@wix/evalforge-evaluator 0.206.0 → 0.208.0

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/build/index.js CHANGED
@@ -10902,7 +10902,7 @@ var MODEL_PRICING = {
10902
10902
  "claude-opus-4-5": { input: 5, output: 25 },
10903
10903
  "claude-sonnet-4-5": { input: 3, output: 15 },
10904
10904
  "claude-haiku-4-5": { input: 1, output: 5 },
10905
- // Anthropic — Claude 4
10905
+ // Anthropic — Claude 4.0 (retired; kept so historical runs still show cost)
10906
10906
  "claude-opus-4": { input: 15, output: 75 },
10907
10907
  "claude-sonnet-4": { input: 3, output: 15 },
10908
10908
  // OpenAI — GPT-5
@@ -11246,13 +11246,13 @@ async function executeWithAiSdk(context) {
11246
11246
  [text, steps, rawUsage] = await Promise.all([
11247
11247
  stream.text,
11248
11248
  stream.steps,
11249
- stream.usage
11249
+ stream.totalUsage
11250
11250
  ]);
11251
11251
  } else {
11252
11252
  const result = await (0, import_ai.generateText)(sdkParams);
11253
11253
  text = result.text;
11254
11254
  steps = result.steps;
11255
- rawUsage = result.usage;
11255
+ rawUsage = result.totalUsage;
11256
11256
  }
11257
11257
  const durationMs = Date.now() - startTime;
11258
11258
  const usage = {
@@ -11263,7 +11263,6 @@ async function executeWithAiSdk(context) {
11263
11263
  const llmTrace = buildLLMTrace2(
11264
11264
  steps,
11265
11265
  durationMs,
11266
- usage,
11267
11266
  modelId,
11268
11267
  provider,
11269
11268
  startTime,
@@ -11412,7 +11411,7 @@ function findToolResultError(step) {
11412
11411
  }
11413
11412
  return null;
11414
11413
  }
11415
- function buildLLMTrace2(steps, totalDurationMs, totalUsage, modelId, provider, executionStartMs, stepTimestamps) {
11414
+ function buildLLMTrace2(steps, totalDurationMs, modelId, provider, executionStartMs, stepTimestamps) {
11416
11415
  const traceSteps = steps.map((step, i) => {
11417
11416
  const stepFinishedAt = stepTimestamps[i] ?? executionStartMs;
11418
11417
  const stepStartedAt = i === 0 ? executionStartMs : stepTimestamps[i - 1] ?? executionStartMs;
@@ -11444,11 +11443,14 @@ function buildLLMTrace2(steps, totalDurationMs, totalUsage, modelId, provider, e
11444
11443
  };
11445
11444
  });
11446
11445
  const totalCostUsd = traceSteps.reduce((sum, s) => sum + s.costUsd, 0);
11447
- const finalTokens = {
11448
- prompt: totalUsage.inputTokens,
11449
- completion: totalUsage.outputTokens,
11450
- total: totalUsage.totalTokens
11451
- };
11446
+ const finalTokens = traceSteps.reduce(
11447
+ (acc, s) => ({
11448
+ prompt: acc.prompt + s.tokenUsage.prompt,
11449
+ completion: acc.completion + s.tokenUsage.completion,
11450
+ total: acc.total + s.tokenUsage.total
11451
+ }),
11452
+ { prompt: 0, completion: 0, total: 0 }
11453
+ );
11452
11454
  return {
11453
11455
  id: (0, import_crypto4.randomUUID)(),
11454
11456
  steps: traceSteps,