@chatluna/v1-shared-adapter 1.0.34 → 1.0.36

package/lib/client.d.ts

@@ -10,5 +10,6 @@ export declare function normalizeOpenAIModelName(modelName: string): string;
 export declare function isEmbeddingModel(modelName: string): boolean;
 export declare function isNonLLMModel(modelName: string): boolean;
 export declare function isImageGenerationModel(modelName: string): boolean;
+export declare function isRerankerModel(modelName: string): boolean;
 export declare function getModelMaxContextSize(info: ModelInfo): number;
 export declare function supportImageInput(modelName: string): boolean;

package/lib/index.cjs

@@ -28,6 +28,7 @@ __export(index_exports, {
   convertMessageToMessageChunk: () => convertMessageToMessageChunk,
   createEmbeddings: () => createEmbeddings,
   createRequestContext: () => createRequestContext,
+  createRerank: () => createRerank,
   createUsageMetadata: () => createUsageMetadata,
   expandReasoningEffortModelVariants: () => expandReasoningEffortModelVariants,
   fetchFileLikeUrl: () => fetchFileLikeUrl,
@@ -40,6 +41,7 @@ __export(index_exports, {
   isEmbeddingModel: () => isEmbeddingModel,
   isImageGenerationModel: () => isImageGenerationModel,
   isNonLLMModel: () => isNonLLMModel,
+  isRerankerModel: () => isRerankerModel,
   langchainMessageToOpenAIMessage: () => langchainMessageToOpenAIMessage,
   langchainMessageToResponseInput: () => langchainMessageToResponseInput,
   messageTypeToOpenAIRole: () => messageTypeToOpenAIRole,
@@ -118,7 +120,7 @@ function isNonLLMModel(modelName) {
   if (modelName.includes("gemini") && modelName.includes("image")) {
     return false;
   }
-  return ["whisper", "tts", "dall-e", "image", "rerank"].some(
+  return ["whisper", "tts", "dall-e", "image"].some(
     (keyword) => modelName.includes(keyword)
   );
 }
@@ -127,6 +129,10 @@ function isImageGenerationModel(modelName) {
   return isNonLLMModel(modelName) && ["dall-e", "image"].some((keyword) => modelName.includes(keyword));
 }
 __name(isImageGenerationModel, "isImageGenerationModel");
+function isRerankerModel(modelName) {
+  return modelName.includes("rerank");
+}
+__name(isRerankerModel, "isRerankerModel");
 function getModelMaxContextSize(info) {
   const maxTokens = info.maxTokens;
   if (maxTokens != null) {
@@ -214,6 +220,7 @@ var import_sse = require("koishi-plugin-chatluna/utils/sse");
 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_langchain = require("koishi-plugin-chatluna/utils/langchain");
 var import_types = require("@langchain/core/utils/types");
 function createUsageMetadata(data) {
   const inputTokenDetails = {
@@ -510,22 +517,34 @@ function processInterleavedThinkMessages(convertedMessages, originalMessages) {
   if (originalMessages.length === 0) {
     return convertedMessages;
   }
+  const hasToolCallRound = convertedMessages.some(
+    (message) => message.role === "assistant" && (message.tool_calls?.length ?? 0) > 0
+  );
   let lastTurnStartIndex = -1;
   for (let i = originalMessages.length - 1; i >= 0; i--) {
     const message = originalMessages[i];
-    if (message.getType() === "human") {
+    if ((0, import_langchain.isChatLunaUserMessage)(message)) {
       lastTurnStartIndex = i;
       break;
     }
   }
+  if (lastTurnStartIndex === -1) {
+    for (let i = originalMessages.length - 1; i >= 0; i--) {
+      const message = originalMessages[i];
+      if (message.getType() === "human") {
+        lastTurnStartIndex = i;
+        break;
+      }
+    }
+  }
   if (lastTurnStartIndex === -1) {
     lastTurnStartIndex = 0;
   }
   return convertedMessages.map((message, index) => {
-    if (index >= lastTurnStartIndex) {
+    if (hasToolCallRound || index >= lastTurnStartIndex) {
       const originalMessage = originalMessages[index];
       const reasoningContent = originalMessage?.additional_kwargs?.reasoning_content;
-      if (reasoningContent) {
+      if (reasoningContent != null) {
         return {
           ...message,
           reasoning_content: reasoningContent
@@ -716,10 +735,10 @@ function removeAdditionalProperties(schema) {
 __name(removeAdditionalProperties, "removeAdditionalProperties");
 function convertMessageToMessageChunk(message) {
   const content = message.content ?? "";
-  const reasoningContent = message.reasoning_content ?? "";
+  const reasoningContent = message.reasoning_content;
   const role = ((message.role?.length ?? 0) > 0 ? message.role : "assistant").toLowerCase();
   const additionalKwargs = {};
