@chatluna/v1-shared-adapter 1.0.27 → 1.0.29

package/lib/index.cjs

@@ -27,7 +27,9 @@ __export(index_exports, {
   convertMessageToMessageChunk: () => convertMessageToMessageChunk,
   createEmbeddings: () => createEmbeddings,
   createRequestContext: () => createRequestContext,
+  createUsageMetadata: () => createUsageMetadata,
   expandReasoningEffortModelVariants: () => expandReasoningEffortModelVariants,
+  fetchFileLikeUrl: () => fetchFileLikeUrl,
   fetchImageUrl: () => fetchImageUrl,
   formatToolToOpenAITool: () => formatToolToOpenAITool,
   formatToolsToOpenAITools: () => formatToolsToOpenAITools,
@@ -38,9 +40,9 @@ __export(index_exports, {
   langchainMessageToOpenAIMessage: () => langchainMessageToOpenAIMessage,
   messageTypeToOpenAIRole: () => messageTypeToOpenAIRole,
   normalizeOpenAIModelName: () => normalizeOpenAIModelName,
+  openAIUsageToUsageMetadata: () => openAIUsageToUsageMetadata,
   parseOpenAIModelNameWithReasoningEffort: () => parseOpenAIModelNameWithReasoningEffort,
   processDeepSeekThinkMessages: () => processDeepSeekThinkMessages,
-  processReasoningContent: () => processReasoningContent,
   processResponse: () => processResponse,
   processStreamResponse: () => processStreamResponse,
   reasoningEffortModelSuffixes: () => reasoningEffortModelSuffixes,
@@ -163,6 +165,8 @@ var imageModelMatchers = [
   "qwen*-omni",
   "qwen-omni",
   "qwen*-vl",
+  "qwen-3.5",
+  "qwen3.5",
   "qvq",
   "o1",
   "o3",
@@ -190,7 +194,38 @@ var import_messages = require("@langchain/core/messages");
 var import_zod_to_json_schema = require("zod-to-json-schema");
 var import_string = require("koishi-plugin-chatluna/utils/string");
 var import_types = require("@langchain/core/utils/types");
-async function langchainMessageToOpenAIMessage(messages, plugin, model, supportImageInput2, removeSystemMessage) {
+function createUsageMetadata(data) {
+  const inputTokenDetails = {
+    ...data.inputAudioTokens != null ? { audio: data.inputAudioTokens } : {},
+    ...data.cacheReadTokens != null ? { cache_read: data.cacheReadTokens } : {},
+    ...data.cacheCreationTokens != null ? { cache_creation: data.cacheCreationTokens } : {}
+  };
+  const outputTokenDetails = {
+    ...data.outputAudioTokens != null ? { audio: data.outputAudioTokens } : {},
+    ...data.reasoningTokens != null ? { reasoning: data.reasoningTokens } : {}
+  };
+  return {
+    input_tokens: data.inputTokens,
+    output_tokens: data.outputTokens,
+    total_tokens: data.totalTokens,
+    ...Object.keys(inputTokenDetails).length > 0 ? { input_token_details: inputTokenDetails } : {},
+    ...Object.keys(outputTokenDetails).length > 0 ? { output_token_details: outputTokenDetails } : {}
+  };
+}
+__name(createUsageMetadata, "createUsageMetadata");
+function openAIUsageToUsageMetadata(usage) {
+  return createUsageMetadata({
+    inputTokens: usage.prompt_tokens,
+    outputTokens: usage.completion_tokens,
+    totalTokens: usage.total_tokens,
+    inputAudioTokens: usage.prompt_tokens_details?.audio_tokens,
+    outputAudioTokens: usage.completion_tokens_details?.audio_tokens,
+    cacheReadTokens: usage.prompt_tokens_details?.cached_tokens,
+    reasoningTokens: usage.completion_tokens_details?.reasoning_tokens
+  });
+}
+__name(openAIUsageToUsageMetadata, "openAIUsageToUsageMetadata");
+async function langchainMessageToOpenAIMessage(messages, plugin, model, supportImageInputType, removeSystemMessage) {
   const result = [];
   const normalizedModel = model ? normalizeOpenAIModelName(model) : model;
   const isDeepseekThinkModel = normalizedModel?.includes("deepseek-reasoner");
@@ -224,7 +259,7 @@ async function langchainMessageToOpenAIMessage(messages, plugin, model, supportI
     }
     const images = rawMessage.additional_kwargs.images;
     const lowerModel = normalizedModel?.toLowerCase() ?? "";
-    if ((lowerModel?.includes("vision") || lowerModel?.includes("gpt-4o") || lowerModel?.includes("claude") || lowerModel?.includes("gemini") || lowerModel?.includes("qwen-vl") || lowerModel?.includes("omni") || lowerModel?.includes("qwen2.5-vl") || lowerModel?.includes("qwen2.5-omni") || lowerModel?.includes("qwen-omni") || lowerModel?.includes("qwen2-vl") || lowerModel?.includes("qwen3.5") || lowerModel?.includes("qvq") || normalizedModel?.includes("o1") || normalizedModel?.includes("o4") || normalizedModel?.includes("o3") || normalizedModel?.includes("gpt-4.1") || normalizedModel?.includes("gpt-5") || supportImageInput2) && images != null) {
