@qwen-code/qwen-code 0.12.5 → 0.12.6

package/cli.js

@@ -133792,6 +133792,10 @@ function normalize2(model) {
   s5 = s5.replace(/-(?:\d?bit|int[48]|bf16|fp16|q[45]|quantized)$/g, "");
   return s5;
 }
+function hasExplicitOutputLimit(model) {
+  const norm = normalize2(model);
+  return OUTPUT_PATTERNS.some(([regex3]) => regex3.test(norm));
+}
 function tokenLimit(model, type = "input") {
   const norm = normalize2(model);
   const patterns = type === "output" ? OUTPUT_PATTERNS : PATTERNS;
@@ -133808,7 +133812,7 @@ var init_tokenLimits = __esm({
     "use strict";
     init_esbuild_shims();
     DEFAULT_TOKEN_LIMIT = 131072;
-    DEFAULT_OUTPUT_TOKEN_LIMIT = 16384;
+    DEFAULT_OUTPUT_TOKEN_LIMIT = 32e3;
     LIMITS = {
       "32k": 32768,
       "64k": 65536,
@@ -133932,6 +133936,7 @@ var init_tokenLimits = __esm({
       // Kimi
       [/^kimi-k2\.5/, LIMITS["32k"]]
     ];
+    __name(hasExplicitOutputLimit, "hasExplicitOutputLimit");
     __name(tokenLimit, "tokenLimit");
   }
 });
@@ -134801,18 +134806,6 @@ var init_modelsConfig = __esm({
             detail: "auto-detected from model"
           };
         }
-        if (!this._generationConfig.samplingParams?.max_tokens) {
-          const outputLimit = tokenLimit(model.id, "output");
-          if (!this._generationConfig.samplingParams) {
-            this._generationConfig.samplingParams = {};
-          }
-          this._generationConfig.samplingParams.max_tokens = outputLimit;
-          const existingSource = this.generationConfigSources["samplingParams"];
-          this.generationConfigSources["samplingParams"] = {
-            kind: "computed",
-            detail: existingSource ? `max_tokens auto-detected from model (other params from ${existingSource.kind})` : "max_tokens auto-detected from model"
-          };
-        }
         if (gc.modalities === void 0) {
           this._generationConfig.modalities = defaultModalities(model.id);
           this.generationConfigSources["modalities"] = {
@@ -142713,6 +142706,7 @@ var init_default = __esm({
     init_openai();
     init_constants2();
     init_runtimeFetchOptions();
+    init_tokenLimits();
     DefaultOpenAICompatibleProvider = class {
       static {
         __name(this, "DefaultOpenAICompatibleProvider");
@@ -142747,15 +142741,66 @@ var init_default = __esm({
       }
       buildRequest(request4, _userPromptId) {
         const extraBody = this.contentGeneratorConfig.extra_body;
+        const requestWithTokenLimits = this.applyOutputTokenLimit(request4);
         return {
-          ...request4,
-          // Preserve all original parameters including sampling params
+          ...requestWithTokenLimits,
           ...extraBody ? extraBody : {}
         };
       }
       getDefaultGenerationConfig() {
         return {};
       }
+      /**
+       * Apply output token limit to a request's max_tokens parameter.
+       *
+       * Purpose:
+       * Some APIs (e.g., OpenAI-compatible) default to a very small max_tokens value,
+       * which can cause responses to be truncated mid-output. This function ensures
+       * a reasonable default is set while respecting user configuration.
+       *
+       * Logic:
+       * 1. If user explicitly configured max_tokens:
+       *    - For known models (in OUTPUT_PATTERNS): use the user's value, but cap at
+       *      model's max output limit to avoid API errors
+       *      (input + max_output > contextWindowSize would cause 400 errors on some APIs)
+       *    - For unknown models (deployment aliases, self-hosted): respect user's
+       *      configured value entirely (backend may support larger limits)
+       * 2. If user didn't configure max_tokens:
+       *    - Use min(modelLimit, DEFAULT_OUTPUT_TOKEN_LIMIT)
+       *    - This provides a conservative default (32K) that avoids truncating output
+       *      while preserving input quota (not occupying too much context window)
+       * 3. If model has no specific limit (tokenLimit returns default):
+       *    - Still apply DEFAULT_OUTPUT_TOKEN_LIMIT as safeguard
+       *
+       * Examples:
+       * - User sets 4K, known model limit 64K → uses 4K (respects user preference)
+       * - User sets 100K, known model limit 64K → uses 64K (capped to avoid API error)
+       * - User sets 100K, unknown model → uses 100K (respects user, backend may support it)
+       * - User not set, model limit 64K → uses 32K (conservative default)
+       * - User not set, model limit 8K → uses 8K (model limit is lower)
+       *
+       * @param request - The chat completion request parameters
+       * @returns The request with max_tokens adjusted according to the logic
+       */
+      applyOutputTokenLimit(request4) {
+        const userMaxTokens = request4.max_tokens;
+        const modelLimit = tokenLimit(request4.model, "output");
+        const isKnownModel = hasExplicitOutputLimit(request4.model);
+        let effectiveMaxTokens;
+        if (userMaxTokens !== void 0 && userMaxTokens !== null) {
+          if (isKnownModel) {
+            effectiveMaxTokens = Math.min(userMaxTokens, modelLimit);
+          } else {
+            effectiveMaxTokens = userMaxTokens;
+          }
+        } else {
+          effectiveMaxTokens = Math.min(modelLimit, DEFAULT_OUTPUT_TOKEN_LIMIT);
+        }
+        return {
+          ...request4,
+          max_tokens: effectiveMaxTokens
+        };
+      }
     };
   }
 });
@@ -142802,16 +142847,13 @@ var init_dashscope = __esm({
     init_contentGenerator();
     init_constants2();
     init_runtimeFetchOptions();
-    init_tokenLimits();
-    DashScopeOpenAICompatibleProvider = class _DashScopeOpenAICompatibleProvider {
+    init_default();
+    DashScopeOpenAICompatibleProvider = class _DashScopeOpenAICompatibleProvider extends DefaultOpenAICompatibleProvider {
       static {
         __name(this, "DashScopeOpenAICompatibleProvider");
       }
-      contentGeneratorConfig;
-      cliConfig;
       constructor(contentGeneratorConfig, cliConfig) {
-        this.cliConfig = cliConfig;
-        this.contentGeneratorConfig = contentGeneratorConfig;
+        super(contentGeneratorConfig, cliConfig);
       }
       static isDashScopeProvider(contentGeneratorConfig) {
         const { authType, baseUrl } = contentGeneratorConfig;
@@ -143003,29 +143045,6 @@ var init_dashscope = __esm({
         }
         return false;
       }
-      /**
-       * Apply output token limit to a request's max_tokens parameter.
-       *
-       * Ensures that existing max_tokens parameters don't exceed the model's maximum output
-       * token limit. Only modifies max_tokens when already present in the request.
-       *
-       * @param request - The chat completion request parameters
-       * @returns The request with max_tokens adjusted to respect the model's limits (if present)
-       */
-      applyOutputTokenLimit(request4) {
-        const currentMaxTokens = request4.max_tokens;
-        if (currentMaxTokens === void 0 || currentMaxTokens === null) {
-          return request4;
-        }
-        const modelLimit = tokenLimit(request4.model, "output");
-        if (currentMaxTokens > modelLimit) {
-          return {
-            ...request4,
-            max_tokens: modelLimit
-          };
-        }
-        return request4;
-      }
       /**
        * Check if cache control should be disabled based on configuration.
        *
@@ -157264,6 +157283,7 @@ var init_anthropicContentGenerator = __esm({
     init_runtimeFetchOptions();
     init_constants2();
     init_debugLogger();
+    init_tokenLimits();
     debugLogger14 = createDebugLogger("ANTHROPIC");
     AnthropicContentGenerator = class {
       static {
@@ -157373,7 +157393,11 @@ var init_anthropicContentGenerator = __esm({
           const requestValue = requestKey ? requestConfig[requestKey] : void 0;
           return configValue !== void 0 ? configValue : requestValue;
         }, "getParam");
-        const maxTokens = getParam("max_tokens", "maxOutputTokens") ?? 1e4;
+        const userMaxTokens = getParam("max_tokens", "maxOutputTokens");
+        const modelId = this.contentGeneratorConfig.model;
+        const modelLimit = tokenLimit(modelId, "output");
+        const isKnownModel = hasExplicitOutputLimit(modelId);
+        const maxTokens = userMaxTokens !== void 0 && userMaxTokens !== null ? isKnownModel ? Math.min(userMaxTokens, modelLimit) : userMaxTokens : Math.min(modelLimit, DEFAULT_OUTPUT_TOKEN_LIMIT);
         return {
           max_tokens: maxTokens,
           temperature: getParam("temperature", "temperature") ?? 1,
@@ -157778,7 +157802,7 @@ __export(geminiContentGenerator_exports, {
   createGeminiContentGenerator: () => createGeminiContentGenerator
 });
 function createGeminiContentGenerator(config2, gcConfig) {
-  const version2 = "0.12.5";
+  const version2 = "0.12.6";
   const userAgent2 = config2.userAgent || `QwenCode/${version2} (${process.platform}; ${process.arch})`;
   const baseHeaders = {
     "User-Agent": userAgent2
@@ -390052,7 +390076,7 @@ __name(getPackageJson, "getPackageJson");
 // packages/cli/src/utils/version.ts
 async function getCliVersion() {
   const pkgJson = await getPackageJson();
-  return "0.12.5";
+  return "0.12.6";
 }
 __name(getCliVersion, "getCliVersion");
 
