@lobehub/chat 1.133.2 → 1.133.4

package/packages/model-bank/src/standard-parameters/index.ts

@@ -1,3 +1,4 @@
+/* eslint-disable sort-keys-fix/sort-keys-fix, typescript-sort-keys/interface */
 import type { Simplify } from 'type-fest';
 import { z } from 'zod';
 
@@ -53,36 +54,14 @@ export const CHAT_MODEL_IMAGE_GENERATION_PARAMS: ModelParamsSchema = {
 
 // 定义顶层的元规范 - 平铺结构
 export const ModelParamsMetaSchema = z.object({
-  aspectRatio: z
-    .object({
-      default: z.string(),
-      description: z.string().optional(),
-      enum: z.array(z.string()),
-      type: z.literal('string').optional(),
-    })
-    .optional(),
-
-  cfg: z
-    .object({
-      default: z.number(),
-      description: z.string().optional(),
-      max: z.number(),
-      min: z.number(),
-      step: z.number(),
-      type: z.literal('number').optional(),
-    })
-    .optional(),
-
-  height: z
-    .object({
-      default: z.number(),
-      description: z.string().optional(),
-      max: z.number(),
-      min: z.number(),
-      step: z.number().optional().default(1),
-      type: z.literal('number').optional(),
-    })
-    .optional(),
+  /**
+   * Prompt 是唯一一个每个模型都有的参数
+   */
+  prompt: z.object({
+    default: z.string().optional().default(''),
+    description: z.string().optional(),
+    type: z.literal('string').optional(),
+  }),
 
   imageUrl: z
     .object({
@@ -106,22 +85,25 @@ export const ModelParamsMetaSchema = z.object({
     })
     .optional(),
 
-  /**
-   * Prompt 是唯一一个每个模型都有的参数
-   */
-  prompt: z.object({
-    default: z.string().optional().default(''),
-    description: z.string().optional(),
-    type: z.literal('string').optional(),
-  }),
+  width: z
+    .object({
+      default: z.number(),
+      description: z.string().optional(),
+      max: z.number(),
+      min: z.number(),
+      step: z.number().optional().default(1),
+      type: z.literal('number').optional(),
+    })
+    .optional(),
 
-  seed: z
+  height: z
     .object({
-      default: z.number().nullable().default(null),
+      default: z.number(),
       description: z.string().optional(),
-      max: z.number().optional().default(MAX_SEED),
-      min: z.number().optional().default(0),
-      type: z.tuple([z.literal('number'), z.literal('null')]).optional(),
+      max: z.number(),
+      min: z.number(),
+      step: z.number().optional().default(1),
+      type: z.literal('number').optional(),
     })
     .optional(),
 
@@ -134,18 +116,27 @@ export const ModelParamsMetaSchema = z.object({
     })
     .optional(),
 
-  steps: z
+  aspectRatio: z
+    .object({
+      default: z.string(),
+      description: z.string().optional(),
+      enum: z.array(z.string()),
+      type: z.literal('string').optional(),
+    })
+    .optional(),
+
+  cfg: z
     .object({
       default: z.number(),
       description: z.string().optional(),
       max: z.number(),
       min: z.number(),
-      step: z.number().optional().default(1),
+      step: z.number(),
       type: z.literal('number').optional(),
     })
     .optional(),
 
-  width: z
+  steps: z
     .object({
       default: z.number(),
       description: z.string().optional(),
@@ -155,6 +146,25 @@ export const ModelParamsMetaSchema = z.object({
       type: z.literal('number').optional(),
     })
     .optional(),
+
+  quality: z
+    .object({
+      default: z.string(),
+      description: z.string().optional(),
+      enum: z.array(z.string()),
+      type: z.literal('string').optional(),
+    })
+    .optional(),
+
+  seed: z
+    .object({
+      default: z.number().nullable().default(null),
+      description: z.string().optional(),
+      max: z.number().optional().default(MAX_SEED),
+      min: z.number().optional().default(0),
+      type: z.tuple([z.literal('number'), z.literal('null')]).optional(),
+    })
+    .optional(),
 });
 // 导出推断出的类型，供定义对象使用
 export type ModelParamsSchema = z.input<typeof ModelParamsMetaSchema>;

package/packages/model-runtime/package.json

@@ -13,6 +13,7 @@
   },
   "dependencies": {
     "@aws-sdk/client-bedrock-runtime": "^3.862.0",
+    "@lobechat/const": "workspace:*",
     "@lobechat/types": "workspace:*",
     "@lobechat/utils": "workspace:*",
     "debug": "^4.4.1",

package/packages/model-runtime/src/core/RouterRuntime/createRuntime.ts

@@ -129,7 +129,7 @@ export const createRouterRuntime = ({
   ...params
 }: CreateRouterRuntimeOptions) => {
   return class UniformRuntime implements LobeRuntimeAI {
-    private _options: ClientOptions & Record<string, any>;
+    public _options: ClientOptions & Record<string, any>;
     private _routers: Routers;
     private _params: any;
     private _id: string;
@@ -148,7 +148,7 @@ export const createRouterRuntime = ({
     }
 