+    if (images != null && (supportImageInput(lowerModel) || supportImageInputType)) {
       msg.content = [
         {
           type: "text",
@@ -376,15 +411,73 @@ async function fetchImageUrl(plugin, content) {
   }
   const ext = url.match(/\.([^.?#]+)(?:[?#]|$)/)?.[1]?.toLowerCase();
   const imageType = (0, import_string.getImageMimeType)(ext);
-  const buffer = await plugin.fetch(url).then((res) => {
-    if (!res.ok) {
-      throw new Error(`Failed to fetch image: ${res.status}`);
-    }
-    return res.arrayBuffer();
-  }).then(Buffer.from);
+  const controller = new AbortController();
+  const timeout = setTimeout(() => controller.abort(), 6e4);
+  const response = await plugin.fetch(url, {
+    signal: controller.signal
+  }).finally(() => {
+    clearTimeout(timeout);
+  });
+  if (!response.ok) {
+    throw new Error(`Failed to fetch image: ${response.status}`);
+  }
+  const buffer = Buffer.from(await response.arrayBuffer());
   return `data:${imageType};base64,${buffer.toString("base64")}`;
 }
 __name(fetchImageUrl, "fetchImageUrl");
+function getFileLikeUrlInfo(content) {
+  switch (content.type) {
+    case "file_url": {
+      const raw = content.file_url;
+      return {
+        url: typeof raw === "string" ? raw : raw.url,
+        mimeType: typeof raw === "string" ? void 0 : raw.mimeType
+      };
+    }
+    case "audio_url": {
+      const raw = content.audio_url;
+      return {
+        url: typeof raw === "string" ? raw : raw.url,
+        mimeType: typeof raw === "string" ? void 0 : raw.mimeType
+      };
+    }
+    case "video_url": {
+      const raw = content.video_url;
+      return {
+        url: typeof raw === "string" ? raw : raw.url,
+        mimeType: typeof raw === "string" ? void 0 : raw.mimeType
+      };
+    }
+  }
+}
+__name(getFileLikeUrlInfo, "getFileLikeUrlInfo");
+async function fetchFileLikeUrl(plugin, content) {
+  const { url, mimeType } = getFileLikeUrlInfo(content);
+  const dataUrlMatch = url.match(/^data:([^;,]+);base64,(.+)$/i);
+  if (dataUrlMatch) {
+    return {
+      buffer: Buffer.from(dataUrlMatch[2], "base64"),
+      mimeType: dataUrlMatch[1] || mimeType || "application/octet-stream"
+    };
+  }
+  const controller = new AbortController();
+  const timeout = setTimeout(() => controller.abort(), 6e4);
+  const response = await plugin.fetch(url, {
+    signal: controller.signal
+  }).finally(() => {
+    clearTimeout(timeout);
+  });
+  if (!response.ok) {
+    throw new Error(`Failed to fetch file: ${response.status}`);
+  }
+  const buffer = Buffer.from(await response.arrayBuffer());
+  const fetchedMimeType = response.headers.get("content-type")?.split(";")[0]?.trim();
+  return {
+    buffer,
+    mimeType: mimeType ?? fetchedMimeType ?? (0, import_string.getMimeTypeFromSource)(url) ?? "application/octet-stream"
+  };
+}
+__name(fetchFileLikeUrl, "fetchFileLikeUrl");
 function messageTypeToOpenAIRole(type) {
   switch (type) {
     case "system":
@@ -478,8 +571,12 @@ function convertMessageToMessageChunk(message) {
     const toolCallChunks = [];
     if (Array.isArray(message.tool_calls)) {
       for (const rawToolCall of message.tool_calls) {
+        let name = rawToolCall.function?.name;
+        if (name != null && name.length < 1) {
+          name = void 0;
+        }
         toolCallChunks.push({
-          name: rawToolCall.function?.name,
+          name,
           args: rawToolCall.function?.arguments,
           id: rawToolCall.id
         });
@@ -518,10 +615,6 @@ function convertDeltaToMessageChunk(delta, defaultRole) {
     additionalKwargs = {
       function_call: delta.function_call
     };
-  } else if (delta.tool_calls) {
-    additionalKwargs = {
-      tool_calls: delta.tool_calls
-    };
   } else {
     additionalKwargs = {};
   }
@@ -534,12 +627,16 @@ function convertDeltaToMessageChunk(delta, defaultRole) {
     const toolCallChunks = [];
     if (Array.isArray(delta.tool_calls)) {
       for (const rawToolCall of delta.tool_calls) {
-        toolCallChunks.push({
+        const toolCall = {
           name: rawToolCall.function?.name,
           args: rawToolCall.function?.arguments,
           id: rawToolCall.id,
           index: rawToolCall.index
-        });
+        };
+        if (toolCall.name != null && toolCall.name.length < 1) {