-  if (reasoningContent.length > 0) {
+  if (reasoningContent != null) {
     additionalKwargs.reasoning_content = reasoningContent;
   }
   if (role === "user") {
@@ -766,7 +785,7 @@ __name(convertMessageToMessageChunk, "convertMessageToMessageChunk");
 function convertDeltaToMessageChunk(delta, defaultRole) {
   const role = ((delta.role?.length ?? 0) > 0 ? delta.role : defaultRole).toLowerCase();
   const content = delta.content ?? "";
-  const reasoningContent = delta.reasoning_content ?? "";
+  const reasoningContent = delta.reasoning_content;
   let additionalKwargs;
   if (delta.function_call) {
     additionalKwargs = {
@@ -775,7 +794,7 @@ function convertDeltaToMessageChunk(delta, defaultRole) {
   } else {
     additionalKwargs = {};
   }
-  if (reasoningContent.length > 0) {
+  if (reasoningContent != null) {
     additionalKwargs.reasoning_content = reasoningContent;
   }
   if (role === "user") {
@@ -909,6 +928,7 @@ async function* processStreamResponse(requestContext, iterator) {
   let errorCount = 0;
   const reasoningState = {
     content: "",
+    seen: false,
     startedAt: Date.now(),
     endedAt: void 0
   };
@@ -962,8 +982,11 @@ async function* processStreamResponse(requestContext, iterator) {
       if (reasoningState.endedAt == null && hasResult) {
         reasoningState.endedAt = Date.now();
       }
-      if (reasoningState.endedAt == null && !hasResult && delta.reasoning_content) {
-        reasoningState.content += delta.reasoning_content;
+      if (Object.hasOwn(delta, "reasoning_content")) {
+        reasoningState.seen = true;
+        if (reasoningState.endedAt == null && !hasResult && typeof delta.reasoning_content === "string") {
+          reasoningState.content += delta.reasoning_content;
+        }
       }
       const messageChunk = convertDeltaToMessageChunk(
         {
@@ -1000,12 +1023,14 @@ async function* processStreamResponse(requestContext, iterator) {
       errorCount++;
     }
   }
-  if (reasoningState.content.length > 0) {
+  if (reasoningState.seen || reasoningState.content.length > 0) {
     const reasoningTime = (reasoningState.endedAt ?? Date.now()) - reasoningState.startedAt;
     yield new import_outputs.ChatGenerationChunk({
       message: new import_messages2.AIMessageChunk({
         content: "",
         additional_kwargs: {
+          // Always emit the field (possibly "") so DeepSeek-V4
+          // thinking mode receives reasoning_content back verbatim.
           reasoning_content: reasoningState.content,
           ...reasoningTime != null ? { reasoning_time: reasoningTime } : {}
         }
@@ -1405,6 +1430,42 @@ async function createEmbeddings(requestContext, params, embeddingUrl = "embeddin
   }
 }
 __name(createEmbeddings, "createEmbeddings");
+async function createRerank(requestContext, params, rerankUrl = "rerank") {
+  const { modelRequester } = requestContext;
+  try {
+    const response = await modelRequester.post(
+      rerankUrl,
+      {
+        model: params.model,
+        query: params.query,
+        documents: params.documents,
+        top_n: params.topN,
+        max_chunks_per_doc: params.maxChunksPerDoc,
+        return_documents: false
+      },
+      {
+        signal: params.signal
+      }
+    );
+    const data = await response.json();
+    if (data.results == null) {
+      throw new import_error.ChatLunaError(
+        import_error.ChatLunaErrorCode.API_REQUEST_FAILED,
+        new Error(`Call Rerank Error: ${JSON.stringify(data)}`)
+      );
+    }
+    return data.results.map((item) => ({
+      index: item.index,
+      relevanceScore: item.relevance_score
+    }));
+  } catch (e) {
+    if (e instanceof import_error.ChatLunaError) {
+      throw e;
+    }
+    throw new import_error.ChatLunaError(import_error.ChatLunaErrorCode.API_REQUEST_FAILED, e);
+  }
+}
+__name(createRerank, "createRerank");
 async function getModels(requestContext, config) {
   const { modelRequester } = requestContext;
   let data;
@@ -1467,6 +1528,7 @@ __name(createRequestContext, "createRequestContext");
   convertMessageToMessageChunk,
   createEmbeddings,
   createRequestContext,
+  createRerank,
   createUsageMetadata,
   expandReasoningEffortModelVariants,
   fetchFileLikeUrl,
@@ -1479,6 +1541,7 @@ __name(createRequestContext, "createRequestContext");
   isEmbeddingModel,
   isImageGenerationModel,
   isNonLLMModel,
+  isRerankerModel,
   langchainMessageToOpenAIMessage,
   langchainMessageToResponseInput,
   messageTypeToOpenAIRole,