     /**
-     * TODO: routers 如果是静态对象，可以提前生成 runtimes, 避免运行时生成开销
+     * TODO: 考虑添加缓存机制，避免重复创建相同配置的 runtimes
      */
     private async createRuntimesByRouters(model?: string): Promise<RuntimeItem[]> {
       // 动态获取 routers，支持传入 model
@@ -181,9 +181,11 @@ export const createRouterRuntime = ({
       for (const runtimeItem of runtimes) {
         const models = runtimeItem.models || [];
         if (models.includes(model)) {
+          console.log(`get runtime ${runtimeItem.id} ${model}`);
           return runtimeItem.runtime;
         }
       }
+
       return runtimes.at(-1)!.runtime;
     }
 

package/packages/model-runtime/src/core/openaiCompatibleFactory/createImage.ts

@@ -3,9 +3,11 @@ import { RuntimeImageGenParamsValue } from 'model-bank';
 import OpenAI from 'openai';
 
 import { CreateImagePayload, CreateImageResponse } from '../../types/image';
+import { getModelPricing } from '../../utils/getModelPricing';
 import { imageUrlToBase64 } from '../../utils/imageToBase64';
 import { convertImageUrlToFile } from '../../utils/openaiHelpers';
 import { parseDataUri } from '../../utils/uriParser';
+import { convertOpenAIImageUsage } from '../usageConverters/openai';
 
 const log = createDebug('lobe-image:openai-compatible');
 
@@ -15,6 +17,7 @@ const log = createDebug('lobe-image:openai-compatible');
 async function generateByImageMode(
   client: OpenAI,
   payload: CreateImagePayload,
+  provider: string,
 ): Promise<CreateImageResponse> {
   const { model, params } = payload;
 
@@ -112,8 +115,15 @@ async function generateByImageMode(
     throw new Error('Invalid image response: missing both b64_json and url fields');
   }
 
+  log('provider: %s', provider);
+
   return {
     imageUrl,
+    ...(img.usage
+      ? {
+          modelUsage: convertOpenAIImageUsage(img.usage, await getModelPricing(model, provider)),
+        }
+      : {}),
   };
 }
 
@@ -218,7 +228,7 @@ async function generateByChatModel(
 export async function createOpenAICompatibleImage(
   client: OpenAI,
   payload: CreateImagePayload,
-  _provider: string, // eslint-disable-line @typescript-eslint/no-unused-vars
+  provider: string, // eslint-disable-line @typescript-eslint/no-unused-vars
 ): Promise<CreateImageResponse> {
   try {
     const { model } = payload;
@@ -229,7 +239,7 @@ export async function createOpenAICompatibleImage(
     }
 
     // Default to traditional images API
-    return await generateByImageMode(client, payload);
+    return await generateByImageMode(client, payload, provider);
   } catch (error) {
     const err = error as Error;
     log('Error in createImage: %O', err);

package/packages/model-runtime/src/core/openaiCompatibleFactory/index.ts

@@ -27,6 +27,7 @@ import { AgentRuntimeError } from '../../utils/createError';
 import { debugResponse, debugStream } from '../../utils/debugStream';
 import { desensitizeUrl } from '../../utils/desensitizeUrl';
 import { getModelPropertyWithFallback } from '../../utils/getFallbackModelProperty';
+import { getModelPricing } from '../../utils/getModelPricing';
 import { handleOpenAIError } from '../../utils/handleOpenAIError';
 import { convertOpenAIMessages, convertOpenAIResponseInputs } from '../../utils/openaiHelpers';
 import { postProcessModelList } from '../../utils/postProcessModelList';
@@ -228,7 +229,11 @@ export const createOpenAICompatibleRuntime = <T extends Record<string, any> = an
         const streamOptions: OpenAIStreamOptions = {
           bizErrorTypeTransformer: chatCompletion?.handleStreamBizErrorType,
           callbacks: options?.callback,
-          provider: this.id,
+          payload: {
+            model: payload.model,
+            pricing: await getModelPricing(payload.model, this.id),
+            provider: this.id,
+          },
         };
 