+          delete toolCall.name;
+        }
+        toolCallChunks.push(toolCall);
       }
     }
     return new import_messages.AIMessageChunk({
@@ -619,25 +716,15 @@ async function buildChatCompletionParams(params, plugin, enableGoogleSearch, sup
   return (0, import_object.deepAssign)({}, base, params.overrideRequestParams ?? {});
 }
 __name(buildChatCompletionParams, "buildChatCompletionParams");
-function processReasoningContent(delta, reasoningState) {
-  if (delta.reasoning_content) {
-    reasoningState.content += delta.reasoning_content;
-    if (reasoningState.time === 0) {
-      reasoningState.time = Date.now();
-    }
-  }
-  if ((delta.reasoning_content == null || delta.reasoning_content === "") && delta.content && delta.content.length > 0 && reasoningState.time > 0 && !reasoningState.isSet) {
-    const reasoningTime = Date.now() - reasoningState.time;
-    reasoningState.time = reasoningTime;
-    reasoningState.isSet = true;
-    return reasoningTime;
-  }
-}
-__name(processReasoningContent, "processReasoningContent");
 async function* processStreamResponse(requestContext, iterator) {
   let defaultRole = "assistant";
   let errorCount = 0;
-  const reasoningState = { content: "", time: 0, isSet: false };
+  const reasoningState = {
+    content: "",
+    startedAt: 0,
+    duration: 0,
+    done: false
+  };
   for await (const event of iterator) {
     const chunk = event.data;
     if (chunk === "[DONE]") break;
@@ -652,16 +739,14 @@ async function* processStreamResponse(requestContext, iterator) {
       }
       const choice = data.choices?.[0];
       if (data.usage) {
+        const usageMetadata = openAIUsageToUsageMetadata(data.usage);
         yield new import_outputs.ChatGenerationChunk({
+          generationInfo: {
+            usage_metadata: usageMetadata
+          },
           message: new import_messages2.AIMessageChunk({
             content: "",
-            response_metadata: {
-              tokenUsage: {
-                promptTokens: data.usage.prompt_tokens,
-                completionTokens: data.usage.completion_tokens,
-                totalTokens: data.usage.total_tokens
-              }
-            }
+            usage_metadata: usageMetadata
           }),
           text: ""
         });
@@ -669,9 +754,16 @@ async function* processStreamResponse(requestContext, iterator) {
       if (!choice) continue;
       const { delta } = choice;
       const messageChunk = convertDeltaToMessageChunk(delta, defaultRole);
-      const reasoningTime = processReasoningContent(delta, reasoningState);
-      if (reasoningTime !== void 0) {
-        messageChunk.additional_kwargs.reasoning_time = reasoningTime;
+      if (delta.reasoning_content) {
+        reasoningState.content += delta.reasoning_content;
+        if (reasoningState.startedAt === 0) {
+          reasoningState.startedAt = Date.now();
+        }
+      }
+      if (!reasoningState.done && reasoningState.startedAt > 0 && ((delta.content?.length ?? 0) > 0 || (delta.tool_calls?.length ?? 0) > 0 || delta.function_call != null)) {
+        reasoningState.duration = Date.now() - reasoningState.startedAt;
+        reasoningState.done = true;
+        messageChunk.additional_kwargs.reasoning_time = reasoningState.duration;
       }
       defaultRole = (delta.role?.length ?? 0) > 0 ? delta.role : defaultRole;
       yield new import_outputs.ChatGenerationChunk({
@@ -679,6 +771,13 @@ async function* processStreamResponse(requestContext, iterator) {
         text: messageChunk.content
       });
     } catch (e) {
+      if (chunk.includes("tool_calls") || chunk.includes("function_call") || chunk.includes("tool_call_id")) {
+        requestContext.modelRequester.logger.error(
+          "error with chunk",
+          chunk
+        );
+        throw new import_error.ChatLunaError(import_error.ChatLunaErrorCode.API_REQUEST_FAILED, e);
+      }
       if (errorCount > 5) {
         requestContext.modelRequester.logger.error(
           "error with chunk",
@@ -690,8 +789,21 @@ async function* processStreamResponse(requestContext, iterator) {
     }
   }
   if (reasoningState.content.length > 0) {
+    if (!reasoningState.done && reasoningState.startedAt > 0) {
+      reasoningState.duration = Date.now() - reasoningState.startedAt;
+      reasoningState.done = true;
+      yield new import_outputs.ChatGenerationChunk({
+        message: new import_messages2.AIMessageChunk({