package/lib/index.mjs

@@ -53,7 +53,7 @@ function isNonLLMModel(modelName) {
   if (modelName.includes("gemini") && modelName.includes("image")) {
     return false;
   }
-  return ["whisper", "tts", "dall-e", "image", "rerank"].some(
+  return ["whisper", "tts", "dall-e", "image"].some(
     (keyword) => modelName.includes(keyword)
   );
 }
@@ -62,6 +62,10 @@ function isImageGenerationModel(modelName) {
   return isNonLLMModel(modelName) && ["dall-e", "image"].some((keyword) => modelName.includes(keyword));
 }
 __name(isImageGenerationModel, "isImageGenerationModel");
+function isRerankerModel(modelName) {
+  return modelName.includes("rerank");
+}
+__name(isRerankerModel, "isRerankerModel");
 function getModelMaxContextSize(info) {
   const maxTokens = info.maxTokens;
   if (maxTokens != null) {
@@ -163,6 +167,7 @@ import {
   getMimeTypeFromSource,
   isMessageContentImageUrl
 } from "koishi-plugin-chatluna/utils/string";
+import { isChatLunaUserMessage } from "koishi-plugin-chatluna/utils/langchain";
 import { isZodSchemaV3 } from "@langchain/core/utils/types";
 function createUsageMetadata(data) {
   const inputTokenDetails = {
@@ -459,22 +464,34 @@ function processInterleavedThinkMessages(convertedMessages, originalMessages) {
   if (originalMessages.length === 0) {
     return convertedMessages;
   }
+  const hasToolCallRound = convertedMessages.some(
+    (message) => message.role === "assistant" && (message.tool_calls?.length ?? 0) > 0
+  );
   let lastTurnStartIndex = -1;
   for (let i = originalMessages.length - 1; i >= 0; i--) {
     const message = originalMessages[i];
-    if (message.getType() === "human") {
+    if (isChatLunaUserMessage(message)) {
       lastTurnStartIndex = i;
       break;
     }
   }
+  if (lastTurnStartIndex === -1) {
+    for (let i = originalMessages.length - 1; i >= 0; i--) {
+      const message = originalMessages[i];
+      if (message.getType() === "human") {
+        lastTurnStartIndex = i;
+        break;
+      }
+    }
+  }
   if (lastTurnStartIndex === -1) {
     lastTurnStartIndex = 0;
   }
   return convertedMessages.map((message, index) => {
-    if (index >= lastTurnStartIndex) {
+    if (hasToolCallRound || index >= lastTurnStartIndex) {
       const originalMessage = originalMessages[index];
       const reasoningContent = originalMessage?.additional_kwargs?.reasoning_content;
-      if (reasoningContent) {
+      if (reasoningContent != null) {
         return {
           ...message,
           reasoning_content: reasoningContent
@@ -665,10 +682,10 @@ function removeAdditionalProperties(schema) {
 __name(removeAdditionalProperties, "removeAdditionalProperties");
 function convertMessageToMessageChunk(message) {
   const content = message.content ?? "";
-  const reasoningContent = message.reasoning_content ?? "";
+  const reasoningContent = message.reasoning_content;
   const role = ((message.role?.length ?? 0) > 0 ? message.role : "assistant").toLowerCase();
   const additionalKwargs = {};
-  if (reasoningContent.length > 0) {
+  if (reasoningContent != null) {
     additionalKwargs.reasoning_content = reasoningContent;
   }
   if (role === "user") {
@@ -715,7 +732,7 @@ __name(convertMessageToMessageChunk, "convertMessageToMessageChunk");
 function convertDeltaToMessageChunk(delta, defaultRole) {
   const role = ((delta.role?.length ?? 0) > 0 ? delta.role : defaultRole).toLowerCase();
   const content = delta.content ?? "";
-  const reasoningContent = delta.reasoning_content ?? "";
+  const reasoningContent = delta.reasoning_content;
   let additionalKwargs;
   if (delta.function_call) {
     additionalKwargs = {