@@ -397613,7 +397637,7 @@ var formatDuration = /* @__PURE__ */ __name((milliseconds) => {
 
 // packages/cli/src/generated/git-commit.ts
 init_esbuild_shims();
-var GIT_COMMIT_INFO = "dbfa5b3e8";
+var GIT_COMMIT_INFO = "ac30c98a2";
 
 // packages/cli/src/utils/systemInfo.ts
 async function getNpmVersion() {
@@ -456290,7 +456314,7 @@ var QwenAgent = class {
   async initialize(args) {
     this.clientCapabilities = args.clientCapabilities;
     const authMethods = buildAuthMethods();
-    const version2 = "0.12.5";
+    const version2 = "0.12.6";
     return {
       protocolVersion: PROTOCOL_VERSION,
       agentInfo: {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@qwen-code/qwen-code",
-  "version": "0.12.5",
+  "version": "0.12.6",
   "description": "Qwen Code - AI-powered coding assistant",
   "repository": {
     "type": "git",
@@ -20,7 +20,7 @@
     "locales"
   ],
   "config": {
-    "sandboxImageUri": "ghcr.io/qwenlm/qwen-code:0.12.5"
+    "sandboxImageUri": "ghcr.io/qwenlm/qwen-code:0.12.6"
   },
   "dependencies": {},
   "optionalDependencies": {