+          content: "",
+          additional_kwargs: {
+            reasoning_time: reasoningState.duration
+          }
+        }),
+        text: ""
+      });
+    }
     requestContext.modelRequester.logger.debug(
-      `reasoning content: ${reasoningState.content}. Use time: ${reasoningState.time / 1e3}s`
+      `reasoning content: ${reasoningState.content}. Use time: ${reasoningState.duration / 1e3}s`
     );
   }
 }
@@ -726,11 +838,15 @@ async function processResponse(requestContext, response) {
       );
     }
     const messageChunk = convertMessageToMessageChunk(choice.message);
+    const usageMetadata = data.usage ? openAIUsageToUsageMetadata(data.usage) : void 0;
+    if (messageChunk instanceof import_messages2.AIMessageChunk) {
+      messageChunk.usage_metadata = usageMetadata;
+    }
     return new import_outputs.ChatGenerationChunk({
       message: messageChunk,
       text: (0, import_string2.getMessageContent)(messageChunk.content),
-      generationInfo: {
-        tokenUsage: data.usage
+      generationInfo: usageMetadata == null ? void 0 : {
+        usage_metadata: usageMetadata
       }
     });
   } catch (e) {
@@ -898,7 +1014,9 @@ __name(createRequestContext, "createRequestContext");
   convertMessageToMessageChunk,
   createEmbeddings,
   createRequestContext,
+  createUsageMetadata,
   expandReasoningEffortModelVariants,
+  fetchFileLikeUrl,
   fetchImageUrl,
   formatToolToOpenAITool,
   formatToolsToOpenAITools,
@@ -909,9 +1027,9 @@ __name(createRequestContext, "createRequestContext");
   langchainMessageToOpenAIMessage,
   messageTypeToOpenAIRole,
   normalizeOpenAIModelName,
+  openAIUsageToUsageMetadata,
   parseOpenAIModelNameWithReasoningEffort,
   processDeepSeekThinkMessages,
-  processReasoningContent,
   processResponse,
   processStreamResponse,
   reasoningEffortModelSuffixes,

package/lib/index.mjs

@@ -114,6 +114,8 @@ var imageModelMatchers = [
   "qwen*-omni",
   "qwen-omni",
   "qwen*-vl",
+  "qwen-3.5",
+  "qwen3.5",
   "qvq",
   "o1",
   "o3",
@@ -151,10 +153,42 @@ import {
 import { zodToJsonSchema } from "zod-to-json-schema";
 import {
   getImageMimeType,
+  getMimeTypeFromSource,
   isMessageContentImageUrl
 } from "koishi-plugin-chatluna/utils/string";
 import { isZodSchemaV3 } from "@langchain/core/utils/types";
-async function langchainMessageToOpenAIMessage(messages, plugin, model, supportImageInput2, removeSystemMessage) {
+function createUsageMetadata(data) {
+  const inputTokenDetails = {
+    ...data.inputAudioTokens != null ? { audio: data.inputAudioTokens } : {},
+    ...data.cacheReadTokens != null ? { cache_read: data.cacheReadTokens } : {},
+    ...data.cacheCreationTokens != null ? { cache_creation: data.cacheCreationTokens } : {}
+  };
+  const outputTokenDetails = {
+    ...data.outputAudioTokens != null ? { audio: data.outputAudioTokens } : {},
+    ...data.reasoningTokens != null ? { reasoning: data.reasoningTokens } : {}
+  };
+  return {
+    input_tokens: data.inputTokens,
+    output_tokens: data.outputTokens,
+    total_tokens: data.totalTokens,
+    ...Object.keys(inputTokenDetails).length > 0 ? { input_token_details: inputTokenDetails } : {},
+    ...Object.keys(outputTokenDetails).length > 0 ? { output_token_details: outputTokenDetails } : {}
+  };
+}
+__name(createUsageMetadata, "createUsageMetadata");
+function openAIUsageToUsageMetadata(usage) {
+  return createUsageMetadata({
+    inputTokens: usage.prompt_tokens,
+    outputTokens: usage.completion_tokens,
+    totalTokens: usage.total_tokens,
+    inputAudioTokens: usage.prompt_tokens_details?.audio_tokens,
+    outputAudioTokens: usage.completion_tokens_details?.audio_tokens,
+    cacheReadTokens: usage.prompt_tokens_details?.cached_tokens,
+    reasoningTokens: usage.completion_tokens_details?.reasoning_tokens
+  });
+}
+__name(openAIUsageToUsageMetadata, "openAIUsageToUsageMetadata");
+async function langchainMessageToOpenAIMessage(messages, plugin, model, supportImageInputType, removeSystemMessage) {
   const result = [];
   const normalizedModel = model ? normalizeOpenAIModelName(model) : model;
   const isDeepseekThinkModel = normalizedModel?.includes("deepseek-reasoner");
@@ -188,7 +222,7 @@ async function langchainMessageToOpenAIMessage(messages, plugin, model, supportI
     }
     const images = rawMessage.additional_kwargs.images;
     const lowerModel = normalizedModel?.toLowerCase() ?? "";