         if (customClient?.createChatCompletionStream) {
@@ -276,7 +281,10 @@ export const createOpenAICompatibleRuntime = <T extends Record<string, any> = an
                   callbacks: streamOptions.callbacks,
                   inputStartAt,
                 })
-              : OpenAIStream(prod, { ...streamOptions, inputStartAt }),
+              : OpenAIStream(prod, {
+                  ...streamOptions,
+                  inputStartAt,
+                }),
             {
               headers: options?.headers,
             },
@@ -320,7 +328,7 @@ export const createOpenAICompatibleRuntime = <T extends Record<string, any> = an
       }
 
       // Use the new createOpenAICompatibleImage function
-      return createOpenAICompatibleImage(this.client, payload, provider);
+      return createOpenAICompatibleImage(this.client, payload, this.id);
     }
 
     async models() {
@@ -453,7 +461,6 @@ export const createOpenAICompatibleRuntime = <T extends Record<string, any> = an
           headers: options?.headers,
           signal: options?.signal,
         });
-
         return mp3.arrayBuffer();
       } catch (error) {
         throw this.handleError(error);
@@ -585,7 +592,11 @@ export const createOpenAICompatibleRuntime = <T extends Record<string, any> = an
       const streamOptions: OpenAIStreamOptions = {
         bizErrorTypeTransformer: chatCompletion?.handleStreamBizErrorType,
         callbacks: options?.callback,
-        provider: this.id,
+        payload: {
+          model: payload.model,
+          pricing: await getModelPricing(payload.model, this.id),
+          provider: this.id,
+        },
       };
 
       if (isStreaming) {

package/packages/model-runtime/src/core/streams/anthropic.ts

@@ -1,9 +1,12 @@
 import Anthropic from '@anthropic-ai/sdk';
 import type { Stream } from '@anthropic-ai/sdk/streaming';
-import { ChatCitationItem, ModelTokensUsage } from '@lobechat/types';
+
+import { ChatCitationItem } from '@/types/message';
 
 import { ChatStreamCallbacks } from '../../types';
+import { convertAnthropicUsage } from '../usageConverters';
 import {
+  ChatPayloadForTransformStream,
   StreamContext,
   StreamProtocolChunk,
   StreamProtocolToolCallChunk,
@@ -17,31 +20,20 @@ import {
 export const transformAnthropicStream = (
   chunk: Anthropic.MessageStreamEvent,
   context: StreamContext,
+  payload?: ChatPayloadForTransformStream,
 ): StreamProtocolChunk | StreamProtocolChunk[] => {
   // maybe need another structure to add support for multiple choices
   switch (chunk.type) {
     case 'message_start': {
       context.id = chunk.message.id;
       context.returnedCitationArray = [];
-      let totalInputTokens = chunk.message.usage?.input_tokens;
-
-      if (
-        chunk.message.usage?.cache_creation_input_tokens ||
-        chunk.message.usage?.cache_read_input_tokens
-      ) {
-        totalInputTokens =
-          chunk.message.usage?.input_tokens +
-          (chunk.message.usage.cache_creation_input_tokens || 0) +
-          (chunk.message.usage.cache_read_input_tokens || 0);
-      }
+      const usage = convertAnthropicUsage(chunk, undefined, payload);
 
-      context.usage = {
-        inputCacheMissTokens: chunk.message.usage?.input_tokens,
-        inputCachedTokens: chunk.message.usage?.cache_read_input_tokens || undefined,
-        inputWriteCacheTokens: chunk.message.usage?.cache_creation_input_tokens || undefined,
-        totalInputTokens,
-        totalOutputTokens: chunk.message.usage?.output_tokens,
-      };
+      if (usage) {
+        context.usage = usage;
+      } else {
+        delete context.usage;
+      }
 
       return { data: chunk.message, id: chunk.message.id, type: 'data' };
     }
@@ -193,26 +185,19 @@ export const transformAnthropicStream = (
     }
 
     case 'message_delta': {
-      const totalOutputTokens =
-        chunk.usage?.output_tokens + (context.usage?.totalOutputTokens || 0);
-      const totalInputTokens = context.usage?.totalInputTokens || 0;
-      const totalTokens = totalInputTokens + totalOutputTokens;
+      const aggregatedUsage = convertAnthropicUsage(chunk, context.usage, payload);
+
+      if (aggregatedUsage) {
+        context.usage = aggregatedUsage;
+      }
 
