAiB,EACxB,QAAQ,CAAC,EAAE,oBAAoB,GAC9B,OAAO,CAAC,IAAI,CAAC,CAQf;AAMD,wBAAgB,kBAAkB,CAChC,MAAM,EAAE,UAAU,EAClB,QAAQ,CAAC,EAAE,oBAAoB,GAC9B,CAAC,GAAG,CAAC,CAEP;AAMD,KAAK,qBAAqB,GAAG,IAAI,CAAC,eAAe,EAAE,aAAa,GAAG,IAAI,CAAC,GAAG;IACzE,WAAW,CAAC,EAAE,MAAM,CAAC;CACtB,CAAC;AAQF,wBAAsB,sBAAsB,CAC1C,UAAU,EAAE,MAAM,EAClB,MAAM,EAAE,qBAAqB,GAAG,IAAI,GAAG,SAAS,sEAMjD;AAED,wBAAgB,cAAc,CAAC,QAAQ,EAAE,aAAa,MA2CrD;AAED,wBAAgB,kBAAkB,CAAC,QAAQ,EAAE,aAAa,GAAG,YAAY,EAAE,CAY1E;AA+FD,KAAK,yBAAyB,GAAG;IAC/B,aAAa,EAAE,MAAM,CAAC;IACtB,eAAe,EAAE,MAAM,EAAE,CAAC;IAC1B,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,MAAM,EAAE,MAAM,EAAE,CAAC;CAClB,CAAC;AAEF,KAAK,iCAAiC,GAAG;IACvC,MAAM,EAAE,yBAAyB,CAAC;IAClC,MAAM,EAAE,yBAAyB,CAAC;CACnC,CAAC;AAmLF,wBAAsB,WAAW,CAAC,OAAO,EAAE,WAAW,GAAG,OAAO,CAAC,IAAI,CAAC,CA8IrE"}
|
|
@@ -36,8 +36,17 @@ function stripFileProtocol(path) {
|
|
|
36
36
|
function createCliRunId(now = new Date()) {
|
|
37
37
|
return `evalrun_${now.toISOString().replace(/[-:.]/g, "").replace("T", "_").replace("Z", "")}`;
|
|
38
38
|
}
|
|
39
|
-
|
|
40
|
-
return
|
|
39
|
+
function createEvalReportDirTimestamp(runId) {
|
|
40
|
+
return runId.startsWith("evalrun_") ? runId.slice("evalrun_".length) : runId;
|
|
41
|
+
}
|
|
42
|
+
function sanitizeEvalReportDirLabel(label) {
|
|
43
|
+
const normalized = label.startsWith("eval:") ? label.slice("eval:".length) : label;
|
|
44
|
+
return normalized.trim().replace(/[^A-Za-z0-9._-]+/g, "-").replace(/-+/g, "-").replace(/^[-._]+|[-._]+$/g, "");
|
|
45
|
+
}
|
|
46
|
+
export function createDefaultEvalReportDir(runId, label) {
|
|
47
|
+
const timestamp = createEvalReportDirTimestamp(runId);
|
|
48
|
+
const suffix = label ? sanitizeEvalReportDirLabel(label) : "";
|
|
49
|
+
return join(".veryfront", "evals", suffix ? `${timestamp}-${suffix}` : timestamp);
|
|
41
50
|
}
|
|
42
51
|
export function createEvalArtifactPaths(reportDir) {
|
|
43
52
|
return {
|
|
@@ -313,8 +322,14 @@ function usdCell(value) {
|
|
|
313
322
|
return `$${value.toFixed(2)}`;
|
|
314
323
|
return `$${value.toFixed(6).replace(/0+$/, "").replace(/\.$/, "")}`;
|
|
315
324
|
}
|
|
325
|
+
function isBlockingEvalResultFailure(result) {
|
|
326
|
+
return !result.skipped && result.pass === false &&
|
|
327
|
+
(result.severity === "gate" || result.severity === "budget");
|
|
328
|
+
}
|
|
316
329
|
function examplePassed(record) {
|
|
317
|
-
|
|
330
|
+
if (!record.completed || record.error)
|
|
331
|
+
return false;
|
|
332
|
+
return [...(record.metrics ?? []), ...(record.checks ?? [])].every((result) => !isBlockingEvalResultFailure(result));
|
|
318
333
|
}
|
|
319
334
|
function usageRows(usage) {
|
|
320
335
|
if (!usage)
|
|
@@ -605,7 +620,8 @@ async function writeEvalModelComparisonArtifacts(comparison, paths) {
|
|
|
605
620
|
}
|
|
606
621
|
async function runEvalModelComparison(input) {
|
|
607
622
|
const runId = createCliRunId();
|
|
608
|
-
const reportDir = input.options.reportDir ??
|
|
623
|
+
const reportDir = input.options.reportDir ??
|
|
624
|
+
createDefaultEvalReportDir(runId, input.evalItem.id);
|
|
609
625
|
const paths = createEvalModelComparisonArtifactPaths(reportDir, input.config.models);
|
|
610
626
|
const provenance = await resolveEvalRunProvenance({
|
|
611
627
|
projectDir: input.projectDir,
|
|
@@ -788,7 +804,7 @@ export async function evalCommand(options) {
|
|
|
788
804
|
return;
|
|
789
805
|
}
|
|
790
806
|
const runId = createCliRunId();
|
|
791
|
-
const artifactPaths = createEvalArtifactPaths(options.reportDir ?? createDefaultEvalReportDir(runId));
|
|
807
|
+
const artifactPaths = createEvalArtifactPaths(options.reportDir ?? createDefaultEvalReportDir(runId, evalItem.id));
|
|
792
808
|
const provenance = await resolveEvalRunProvenance({
|
|
793
809
|
projectDir,
|
|
794
810
|
frameworkVersion: VERSION,
|
package/esm/deno.js
CHANGED