@qwen-code/qwen-code 0.19.8 → 0.19.9

package/chunks/{chunk-HW4N45EE.js → chunk-524RQP4Q.js}

@@ -2,7 +2,7 @@
 "use strict";
 import {
   init_esbuild_shims
-} from "./chunk-A4BMJM77.js";
+} from "./chunk-5O2XNYP6.js";
 import {
   __name
 } from "./chunk-J2S4EL5Y.js";
@@ -12,6 +12,17 @@ init_esbuild_shims();
 var DEFAULT_TOKEN_LIMIT = 2e5;
 var DEFAULT_OUTPUT_TOKEN_LIMIT = 32e3;
 var ESCALATED_MAX_TOKENS = 64e3;
+var OUTPUT_TOKEN_CEILING = ESCALATED_MAX_TOKENS;
+var MIN_CLAMPED_OUTPUT_TOKENS = 4e3;
+function outputClampMargin(contextWindowSize) {
+  return Math.max(1e4, Math.round(0.05 * contextWindowSize));
+}
+__name(outputClampMargin, "outputClampMargin");
+function clampOutputTokensToWindow(outputCeiling, contextWindowSize, promptTokens) {
+  const room = contextWindowSize - promptTokens - outputClampMargin(contextWindowSize);
+  return Math.min(outputCeiling, Math.max(MIN_CLAMPED_OUTPUT_TOKENS, room));
+}
+__name(clampOutputTokensToWindow, "clampOutputTokensToWindow");
 function parsePositiveIntegerEnvValue(raw) {
   if (raw === void 0) return void 0;
   const trimmed = raw.trim();
@@ -190,12 +201,6 @@ function hasExplicitOutputLimit(model) {
   return OUTPUT_PATTERNS.some(([regex]) => regex.test(norm));
 }
 __name(hasExplicitOutputLimit, "hasExplicitOutputLimit");
-function escalatedOutputTokenLimit(model, contextWindowSize) {
-  const escalated = Math.max(ESCALATED_MAX_TOKENS, tokenLimit(model, "output"));
-  const window = contextWindowSize !== void 0 && contextWindowSize > 0 ? contextWindowSize : DEFAULT_TOKEN_LIMIT;
-  return Math.min(escalated, Math.floor(window / 2));
-}
-__name(escalatedOutputTokenLimit, "escalatedOutputTokenLimit");
 function knownTokenLimit(model, type = "input") {
   return findTokenLimit(model, type);
 }
@@ -204,6 +209,17 @@ function tokenLimit(model, type = "input") {
   return knownTokenLimit(model, type) ?? (type === "output" ? DEFAULT_OUTPUT_TOKEN_LIMIT : DEFAULT_TOKEN_LIMIT);
 }
 __name(tokenLimit, "tokenLimit");
+function defaultOutputCeiling(model) {
+  return Math.min(tokenLimit(model, "output"), OUTPUT_TOKEN_CEILING);
+}
+__name(defaultOutputCeiling, "defaultOutputCeiling");
+function reconcileMaxTokens(configMaxTokens, requestMaxTokens) {
+  if (typeof configMaxTokens === "number" && typeof requestMaxTokens === "number") {
+    return Math.min(configMaxTokens, requestMaxTokens);
+  }
+  return void 0;
+}
+__name(reconcileMaxTokens, "reconcileMaxTokens");
 
 // packages/core/src/core/openaiContentGenerator/constants.ts
 init_esbuild_shims();
@@ -258,12 +274,17 @@ export {
   DEFAULT_TOKEN_LIMIT,
   DEFAULT_OUTPUT_TOKEN_LIMIT,
   ESCALATED_MAX_TOKENS,
+  OUTPUT_TOKEN_CEILING,
+  MIN_CLAMPED_OUTPUT_TOKENS,
+  outputClampMargin,
+  clampOutputTokensToWindow,
   parsePositiveIntegerEnvValue,
   normalize,
   hasExplicitOutputLimit,
-  escalatedOutputTokenLimit,
   knownTokenLimit,
   tokenLimit,
+  defaultOutputCeiling,
+  reconcileMaxTokens,
   resolveRequestTimeout,
   DEFAULT_STREAM_IDLE_TIMEOUT_MS,
   QWEN_STREAM_IDLE_TIMEOUT_MS_ENV,

package/chunks/{chunk-3GONHQOA.js → chunk-57OAHC2Y.js}

@@ -2,7 +2,7 @@
 "use strict";
 import {
   init_esbuild_shims
-} from "./chunk-A4BMJM77.js";
+} from "./chunk-5O2XNYP6.js";
 import {
   __commonJS,
   __name,

package/chunks/{chunk-JZSY4WP3.js → chunk-5DTTYNB7.js}

@@ -2,7 +2,7 @@
 "use strict";
 import {
   init_esbuild_shims
-} from "./chunk-A4BMJM77.js";
+} from "./chunk-5O2XNYP6.js";
 import {
   __name
 } from "./chunk-J2S4EL5Y.js";