-    if ((lowerModel?.includes("vision") || lowerModel?.includes("gpt-4o") || lowerModel?.includes("claude") || lowerModel?.includes("gemini") || lowerModel?.includes("qwen-vl") || lowerModel?.includes("omni") || lowerModel?.includes("qwen2.5-vl") || lowerModel?.includes("qwen2.5-omni") || lowerModel?.includes("qwen-omni") || lowerModel?.includes("qwen2-vl") || lowerModel?.includes("qwen3.5") || lowerModel?.includes("qvq") || normalizedModel?.includes("o1") || normalizedModel?.includes("o4") || normalizedModel?.includes("o3") || normalizedModel?.includes("gpt-4.1") || normalizedModel?.includes("gpt-5") || supportImageInput2) && images != null) {
+    if (images != null && (supportImageInput(lowerModel) || supportImageInputType)) {
       msg.content = [
         {
           type: "text",
@@ -340,15 +374,73 @@ async function fetchImageUrl(plugin, content) {
   }
   const ext = url.match(/\.([^.?#]+)(?:[?#]|$)/)?.[1]?.toLowerCase();
   const imageType = getImageMimeType(ext);
-  const buffer = await plugin.fetch(url).then((res) => {
-    if (!res.ok) {
-      throw new Error(`Failed to fetch image: ${res.status}`);
-    }
-    return res.arrayBuffer();
-  }).then(Buffer.from);
+  const controller = new AbortController();
+  const timeout = setTimeout(() => controller.abort(), 6e4);
+  const response = await plugin.fetch(url, {
+    signal: controller.signal
+  }).finally(() => {
+    clearTimeout(timeout);
+  });
+  if (!response.ok) {
+    throw new Error(`Failed to fetch image: ${response.status}`);
+  }
+  const buffer = Buffer.from(await response.arrayBuffer());
   return `data:${imageType};base64,${buffer.toString("base64")}`;
 }
 __name(fetchImageUrl, "fetchImageUrl");
+function getFileLikeUrlInfo(content) {
+  switch (content.type) {
+    case "file_url": {
+      const raw = content.file_url;
+      return {
+        url: typeof raw === "string" ? raw : raw.url,
+        mimeType: typeof raw === "string" ? void 0 : raw.mimeType
+      };
+    }
+    case "audio_url": {
+      const raw = content.audio_url;
+      return {
+        url: typeof raw === "string" ? raw : raw.url,
+        mimeType: typeof raw === "string" ? void 0 : raw.mimeType
+      };
+    }
+    case "video_url": {
+      const raw = content.video_url;
+      return {
+        url: typeof raw === "string" ? raw : raw.url,
+        mimeType: typeof raw === "string" ? void 0 : raw.mimeType
+      };
+    }
+  }
+}
+__name(getFileLikeUrlInfo, "getFileLikeUrlInfo");
+async function fetchFileLikeUrl(plugin, content) {
+  const { url, mimeType } = getFileLikeUrlInfo(content);
+  const dataUrlMatch = url.match(/^data:([^;,]+);base64,(.+)$/i);
+  if (dataUrlMatch) {
+    return {
+      buffer: Buffer.from(dataUrlMatch[2], "base64"),
+      mimeType: dataUrlMatch[1] || mimeType || "application/octet-stream"
+    };
+  }
+  const controller = new AbortController();
+  const timeout = setTimeout(() => controller.abort(), 6e4);
+  const response = await plugin.fetch(url, {
+    signal: controller.signal
+  }).finally(() => {
+    clearTimeout(timeout);
+  });
+  if (!response.ok) {
+    throw new Error(`Failed to fetch file: ${response.status}`);
+  }
+  const buffer = Buffer.from(await response.arrayBuffer());
+  const fetchedMimeType = response.headers.get("content-type")?.split(";")[0]?.trim();
+  return {
+    buffer,
+    mimeType: mimeType ?? fetchedMimeType ?? getMimeTypeFromSource(url) ?? "application/octet-stream"
+  };
+}
+__name(fetchFileLikeUrl, "fetchFileLikeUrl");
 function messageTypeToOpenAIRole(type) {
   switch (type) {
     case "system":
@@ -442,8 +534,12 @@ function convertMessageToMessageChunk(message) {
     const toolCallChunks = [];
     if (Array.isArray(message.tool_calls)) {
       for (const rawToolCall of message.tool_calls) {
+        let name = rawToolCall.function?.name;
+        if (name != null && name.length < 1) {
+          name = void 0;
+        }
         toolCallChunks.push({
-          name: rawToolCall.function?.name,
+          name,
           args: rawToolCall.function?.arguments,
           id: rawToolCall.id
         });
@@ -482,10 +578,6 @@ function convertDeltaToMessageChunk(delta, defaultRole) {
     additionalKwargs = {
       function_call: delta.function_call
     };
-  } else if (delta.tool_calls) {
-    additionalKwargs = {
-      tool_calls: delta.tool_calls
-    };
   } else {