-      if (totalTokens > 0) {
+      if (aggregatedUsage && (aggregatedUsage.totalTokens ?? 0) > 0) {
         return [
           { data: chunk.delta.stop_reason, id: context.id, type: 'stop' },
-          {
-            data: {
-              ...context.usage,
-              totalInputTokens,
-              totalOutputTokens,
-              totalTokens,
-            } as ModelTokensUsage,
-            id: context.id,
-            type: 'usage',
-          },
+          { data: aggregatedUsage, id: context.id, type: 'usage' },
         ];
       }
+
       return { data: chunk.delta.stop_reason, id: context.id, type: 'stop' };
     }
 
@@ -241,20 +226,24 @@ export interface AnthropicStreamOptions {
   callbacks?: ChatStreamCallbacks;
   enableStreaming?: boolean; // 选择 TPS 计算方式（非流式时传 false）
   inputStartAt?: number;
+  payload?: ChatPayloadForTransformStream;
 }
 
 export const AnthropicStream = (
   stream: Stream<Anthropic.MessageStreamEvent> | ReadableStream,
-  { callbacks, inputStartAt, enableStreaming = true }: AnthropicStreamOptions = {},
+  { callbacks, inputStartAt, enableStreaming = true, payload }: AnthropicStreamOptions = {},
 ) => {
   const streamStack: StreamContext = { id: '' };
 
   const readableStream =
     stream instanceof ReadableStream ? stream : convertIterableToStream(stream);
 
+  const transformWithPayload: typeof transformAnthropicStream = (chunk, ctx) =>
+    transformAnthropicStream(chunk, ctx, payload);
+
   return readableStream
     .pipeThrough(
-      createTokenSpeedCalculator(transformAnthropicStream, {
+      createTokenSpeedCalculator(transformWithPayload, {
         enableStreaming: enableStreaming,
         inputStartAt,
         streamStack,

package/packages/model-runtime/src/core/streams/google/google-ai.test.ts

@@ -281,7 +281,7 @@ describe('GoogleGenerativeAIStream', () => {
       // usage
       'id: chat_1\n',
       'event: usage\n',
-      `data: {"inputCachedTokens":14286,"inputTextTokens":15725,"outputImageTokens":0,"outputTextTokens":1053,"totalInputTokens":15725,"totalOutputTokens":1053,"totalTokens":16778}\n\n`,
+      `data: {"inputCacheMissTokens":1439,"inputCachedTokens":14286,"inputTextTokens":15725,"outputImageTokens":0,"outputTextTokens":1053,"totalInputTokens":15725,"totalOutputTokens":1053,"totalTokens":16778}\n\n`,
     ]);
   });
 

package/packages/model-runtime/src/core/streams/google/index.ts

@@ -1,9 +1,11 @@
 import { GenerateContentResponse } from '@google/genai';
-import { GroundingSearch, ModelTokensUsage } from '@lobechat/types';
+import { GroundingSearch } from '@lobechat/types';
 
 import { ChatStreamCallbacks } from '../../../types';
 import { nanoid } from '../../../utils/uuid';
+import { convertGoogleAIUsage } from '../../usageConverters/google-ai';
 import {
+  ChatPayloadForTransformStream,
   StreamContext,
   StreamProtocolChunk,
   StreamToolCallChunkData,
@@ -28,6 +30,7 @@ const getBlockReasonMessage = (blockReason: string): string => {
 const transformGoogleGenerativeAIStream = (
   chunk: GenerateContentResponse,
   context: StreamContext,
+  payload?: ChatPayloadForTransformStream,
 ): StreamProtocolChunk | StreamProtocolChunk[] => {
   // Handle injected internal error marker to pass through detailed error info
   if ((chunk as any)?.[LOBE_ERROR_KEY]) {
@@ -60,46 +63,15 @@ const transformGoogleGenerativeAIStream = (
 