@@ -724,7 +741,7 @@ function convertDeltaToMessageChunk(delta, defaultRole) {
   } else {
     additionalKwargs = {};
   }
-  if (reasoningContent.length > 0) {
+  if (reasoningContent != null) {
     additionalKwargs.reasoning_content = reasoningContent;
   }
   if (role === "user") {
@@ -858,6 +875,7 @@ async function* processStreamResponse(requestContext, iterator) {
   let errorCount = 0;
   const reasoningState = {
     content: "",
+    seen: false,
     startedAt: Date.now(),
     endedAt: void 0
   };
@@ -911,8 +929,11 @@ async function* processStreamResponse(requestContext, iterator) {
       if (reasoningState.endedAt == null && hasResult) {
         reasoningState.endedAt = Date.now();
       }
-      if (reasoningState.endedAt == null && !hasResult && delta.reasoning_content) {
-        reasoningState.content += delta.reasoning_content;
+      if (Object.hasOwn(delta, "reasoning_content")) {
+        reasoningState.seen = true;
+        if (reasoningState.endedAt == null && !hasResult && typeof delta.reasoning_content === "string") {
+          reasoningState.content += delta.reasoning_content;
+        }
       }
       const messageChunk = convertDeltaToMessageChunk(
         {
@@ -949,12 +970,14 @@ async function* processStreamResponse(requestContext, iterator) {
       errorCount++;
     }
   }
-  if (reasoningState.content.length > 0) {
+  if (reasoningState.seen || reasoningState.content.length > 0) {
     const reasoningTime = (reasoningState.endedAt ?? Date.now()) - reasoningState.startedAt;
     yield new ChatGenerationChunk({
       message: new AIMessageChunk2({
         content: "",
         additional_kwargs: {
+          // Always emit the field (possibly "") so DeepSeek-V4
+          // thinking mode receives reasoning_content back verbatim.
           reasoning_content: reasoningState.content,
           ...reasoningTime != null ? { reasoning_time: reasoningTime } : {}
         }
@@ -1354,6 +1377,42 @@ async function createEmbeddings(requestContext, params, embeddingUrl = "embeddin
   }
 }
 __name(createEmbeddings, "createEmbeddings");
+async function createRerank(requestContext, params, rerankUrl = "rerank") {
+  const { modelRequester } = requestContext;
+  try {
+    const response = await modelRequester.post(
+      rerankUrl,
+      {
+        model: params.model,
+        query: params.query,
+        documents: params.documents,
+        top_n: params.topN,
+        max_chunks_per_doc: params.maxChunksPerDoc,
+        return_documents: false
+      },
+      {
+        signal: params.signal
+      }
+    );
+    const data = await response.json();
+    if (data.results == null) {
+      throw new ChatLunaError(
+        ChatLunaErrorCode.API_REQUEST_FAILED,
+        new Error(`Call Rerank Error: ${JSON.stringify(data)}`)
+      );
+    }
+    return data.results.map((item) => ({
+      index: item.index,
+      relevanceScore: item.relevance_score
+    }));
+  } catch (e) {
+    if (e instanceof ChatLunaError) {
+      throw e;
+    }
+    throw new ChatLunaError(ChatLunaErrorCode.API_REQUEST_FAILED, e);
+  }
+}
+__name(createRerank, "createRerank");
 async function getModels(requestContext, config) {
   const { modelRequester } = requestContext;
   let data;
@@ -1415,6 +1474,7 @@ export {
   convertMessageToMessageChunk,
   createEmbeddings,
   createRequestContext,
+  createRerank,
   createUsageMetadata,
   expandReasoningEffortModelVariants,
   fetchFileLikeUrl,
@@ -1427,6 +1487,7 @@ export {
   isEmbeddingModel,
   isImageGenerationModel,
   isNonLLMModel,
+  isRerankerModel,
   langchainMessageToOpenAIMessage,
   langchainMessageToResponseInput,
   messageTypeToOpenAIRole,