     additionalKwargs = {};
   }
@@ -498,12 +590,16 @@ function convertDeltaToMessageChunk(delta, defaultRole) {
     const toolCallChunks = [];
     if (Array.isArray(delta.tool_calls)) {
       for (const rawToolCall of delta.tool_calls) {
-        toolCallChunks.push({
+        const toolCall = {
           name: rawToolCall.function?.name,
           args: rawToolCall.function?.arguments,
           id: rawToolCall.id,
           index: rawToolCall.index
-        });
+        };
+        if (toolCall.name != null && toolCall.name.length < 1) {
+          delete toolCall.name;
+        }
+        toolCallChunks.push(toolCall);
       }
     }
     return new AIMessageChunk({
@@ -583,25 +679,15 @@ async function buildChatCompletionParams(params, plugin, enableGoogleSearch, sup
   return deepAssign({}, base, params.overrideRequestParams ?? {});
 }
 __name(buildChatCompletionParams, "buildChatCompletionParams");
-function processReasoningContent(delta, reasoningState) {
-  if (delta.reasoning_content) {
-    reasoningState.content += delta.reasoning_content;
-    if (reasoningState.time === 0) {
-      reasoningState.time = Date.now();
-    }
-  }
-  if ((delta.reasoning_content == null || delta.reasoning_content === "") && delta.content && delta.content.length > 0 && reasoningState.time > 0 && !reasoningState.isSet) {
-    const reasoningTime = Date.now() - reasoningState.time;
-    reasoningState.time = reasoningTime;
-    reasoningState.isSet = true;
-    return reasoningTime;
-  }
-}
-__name(processReasoningContent, "processReasoningContent");
 async function* processStreamResponse(requestContext, iterator) {
   let defaultRole = "assistant";
   let errorCount = 0;
-  const reasoningState = { content: "", time: 0, isSet: false };
+  const reasoningState = {
+    content: "",
+    startedAt: 0,
+    duration: 0,
+    done: false
+  };
   for await (const event of iterator) {
     const chunk = event.data;
     if (chunk === "[DONE]") break;
@@ -616,16 +702,14 @@ async function* processStreamResponse(requestContext, iterator) {
       }
       const choice = data.choices?.[0];
       if (data.usage) {
+        const usageMetadata = openAIUsageToUsageMetadata(data.usage);
         yield new ChatGenerationChunk({
+          generationInfo: {
+            usage_metadata: usageMetadata
+          },
           message: new AIMessageChunk2({
             content: "",
-            response_metadata: {
-              tokenUsage: {
-                promptTokens: data.usage.prompt_tokens,
-                completionTokens: data.usage.completion_tokens,
-                totalTokens: data.usage.total_tokens
-              }
-            }
+            usage_metadata: usageMetadata
           }),
           text: ""
         });
@@ -633,9 +717,16 @@ async function* processStreamResponse(requestContext, iterator) {
       if (!choice) continue;
       const { delta } = choice;
       const messageChunk = convertDeltaToMessageChunk(delta, defaultRole);
-      const reasoningTime = processReasoningContent(delta, reasoningState);
-      if (reasoningTime !== void 0) {
-        messageChunk.additional_kwargs.reasoning_time = reasoningTime;
+      if (delta.reasoning_content) {
+        reasoningState.content += delta.reasoning_content;
+        if (reasoningState.startedAt === 0) {
+          reasoningState.startedAt = Date.now();
+        }
+      }
+      if (!reasoningState.done && reasoningState.startedAt > 0 && ((delta.content?.length ?? 0) > 0 || (delta.tool_calls?.length ?? 0) > 0 || delta.function_call != null)) {
+        reasoningState.duration = Date.now() - reasoningState.startedAt;
+        reasoningState.done = true;
+        messageChunk.additional_kwargs.reasoning_time = reasoningState.duration;
       }
       defaultRole = (delta.role?.length ?? 0) > 0 ? delta.role : defaultRole;
       yield new ChatGenerationChunk({
@@ -643,6 +734,13 @@ async function* processStreamResponse(requestContext, iterator) {
         text: messageChunk.content
       });
     } catch (e) {
+      if (chunk.includes("tool_calls") || chunk.includes("function_call") || chunk.includes("tool_call_id")) {
+        requestContext.modelRequester.logger.error(
+          "error with chunk",
+          chunk
+        );
+        throw new ChatLunaError(ChatLunaErrorCode.API_REQUEST_FAILED, e);
+      }
       if (errorCount > 5) {
         requestContext.modelRequester.logger.error(
