@m4trix/evals 0.15.0 → 0.17.0

package/dist/cli-simple.js

@@ -6,7 +6,7 @@ import { resolve, relative, join, parse, dirname } from 'path';
 import * as jitiModule from 'jiti';
 import { writeFile, readdir, readFile, mkdir, appendFile } from 'fs/promises';
 import { pathToFileURL } from 'url';
-import { diffLines } from 'diff';
+import { diffString } from 'json-diff';
 import React2, { useState, useEffect, useCallback } from 'react';
 import { render, Box, Text } from 'ink';
 import { jsxs, jsx, Fragment } from 'react/jsx-runtime';
@@ -260,45 +260,46 @@ async function collectTestCasesFromFiles(config) {
   );
   return found.flat();
 }
-function toJsonLines(value) {
+function createDiffString(expected, actual, diffOptions) {
+  const opts = { ...diffOptions, color: false };
+  const result = diffString(expected, actual, opts);
+  return typeof result === "string" ? result : "";
+}
+function formatLogMessage(msg) {
+  if (typeof msg === "string")
+    return msg;
   try {
-    return JSON.stringify(value, null, 2);
+    if (msg !== null && typeof msg === "object") {
+      return JSON.stringify(msg, null, 2);
+    }
+    return String(msg);
   } catch {
-    return String(value);
+    return String(msg);
   }
 }
-function formatDiffString(changes) {
-  const lines = [];
-  for (const part of changes) {
-    const prefix = part.added ? "+" : part.removed ? "-" : " ";
-    const partLines = part.value.split("\n");
-    if (partLines[partLines.length - 1] === "") {
-      partLines.pop();
-    }
-    for (const line of partLines) {
-      lines.push(`${prefix} ${line}`);
-    }
-  }
-  return lines.join("\n");
+function createLogEntry(message, options) {
+  return {
+    type: "log",
+    label: options?.label,
+    message: formatLogMessage(message)
+  };
 }
-function createDiffString(expected, actual) {
-  const expectedStr = toJsonLines(expected);
-  const actualStr = toJsonLines(actual);
-  const changes = diffLines(expectedStr, actualStr);
-  return formatDiffString(changes);
+function getLogLines(entry) {
+  return entry.message.split("\n");
 }
 function createDiffLogEntry(expected, actual, options) {
-  const diff = createDiffString(expected, actual);
+  const { label, ...diffOpts } = options ?? {};
+  const diff = createDiffString(expected, actual, diffOpts);
   return {
     type: "diff",
-    label: options?.label,
+    label,
     expected,
     actual,
     diff: diff || "(no differences)"
   };
 }
 function getDiffLines(entry) {
-  const raw = createDiffString(entry.expected, entry.actual) || "(no differences)";
+  const raw = entry.diff || "(no differences)";
   return raw.split("\n").map((line) => {
     const trimmed = line.trimStart();
     if (trimmed.startsWith("-") && !trimmed.startsWith("---")) {
@@ -560,6 +561,7 @@ function processOneTestCase(task, testCaseItem, totalEvaluations, publishEvent,
     const reruns = typeof testCaseItem.testCase.getReruns === "function" ? testCaseItem.testCase.getReruns() : 1;
     const rerunPassed = [];
     for (let r = 0; r < reruns; r++) {
+      const evaluatorRunId = `run-${randomUUID()}`;
       const started = Date.now();
       const evaluatorScores = [];
       let testCaseError;
@@ -574,6 +576,9 @@ function processOneTestCase(task, testCaseItem, totalEvaluations, publishEvent,
           const logDiff = (expected, actual, options) => {
             logs.push(createDiffLogEntry(expected, actual, options));
           };
+          const log = (message, options) => {
+            logs.push(createLogEntry(message, options));
+          };
           const ctx = yield* Effect.promise(
             () => Promise.resolve(evaluator.resolveContext())
           );
@@ -583,7 +588,13 @@ function processOneTestCase(task, testCaseItem, totalEvaluations, publishEvent,
                 input: testCaseItem.testCase.getInput(),
                 ctx,
                 output,
-                logDiff
+                meta: {
+                  triggerId: task.triggerId,
+                  runId: evaluatorRunId,
+                  datasetId: task.datasetId
+                },
+                logDiff,
+                log
               })
             )
           );
@@ -1041,6 +1052,7 @@ var EffectRunner = class {
       (sum, tc) => sum + (typeof tc.testCase.getReruns === "function" ? tc.testCase.getReruns() : 1),
       0
     );
+    const triggerId = request.triggerId ?? `trg-${randomUUID()}`;
     const runId = `run-${randomUUID()}`;
     const artifactPath = createArtifactPath(
       this.config.artifactDirectory,
@@ -1082,6 +1094,7 @@ var EffectRunner = class {
     await Effect.runPromise(
       Queue.offer(this.runQueue, {
         runId,
+        triggerId,
         datasetId: request.datasetId,
         dataset: dataset.dataset,
         evaluators: selectedEvaluators,
@@ -1756,7 +1769,7 @@ function RunView({
                   },
                   lineIdx
                 )
-              ) }, logIdx) : null
+              ) }, logIdx) : log.type === "log" ? /* @__PURE__ */ jsx(Box, { flexDirection: "column", children: getLogLines(log).map((line, lineIdx) => /* @__PURE__ */ jsx(Text, { color: "gray", children: line }, lineIdx)) }, logIdx) : null
             ) })
           ]
         },
@@ -2260,6 +2273,10 @@ async function runSimpleEvalCommandPlain(runner, datasetName, evaluatorPattern)
                   const colored = useColor && type === "remove" ? colorize(`      ${line}`, ansi2.red) : useColor && type === "add" ? colorize(`      ${line}`, ansi2.green) : `      ${line}`;
                   lines.push(colored);
                 }
+              } else if (log.type === "log") {
+                for (const line of getLogLines(log)) {
+                  lines.push(`      ${line}`);
+                }
               }
             }
           }