@thanh01.pmt/curriculum-kit 1.4.4 → 1.4.5

package/dist/index.cjs

@@ -530,7 +530,9 @@ Guidelines:
     if (isNvidiaPreferred || uniqueNvidiaModels.length > 0) {
       for (const model of uniqueNvidiaModels) {
         try {
-          const idle = createStreamAbortController({ idleMs: options.idleTimeoutMs, totalMs: options.timeoutMs });
+          const nvdIdle = options.idleTimeoutMs ?? 18e4;
+          const nvdTotal = options.timeoutMs ?? 6e5;
+          const idle = createStreamAbortController({ idleMs: nvdIdle, totalMs: nvdTotal });
           const res = await fetch("https://integrate.api.nvidia.com/v1/chat/completions", {
             method: "POST",
             headers: {
@@ -727,7 +729,9 @@ Guidelines:
       const { provider, model } = target;
       if (provider === "nvidia" && nvidiaKey && isProviderEnabled("nvidia")) {
         try {
-          const idle = createStreamAbortController({ idleMs: options.idleTimeoutMs, totalMs: options.timeoutMs });
+          const nvdIdle = options.idleTimeoutMs ?? 18e4;
+          const nvdTotal = options.timeoutMs ?? 6e5;
+          const idle = createStreamAbortController({ idleMs: nvdIdle, totalMs: nvdTotal });
           const res = await fetch("https://integrate.api.nvidia.com/v1/chat/completions", {
             method: "POST",
             headers: {
@@ -881,8 +885,8 @@ exports.DEFAULT_ARTIFACT_STREAM_IDLE_MS = void 0; exports.DEFAULT_ARTIFACT_STREA
 var init_streamRunner = __esm({
   "src/ai/streamRunner.ts"() {
     init_errors();
-    exports.DEFAULT_ARTIFACT_STREAM_IDLE_MS = 45e3;
-    exports.DEFAULT_ARTIFACT_STREAM_TOTAL_MS = 42e4;
+    exports.DEFAULT_ARTIFACT_STREAM_IDLE_MS = 9e4;
+    exports.DEFAULT_ARTIFACT_STREAM_TOTAL_MS = 48e4;
   }
 });
 
@@ -27083,20 +27087,34 @@ var DeterministicPipelineRunner = class {
 
 // src/services/prefillService.ts
 init_streamRunner();
-var DEFAULT_STREAM_IDLE_MS = 45e3;
-var DEFAULT_STREAM_TOTAL_MS = 42e4;
+var DEFAULT_STREAM_IDLE_MS = 9e4;
+var DEFAULT_STREAM_TOTAL_MS = 48e4;
+var LAYER_IDLE_BUDGET_MS = {
+  1: 9e4,
+  2: 9e4,
+  3: 18e4
+  // Layer 3 synthesizes all accumulated context + pathways (3 minutes idle protection)
+};
 var LAYER_TOTAL_BUDGET_MS = {
-  1: 42e4,
-  2: 42e4,
-  3: 42e4
+  1: 48e4,
+  2: 48e4,
+  3: 6e5
 };
 function resolveStreamBudget(layer, opts) {
   const envIdle = Number(process.env.WIZARD_PREFILL_IDLE_TIMEOUT_MS);
   const envTotal = Number(process.env.WIZARD_PREFILL_TOTAL_TIMEOUT_MS);
+  const isNvidia = Boolean(
+    opts?.provider && opts.provider.toLowerCase().includes("nvidia") || opts?.model && (opts.model.toLowerCase().includes("nvidia") || opts.model.toLowerCase().includes("nemotron"))
+  );
+  const layerIdle = layer !== void 0 ? LAYER_IDLE_BUDGET_MS[layer] : void 0;
+  const baseIdle = layerIdle ?? DEFAULT_STREAM_IDLE_MS;
+  const computedDefaultIdle = isNvidia ? Math.max(baseIdle, 18e4) : baseIdle;
   const layerTotal = layer !== void 0 ? LAYER_TOTAL_BUDGET_MS[layer] : void 0;
+  const baseTotal = layerTotal ?? DEFAULT_STREAM_TOTAL_MS;
+  const computedDefaultTotal = isNvidia ? Math.max(baseTotal, 6e5) : baseTotal;
   return {
-    idleMs: opts?.idleMs ?? (Number.isFinite(envIdle) && envIdle > 0 ? envIdle : DEFAULT_STREAM_IDLE_MS),
-    totalMs: opts?.totalMs ?? (Number.isFinite(envTotal) && envTotal > 0 ? envTotal : layerTotal ?? DEFAULT_STREAM_TOTAL_MS)
+    idleMs: opts?.idleMs ?? (Number.isFinite(envIdle) && envIdle > 0 ? envIdle : computedDefaultIdle),
+    totalMs: opts?.totalMs ?? (Number.isFinite(envTotal) && envTotal > 0 ? envTotal : computedDefaultTotal)
   };
 }
 function createStreamChunkExtractor() {
@@ -27439,7 +27457,12 @@ Return concise JSON matching:
     options.model,
     options.provider,
     // Research may run live web grounding — generous budget, still idle-guarded.
-    resolveStreamBudget(void 0, { idleMs: options.idleTimeoutMs, totalMs: options.timeoutMs })
+    resolveStreamBudget(void 0, {
+      idleMs: options.idleTimeoutMs,
+      totalMs: options.timeoutMs,
+      model: options.model,
+      provider: options.provider
+    })
   );
   let parsedResearch = {
     groundTruthSummary: `Verified ecosystem standards for ${projectName}`,
@@ -27642,9 +27665,14 @@ REQUIREMENT: Synthesize ALL parameters above \u2014 every single line in FULL AC
     },
     options.model,
     options.provider,
-    // RC-W1: layer-aware budget — Layer 2 has the largest output and free-tier
-    // models may think/stream for minutes. Idle window still catches hangs.
-    resolveStreamBudget(targetLayer, { idleMs: options.idleTimeoutMs, totalMs: options.timeoutMs })
+    // RC-W1: layer-aware budget — Layer 3 synthesizes full context and free-tier
+    // models (especially NVIDIA NIM) may think/stream for minutes. Idle window still catches hangs.
+    resolveStreamBudget(targetLayer, {
+      idleMs: options.idleTimeoutMs,
+      totalMs: options.timeoutMs,
+      model: options.model,
+      provider: options.provider
+    })
   );
   const parsed = safeParseJson(rawContent);
   return {
@@ -30220,6 +30248,7 @@ exports.HandoutSectionSchema = HandoutSectionSchema;
 exports.InstructionSectionSchema = InstructionSectionSchema;
 exports.InstructionStepSchema = InstructionStepSchema;
 exports.JudgeCriterionSchema = JudgeCriterionSchema;
+exports.LAYER_IDLE_BUDGET_MS = LAYER_IDLE_BUDGET_MS;
 exports.LAYER_TOTAL_BUDGET_MS = LAYER_TOTAL_BUDGET_MS;
 exports.LESSON_PLAN_TEMPLATE = LESSON_PLAN_TEMPLATE;
 exports.LLMJudgeEngine = LLMJudgeEngine;