           "error with chunk",
@@ -654,8 +752,21 @@ async function* processStreamResponse(requestContext, iterator) {
     }
   }
   if (reasoningState.content.length > 0) {
+    if (!reasoningState.done && reasoningState.startedAt > 0) {
+      reasoningState.duration = Date.now() - reasoningState.startedAt;
+      reasoningState.done = true;
+      yield new ChatGenerationChunk({
+        message: new AIMessageChunk2({
+          content: "",
+          additional_kwargs: {
+            reasoning_time: reasoningState.duration
+          }
+        }),
+        text: ""
+      });
+    }
     requestContext.modelRequester.logger.debug(
-      `reasoning content: ${reasoningState.content}. Use time: ${reasoningState.time / 1e3}s`
+      `reasoning content: ${reasoningState.content}. Use time: ${reasoningState.duration / 1e3}s`
     );
   }
 }
@@ -690,11 +801,15 @@ async function processResponse(requestContext, response) {
       );
     }
     const messageChunk = convertMessageToMessageChunk(choice.message);
+    const usageMetadata = data.usage ? openAIUsageToUsageMetadata(data.usage) : void 0;
+    if (messageChunk instanceof AIMessageChunk2) {
+      messageChunk.usage_metadata = usageMetadata;
+    }
     return new ChatGenerationChunk({
       message: messageChunk,
       text: getMessageContent(messageChunk.content),
-      generationInfo: {
-        tokenUsage: data.usage
+      generationInfo: usageMetadata == null ? void 0 : {
+        usage_metadata: usageMetadata
       }
     });
   } catch (e) {
@@ -861,7 +976,9 @@ export {
   convertMessageToMessageChunk,
   createEmbeddings,
   createRequestContext,
+  createUsageMetadata,
   expandReasoningEffortModelVariants,
+  fetchFileLikeUrl,
   fetchImageUrl,
   formatToolToOpenAITool,
   formatToolsToOpenAITools,
@@ -872,9 +989,9 @@ export {
   langchainMessageToOpenAIMessage,
   messageTypeToOpenAIRole,
   normalizeOpenAIModelName,
+  openAIUsageToUsageMetadata,
   parseOpenAIModelNameWithReasoningEffort,
   processDeepSeekThinkMessages,
-  processReasoningContent,
   processResponse,
   processStreamResponse,
   reasoningEffortModelSuffixes,

package/lib/requester.d.ts

@@ -37,14 +37,6 @@ export declare function buildChatCompletionParams(params: ModelRequestParams, pl
         include_usage: boolean;
     };
 } & Record<string, any>>;
-export declare function processReasoningContent(delta: {
-    reasoning_content?: string;
-    content?: string;
-}, reasoningState: {
-    content: string;
-    time: number;
-    isSet: boolean;
-}): number;
 export declare function processStreamResponse<T extends ClientConfig, R extends ChatLunaPlugin.Config>(requestContext: RequestContext<T, R>, iterator: AsyncGenerator<SSEEvent, string, unknown>): AsyncGenerator<ChatGenerationChunk, void, unknown>;
 export declare function processResponse<T extends ClientConfig, R extends ChatLunaPlugin.Config>(requestContext: RequestContext<T, R>, response: Response): Promise<ChatGenerationChunk>;
 export declare function completionStream<T extends ClientConfig, R extends ChatLunaPlugin.Config>(requestContext: RequestContext<T, R>, params: ModelRequestParams, completionUrl?: string, enableGoogleSearch?: boolean, supportImageInput?: boolean): AsyncGenerator<ChatGenerationChunk>;

package/lib/types.d.ts

@@ -7,6 +7,7 @@ export interface ChatCompletionResponse {
             role?: string;
             reasoning_content?: string;
             function_call?: ChatCompletionRequestMessageToolCall;
+            tool_calls?: ChatCompletionRequestMessageToolCall[];
         };
         message: ChatCompletionResponseMessage;
     }[];
@@ -14,11 +15,24 @@ export interface ChatCompletionResponse {
     object: string;
     created: number;
     model: string;
-    usage: {
-        prompt_tokens: number;
-        completion_tokens: number;
-        total_tokens: number;
-    };
+    usage?: ChatCompletionUsage;
+}
+export interface ChatCompletionPromptTokensDetails {
+    audio_tokens?: number;
+    cached_tokens?: number;
+}
+export interface ChatCompletionCompletionTokensDetails {
+    reasoning_tokens?: number;
+    audio_tokens?: number;
+    accepted_prediction_tokens?: number;
+    rejected_prediction_tokens?: number;
+}
+export interface ChatCompletionUsage {
+    prompt_tokens: number;
+    completion_tokens: number;
+    total_tokens: number;
+    prompt_tokens_details?: ChatCompletionPromptTokensDetails;
