@semiont/jobs 0.5.11 → 0.5.13

package/dist/index.d.ts

@@ -202,7 +202,8 @@ interface DetectionResult {
  * Generation job progress
  */
 interface YieldProgress {
-    stage: 'fetching' | 'generating' | 'creating' | 'linking';
+    /** The two real generation transitions — LLM call running, then persisting. */
+    stage: 'generating' | 'creating';
     percentage: number;
     message?: string;
 }

package/dist/index.js

@@ -419,6 +419,41 @@ function isFailedJob(job) {
 function isCancelledJob(job) {
   return job.status === "cancelled";
 }
+
+// src/workers/inference-call.ts
+var INFERENCE_TIMEOUT_MS = 10 * 6e4;
+async function withTimeout(work, label) {
+  let timer;
+  const timedOut = new Promise((_, reject) => {
+    timer = setTimeout(() => {
+      reject(new Error(
+        `Inference call timed out after ${INFERENCE_TIMEOUT_MS / 6e4} minutes (${label}) \u2014 failing the job to keep the claim loop live`
+      ));
+    }, INFERENCE_TIMEOUT_MS);
+    timer.unref?.();
+  });
+  try {
+    return await Promise.race([work, timedOut]);
+  } catch (err) {
+    work.catch(() => {
+    });
+    throw err;
+  } finally {
+    clearTimeout(timer);
+  }
+}
+function boundedGenerate(client, prompt, maxTokens, temperature, options) {
+  return withTimeout(
+    client.generateText(prompt, maxTokens, temperature, options),
+    `${client.type}:${client.modelId}`
+  );
+}
+function boundedGenerateWithMetadata(client, prompt, maxTokens, temperature, options) {
+  return withTimeout(
+    client.generateTextWithMetadata(prompt, maxTokens, temperature, options),
+    `${client.type}:${client.modelId}`
+  );
+}
 function languageName(tag) {
   return getLocaleEnglishName(tag) || tag;
 }
@@ -900,7 +935,7 @@ var AnnotationDetection = class {
    */
   static async detectComments(content, client, instructions, tone, density, language, sourceLanguage) {
     const prompt = MotivationPrompts.buildCommentPrompt(content, instructions, tone, density, language, sourceLanguage);
-    const response = await client.generateTextWithMetadata(prompt, 3e3, 0.4, { format: "json" });
+    const response = await boundedGenerateWithMetadata(client, prompt, 3e3, 0.4, { format: "json" });
     assertNotTruncated(response, "comment");
     return MotivationParsers.parseComments(response.text, content);
   }
@@ -913,7 +948,7 @@ var AnnotationDetection = class {
    */
   static async detectHighlights(content, client, instructions, density, sourceLanguage) {
     const prompt = MotivationPrompts.buildHighlightPrompt(content, instructions, density, sourceLanguage);
-    const response = await client.generateTextWithMetadata(prompt, 2e3, 0.3, { format: "json" });
+    const response = await boundedGenerateWithMetadata(client, prompt, 2e3, 0.3, { format: "json" });
     assertNotTruncated(response, "highlight");
     return MotivationParsers.parseHighlights(response.text, content);
   }
@@ -926,7 +961,7 @@ var AnnotationDetection = class {
    */
   static async detectAssessments(content, client, instructions, tone, density, language, sourceLanguage) {
     const prompt = MotivationPrompts.buildAssessmentPrompt(content, instructions, tone, density, language, sourceLanguage);
-    const response = await client.generateTextWithMetadata(prompt, 3e3, 0.3, { format: "json" });
+    const response = await boundedGenerateWithMetadata(client, prompt, 3e3, 0.3, { format: "json" });
     assertNotTruncated(response, "assessment");
     return MotivationParsers.parseAssessments(response.text, content);
   }
@@ -957,7 +992,7 @@ var AnnotationDetection = class {
       categoryInfo.examples,
       sourceLanguage
     );
-    const response = await client.generateTextWithMetadata(prompt, 4e3, 0.2, { format: "json" });
+    const response = await boundedGenerateWithMetadata(client, prompt, 4e3, 0.2, { format: "json" });
     assertNotTruncated(response, "tag");
     const parsedTags = MotivationParsers.parseTags(response.text);
     return MotivationParsers.validateTagOffsets(parsedTags, content, category);
@@ -1013,7 +1048,8 @@ If no entities are found, respond with an empty array [].
 Example output:
 [{"exact":"Alice","entityType":"Person","prefix":"","suffix":" went to"},{"exact":"Paris","entityType":"Location","prefix":"went to ","suffix":" yesterday"}]`;
   logger.debug("Sending entity extraction request", { entityTypes: entityTypesDescription });
-  const response = await client.generateTextWithMetadata(
+  const response = await boundedGenerateWithMetadata(
+    client,
     prompt,
     4e3,
     // Increased to handle many entities without truncation
@@ -1263,7 +1299,7 @@ ${formatRequirements}`;
     temperature: finalTemperature,
     maxTokens: finalMaxTokens
   });
-  const response = await client.generateText(prompt, finalMaxTokens, finalTemperature);
+  const response = await boundedGenerate(client, prompt, finalMaxTokens, finalTemperature);
   logger.debug("Got response from inference", { responseLength: response.length });
   const result = parseResponse(response);
   logger.debug("Parsed response", {
@@ -1623,10 +1659,9 @@ async function processGenerationJob(inferenceClient, params, onProgress, logger)
       `Unsupported outputMediaType for generation: ${outputMediaType}. Generation produces ${GENERATABLE_MEDIA_TYPES.join(" or ")}.`
     );
   }
-  onProgress(20, "Fetching context...", "fetching");
   const title = params.title ?? "Untitled";
   const entityTypes = (params.entityTypes ?? []).map(String);
-  onProgress(40, "Generating resource...", "generating");
+  onProgress(5, "Generating resource...", "generating");
   const generated = await generateResourceFromTopic(
     title,
     entityTypes,
@@ -1650,7 +1685,7 @@ async function processGenerationJob(inferenceClient, params, onProgress, logger)
     content = resolved.content;
     citations = resolved.citations;
   }
-  onProgress(85, "Creating resource...", "creating");
+  onProgress(95, "Creating resource...", "creating");
   return {
     content,
     title: generated.title ?? title,