   // maybe need another structure to add support for multiple choices
   const candidate = chunk.candidates?.[0];
-  const usage = chunk.usageMetadata;
+  const { usageMetadata } = chunk;
   const usageChunks: StreamProtocolChunk[] = [];
-  if (candidate?.finishReason && usage) {
-    // totalTokenCount = promptTokenCount + candidatesTokenCount + thoughtsTokenCount
-    const reasoningTokens = usage.thoughtsTokenCount;
-
-    const candidatesDetails = usage.candidatesTokensDetails;
-    const candidatesTotal =
-      usage.candidatesTokenCount ??
-      candidatesDetails?.reduce((s: number, i: any) => s + (i?.tokenCount ?? 0), 0) ??
-      0;
-
-    const outputImageTokens =
-      candidatesDetails?.find((i: any) => i.modality === 'IMAGE')?.tokenCount ?? 0;
-    const outputTextTokens =
-      candidatesDetails?.find((i: any) => i.modality === 'TEXT')?.tokenCount ??
-      Math.max(0, candidatesTotal - outputImageTokens);
-
-    const totalOutputTokens = candidatesTotal + (reasoningTokens ?? 0);
-
-    usageChunks.push(
-      { data: candidate.finishReason, id: context?.id, type: 'stop' },
-      {
-        data: {
-          inputCachedTokens: usage.cachedContentTokenCount,
-          inputImageTokens: usage.promptTokensDetails?.find((i) => i.modality === 'IMAGE')
-            ?.tokenCount,
-          inputTextTokens: usage.promptTokensDetails?.find((i) => i.modality === 'TEXT')
-            ?.tokenCount,
-          outputImageTokens,
-          outputReasoningTokens: reasoningTokens,
-          outputTextTokens,
-          totalInputTokens: usage.promptTokenCount,
-          totalOutputTokens,
-          totalTokens: usage.totalTokenCount,
-        } as ModelTokensUsage,
-        id: context?.id,
-        type: 'usage',
-      },
-    );
+  if (candidate?.finishReason && usageMetadata) {
+    usageChunks.push({ data: candidate.finishReason, id: context?.id, type: 'stop' });
+
+    const convertedUsage = convertGoogleAIUsage(usageMetadata, payload?.pricing);
+    if (convertedUsage) {
+      usageChunks.push({ data: convertedUsage, id: context?.id, type: 'usage' });
+    }
   }
 
   const functionCalls = chunk.functionCalls;
@@ -213,17 +185,21 @@ export interface GoogleAIStreamOptions {
   callbacks?: ChatStreamCallbacks;
   enableStreaming?: boolean; // 选择 TPS 计算方式（非流式时传 false）
   inputStartAt?: number;
+  payload?: ChatPayloadForTransformStream;
 }
 
 export const GoogleGenerativeAIStream = (
   rawStream: ReadableStream<GenerateContentResponse>,
-  { callbacks, inputStartAt, enableStreaming = true }: GoogleAIStreamOptions = {},
+  { callbacks, inputStartAt, enableStreaming = true, payload }: GoogleAIStreamOptions = {},
 ) => {
   const streamStack: StreamContext = { id: 'chat_' + nanoid() };
 
+  const transformWithPayload: typeof transformGoogleGenerativeAIStream = (chunk, ctx) =>
+    transformGoogleGenerativeAIStream(chunk, ctx, payload);
+
   return rawStream
     .pipeThrough(
-      createTokenSpeedCalculator(transformGoogleGenerativeAIStream, {
+      createTokenSpeedCalculator(transformWithPayload, {
         enableStreaming: enableStreaming,
         inputStartAt,
         streamStack,

package/packages/model-runtime/src/core/streams/openai/openai.test.ts

@@ -360,14 +360,9 @@ describe('OpenAIStream', () => {
     }
 
     expect(chunks).toEqual(
-      [
-        'id: 1',
-        'event: text',
-        `data: "Hello"\n`,
-        'id: 1',
-        'event: data',
-        `data: {"id":"1"}\n`,
-      ].map((i) => `${i}\n`),
+      ['id: 1', 'event: text', `data: "Hello"\n`, 'id: 1', 'event: data', `data: {"id":"1"}\n`].map(
+        (i) => `${i}\n`,
+      ),
     );
   });
 
@@ -413,7 +408,9 @@ describe('OpenAIStream', () => {
 
     const protocolStream = OpenAIStream(mockOpenAIStream, {
       bizErrorTypeTransformer: () => AgentRuntimeErrorType.PermissionDenied,
-      provider: 'grok',
+      payload: {
+        provider: 'grok',
+      },
     });
 
     const decoder = new TextDecoder();
@@ -2481,4 +2478,4 @@ describe('OpenAIStream', () => {
       `data: "${base64_2}"\n\n`,
     ]);
   });
-});
+});

package/packages/model-runtime/src/core/streams/openai/openai.ts

@@ -4,8 +4,9 @@ import type { Stream } from 'openai/streaming';
 