+    completion_tokens_details?: ChatCompletionCompletionTokensDetails;
 }
 export interface ChatCompletionTextPart {
     type: 'text';

package/lib/utils.d.ts

@@ -1,15 +1,54 @@
-import { AIMessageChunk, BaseMessage, ChatMessageChunk, FunctionMessageChunk, HumanMessageChunk, MessageContentImageUrl, MessageType, SystemMessageChunk, ToolMessageChunk } from '@langchain/core/messages';
+import { AIMessageChunk, BaseMessage, ChatMessageChunk, FunctionMessageChunk, HumanMessageChunk, MessageContentComplex, MessageContentImageUrl, MessageType, SystemMessageChunk, type UsageMetadata, ToolMessageChunk } from '@langchain/core/messages';
 import { StructuredTool } from '@langchain/core/tools';
 import { JsonSchema7Type } from 'zod-to-json-schema';
-import { ChatCompletionResponseMessage, ChatCompletionResponseMessageRoleEnum, ChatCompletionTool } from './types';
+import { ChatCompletionUsage, ChatCompletionResponseMessage, ChatCompletionResponseMessageRoleEnum, ChatCompletionTool } from './types';
 import { ChatLunaPlugin } from 'koishi-plugin-chatluna/services/chat';
-export declare function langchainMessageToOpenAIMessage(messages: BaseMessage[], plugin: ChatLunaPlugin, model?: string, supportImageInput?: boolean, removeSystemMessage?: boolean): Promise<ChatCompletionResponseMessage[]>;
+export declare function createUsageMetadata(data: {
+    inputTokens: number;
+    outputTokens: number;
+    totalTokens: number;
+    inputAudioTokens?: number;
+    outputAudioTokens?: number;
+    cacheReadTokens?: number;
+    cacheCreationTokens?: number;
+    reasoningTokens?: number;
+}): UsageMetadata;
+export declare function openAIUsageToUsageMetadata(usage: ChatCompletionUsage): UsageMetadata;
+export declare function langchainMessageToOpenAIMessage(messages: BaseMessage[], plugin: ChatLunaPlugin, model?: string, supportImageInputType?: boolean, removeSystemMessage?: boolean): Promise<ChatCompletionResponseMessage[]>;
 export declare function processDeepSeekThinkMessages(convertedMessages: ChatCompletionResponseMessage[], originalMessages: BaseMessage[]): ChatCompletionResponseMessage[];
 export declare function transformSystemMessages(messages: ChatCompletionResponseMessage[]): ChatCompletionResponseMessage[];
 export declare function fetchImageUrl(plugin: ChatLunaPlugin, content: MessageContentImageUrl): Promise<string>;
+type MessageContentFileLike = MessageContentComplex & ({
+    type: 'file_url';
+    file_url: string | {
+        url: string;
+        mimeType?: string;
+    };
+} | {
+    type: 'audio_url';
+    audio_url: string | {
+        url: string;
+        mimeType?: string;
+    };
+} | {
+    type: 'video_url';
+    video_url: string | {
+        url: string;
+        mimeType?: string;
+    };
+});
+/**
+ * Fetch file/audio/video content and return decoded bytes.
+ * If the source is a base64 data URL, it is decoded directly.
+ */
+export declare function fetchFileLikeUrl(plugin: ChatLunaPlugin, content: MessageContentFileLike): Promise<{
+    buffer: Buffer<ArrayBuffer>;
+    mimeType: string;
+}>;
 export declare function messageTypeToOpenAIRole(type: MessageType): ChatCompletionResponseMessageRoleEnum;
 export declare function formatToolsToOpenAITools(tools: StructuredTool[], includeGoogleSearch: boolean): ChatCompletionTool[];
 export declare function formatToolToOpenAITool(tool: StructuredTool): ChatCompletionTool;
 export declare function removeAdditionalProperties(schema: JsonSchema7Type): JsonSchema7Type;
 export declare function convertMessageToMessageChunk(message: ChatCompletionResponseMessage): HumanMessageChunk | AIMessageChunk | SystemMessageChunk | FunctionMessageChunk | ToolMessageChunk | ChatMessageChunk;
 export declare function convertDeltaToMessageChunk(delta: Record<string, any>, defaultRole?: ChatCompletionResponseMessageRoleEnum): HumanMessageChunk | AIMessageChunk | SystemMessageChunk | FunctionMessageChunk | ToolMessageChunk | ChatMessageChunk;
+export {};

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@chatluna/v1-shared-adapter",
   "description": "chatluna shared adapter",
-  "version": "1.0.27",
+  "version": "1.0.29",
   "main": "lib/index.cjs",
   "module": "lib/index.mjs",
   "typings": "lib/index.d.ts",
@@ -70,6 +70,6 @@
   },
   "peerDependencies": {
     "koishi": "^4.18.9",
-    "koishi-plugin-chatluna": "^1.3.22"
+    "koishi-plugin-chatluna": "^1.3.33"
   }
 }