package/lib/requester.d.ts

@@ -1,5 +1,5 @@
 import { ChatGenerationChunk } from '@langchain/core/outputs';
-import { EmbeddingsRequestParams, ModelRequester, ModelRequestParams } from 'koishi-plugin-chatluna/llm-core/platform/api';
+import { EmbeddingsRequestParams, ModelRequester, ModelRequestParams, RerankerRequestParams, RerankerResult } from 'koishi-plugin-chatluna/llm-core/platform/api';
 import { ClientConfig } from 'koishi-plugin-chatluna/llm-core/platform/config';
 import { SSEEvent } from 'koishi-plugin-chatluna/utils/sse';
 import { type ResponseBuiltinTool, ResponseObject, ResponseOutputItem } from './types';
@@ -54,7 +54,7 @@ export declare function buildResponseParams(params: ModelRequestParams, plugin:
     top_p: number;
     prompt_cache_key: string;
     reasoning: {
-        effort: "minimal" | "low" | "medium" | "high" | "max" | "xhigh";
+        effort: "low" | "high" | "minimal" | "medium" | "max" | "xhigh";
     };
     stream: boolean;
     stream_options: {
@@ -73,6 +73,7 @@ export declare function completion<T extends ClientConfig, R extends ChatLunaPlu
 export declare function responseApiCompletionStream<T extends ClientConfig, R extends ChatLunaPlugin.Config>(requestContext: RequestContext<T, R>, params: ModelRequestParams, opts?: ResponseToolOptions, supportImageInput?: boolean, imageProvider?: ResponseImageProvider): AsyncGenerator<ChatGenerationChunk>;
 export declare function responseApiCompletion<T extends ClientConfig, R extends ChatLunaPlugin.Config>(requestContext: RequestContext<T, R>, params: ModelRequestParams, opts?: ResponseToolOptions, supportImageInput?: boolean, imageProvider?: ResponseImageProvider): Promise<ChatGenerationChunk>;
 export declare function createEmbeddings<T extends ClientConfig, R extends ChatLunaPlugin.Config>(requestContext: RequestContext<T, R>, params: EmbeddingsRequestParams, embeddingUrl?: string): Promise<number[] | number[][]>;
+export declare function createRerank<T extends ClientConfig, R extends ChatLunaPlugin.Config>(requestContext: RequestContext<T, R>, params: RerankerRequestParams, rerankUrl?: string): Promise<RerankerResult[]>;
 export declare function getModels<T extends ClientConfig, R extends ChatLunaPlugin.Config>(requestContext: RequestContext<T, R>, config?: RunnableConfig): Promise<string[]>;
 export declare function createRequestContext<T extends ClientConfig, R extends ChatLunaPlugin.Config>(ctx: Context, config: T, pluginConfig: R, plugin: ChatLunaPlugin, modelRequester: ModelRequester<T, R>): RequestContext<T, R>;
 export {};

package/lib/types.d.ts

@@ -325,3 +325,27 @@ export interface CreateEmbeddingResponseUsage {
     total_tokens: number;
 }
 export type ChatCompletionResponseMessageRoleEnum = 'system' | 'assistant' | 'user' | 'function' | 'tool';
+export interface CreateRerankRequest {
+    model: string;
+    query: string;
+    documents: string[];
+    top_n?: number;
+    max_chunks_per_doc?: number;
+    return_documents?: boolean;
+}
+export interface RerankResultItem {
+    index: number;
+    relevance_score: number;
+    document?: {
+        text: string;
+    };
+}
+export interface CreateRerankResponse {
+    id?: string;
+    model?: string;
+    results: RerankResultItem[];
+    usage?: {
+        prompt_tokens?: number;
+        total_tokens?: number;
+    };
+}

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@chatluna/v1-shared-adapter",
   "description": "chatluna shared adapter",
-  "version": "1.0.34",
+  "version": "1.0.36",
   "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": "^1.4.0-alpha.9"
+    "koishi-plugin-chatluna": "^1.4.0-alpha.13"
   }
 }