 import { ChatStreamCallbacks } from '../../../types';
 import { AgentRuntimeErrorType, ILobeAgentRuntimeErrorType } from '../../../types/error';
-import { convertUsage } from '../../../utils/usageConverter';
+import { convertOpenAIUsage } from '../../usageConverters';
 import {
+  ChatPayloadForTransformStream,
   FIRST_CHUNK_ERROR_KEY,
   StreamContext,
   StreamProtocolChunk,
@@ -44,7 +45,7 @@ const processMarkdownBase64Images = (text: string): { cleanedText: string; urls:
 const transformOpenAIStream = (
   chunk: OpenAI.ChatCompletionChunk,
   streamContext: StreamContext,
-  provider?: string,
+  payload?: ChatPayloadForTransformStream,
 ): StreamProtocolChunk | StreamProtocolChunk[] => {
   // handle the first chunk error
   if (FIRST_CHUNK_ERROR_KEY in chunk) {
@@ -75,7 +76,7 @@ const transformOpenAIStream = (
     if (!Array.isArray(chunk.choices) || chunk.choices.length === 0) {
       if (chunk.usage) {
         const usage = chunk.usage;
-        return { data: convertUsage(usage, provider), id: chunk.id, type: 'usage' };
+        return { data: convertOpenAIUsage(usage, payload), id: chunk.id, type: 'usage' };
       }
 
       return { data: chunk, id: chunk.id, type: 'data' };
@@ -232,7 +233,7 @@ const transformOpenAIStream = (
 
       if (chunk.usage) {
         const usage = chunk.usage;
-        return { data: convertUsage(usage, provider), id: chunk.id, type: 'usage' };
+        return { data: convertOpenAIUsage(usage, payload), id: chunk.id, type: 'usage' };
       }
 
       // xAI Live Search 功能返回引用源
@@ -312,7 +313,7 @@ const transformOpenAIStream = (
         // 如果 content 是空字符串但 chunk 带有 usage，则优先返回 usage（例如 Gemini image-preview 最终会在单独的 chunk 中返回 usage）
         if (content === '' && chunk.usage) {
           const usage = chunk.usage;
-          return { data: convertUsage(usage, provider), id: chunk.id, type: 'usage' };
+          return { data: convertOpenAIUsage(usage, payload), id: chunk.id, type: 'usage' };
         }
 
         // 判断是否有 citations 内容，更新 returnedCitation 状态
@@ -387,7 +388,7 @@ const transformOpenAIStream = (
     // litellm 的返回结果中，存在 delta 为空，但是有 usage 的情况
     if (chunk.usage) {
       const usage = chunk.usage;
-      return { data: convertUsage(usage, provider), id: chunk.id, type: 'usage' };
+      return { data: convertOpenAIUsage(usage, payload), id: chunk.id, type: 'usage' };
     }
 
     // 其余情况下，返回 delta 和 index
@@ -426,23 +427,25 @@ export interface OpenAIStreamOptions {
   callbacks?: ChatStreamCallbacks;
   enableStreaming?: boolean; // 选择 TPS 计算方式（非流式时传 false）
   inputStartAt?: number;
-  provider?: string;
+  payload?: ChatPayloadForTransformStream;
 }
 
 export const OpenAIStream = (
   stream: Stream<OpenAI.ChatCompletionChunk> | ReadableStream,
   {
     callbacks,
-    provider,
     bizErrorTypeTransformer,
+    payload,
     inputStartAt,
     enableStreaming = true,
   }: OpenAIStreamOptions = {},
 ) => {
-  const streamStack: StreamContext = { id: '' };
+  const streamStack: StreamContext = {
+    id: '',
+  };
 
   const transformWithProvider = (chunk: OpenAI.ChatCompletionChunk, streamContext: StreamContext) =>
-    transformOpenAIStream(chunk, streamContext, provider);
+    transformOpenAIStream(chunk, streamContext, payload);
 
   const readableStream =
     stream instanceof ReadableStream ? stream : convertIterableToStream(stream);
@@ -452,7 +455,7 @@ export const OpenAIStream = (
       // 1. handle the first error if exist
       // provider like huggingface or minimax will return error in the stream,
       // so in the first Transformer, we need to handle the error
-      .pipeThrough(createFirstErrorHandleTransformer(bizErrorTypeTransformer, provider))
+      .pipeThrough(createFirstErrorHandleTransformer(bizErrorTypeTransformer, payload?.provider))
       .pipeThrough(
         createTokenSpeedCalculator(transformWithProvider, {
           enableStreaming: enableStreaming,