♻️ refactor: improve model fetch behavior (#6055)

* ♻️ refactor: refactor model fetch behavior * 🐛 fix: fix gemini thinking model set functionCall tag * 🐛 fix: fix Spark fc tag missing * ✨ feat: add model fetch support for Tencent Cloud * ♻️ refactor: refactor import `LOBE_DEFAULT_MODEL_LIST` with async method - Part 1 * ♻️ refactor: refactor import LOBE_DEFAULT_MODEL_LIST with async method - Part 2 * 🔨 chore: add vision, reasoning, fc default value * ♻️ refactor: improve knownModel matching * 🐛 fix: fix ci error
lobehub · Feb 14, 2025 · 4c2aaf6 · 4c2aaf6
1 parent e684f10
commit 4c2aaf6
Show file tree

Hide file tree

Showing 38 changed files with 1,119 additions and 553 deletions.
diff --git a/src/config/aiModels/spark.ts b/src/config/aiModels/spark.ts
@@ -32,6 +32,9 @@ const sparkChatModels: AIChatModelCard[] = [
     type: 'chat',
   },
   {
+    abilities: {
+      functionCall: true,
+    },
     contextWindowTokens: 8192,
     description:
       'Spark Max 为功能最为全面的版本，支持联网搜索及众多内置插件。其全面优化的核心能力以及系统角色设定和函数调用功能，使其在各种复杂应用场景中的表现极为优异和出色。',
@@ -42,6 +45,9 @@ const sparkChatModels: AIChatModelCard[] = [
     type: 'chat',
   },
   {
+    abilities: {
+      functionCall: true,
+    },
     contextWindowTokens: 32_768,
     description:
       'Spark Max 32K 配置了大上下文处理能力，更强的上下文理解和逻辑推理能力，支持32K tokens的文本输入，适用于长文档阅读、私有知识问答等场景',
@@ -52,6 +58,9 @@ const sparkChatModels: AIChatModelCard[] = [
     type: 'chat',
   },
   {
+    abilities: {
+      functionCall: true,
+    },
     contextWindowTokens: 8192,
     description:
       'Spark Ultra 是星火大模型系列中最为强大的版本，在升级联网搜索链路同时，提升对文本内容的理解和总结能力。它是用于提升办公生产力和准确响应需求的全方位解决方案，是引领行业的智能产品。',

diff --git a/src/libs/agent-runtime/ai360/index.ts b/src/libs/agent-runtime/ai360/index.ts
@@ -1,7 +1,7 @@
 import { ModelProvider } from '../types';
 import { LobeOpenAICompatibleFactory } from '../utils/openaiCompatibleFactory';
 
-import { LOBE_DEFAULT_MODEL_LIST } from '@/config/aiModels';
+import type { ChatModelCard } from '@/types/llm';
 
 export interface Ai360ModelCard {
   id: string;
@@ -22,28 +22,44 @@ export const LobeAi360AI = LobeOpenAICompatibleFactory({
   debug: {
     chatCompletion: () => process.env.DEBUG_AI360_CHAT_COMPLETION === '1',
   },
-  models: {
-    transformModel: (m) => {
-      const reasoningKeywords = [
-        '360gpt2-o1',
-        '360zhinao2-o1',
-      ];
+  models: async ({ client }) => {
+    const { LOBE_DEFAULT_MODEL_LIST } = await import('@/config/aiModels');
 
-      const model = m as unknown as Ai360ModelCard;
+    const reasoningKeywords = [
+      '360gpt2-o1',
+      '360zhinao2-o1',
+    ];
 
-      return {
-        contextWindowTokens: model.total_tokens,
-        displayName: LOBE_DEFAULT_MODEL_LIST.find((m) => model.id === m.id)?.displayName ?? undefined,
-        enabled: LOBE_DEFAULT_MODEL_LIST.find((m) => model.id === m.id)?.enabled || false,
-        functionCall: model.id === '360gpt-pro',
-        id: model.id,
-        maxTokens:
-          typeof model.max_tokens === 'number'
-            ? model.max_tokens
-            : undefined,
-        reasoning: reasoningKeywords.some(keyword => model.id.toLowerCase().includes(keyword)),
-      };
-    },
+    const modelsPage = await client.models.list() as any;
+    const modelList: Ai360ModelCard[] = modelsPage.data;
+
+    return modelList
+      .map((model) => {
+        const knownModel = LOBE_DEFAULT_MODEL_LIST.find((m) => model.id.toLowerCase() === m.id.toLowerCase());
+
+        return {
+          contextWindowTokens: model.total_tokens,
+          displayName: knownModel?.displayName ?? undefined,
+          enabled: knownModel?.enabled || false,
+          functionCall:
+            model.id === '360gpt-pro'
+            || knownModel?.abilities?.functionCall
+            || false,
+          id: model.id,
+          maxTokens:
+            typeof model.max_tokens === 'number'
+              ? model.max_tokens
+              : undefined,
+          reasoning:
+            reasoningKeywords.some(keyword => model.id.toLowerCase().includes(keyword))
+            || knownModel?.abilities?.reasoning
+            || false,
+          vision:
+            knownModel?.abilities?.vision
+            || false,
+        };
+      })
+      .filter(Boolean) as ChatModelCard[];
   },
   provider: ModelProvider.Ai360,
 });
diff --git a/src/libs/agent-runtime/anthropic/index.ts b/src/libs/agent-runtime/anthropic/index.ts
@@ -13,7 +13,6 @@ import { buildAnthropicMessages, buildAnthropicTools } from '../utils/anthropicH
 import { StreamingResponse } from '../utils/response';
 import { AnthropicStream } from '../utils/streams';
 
-import { LOBE_DEFAULT_MODEL_LIST } from '@/config/aiModels';
 import type { ChatModelCard } from '@/types/llm';
 
 export interface AnthropicModelCard {
@@ -114,6 +113,8 @@ export class LobeAnthropicAI implements LobeRuntimeAI {
   }
 
   async models() {
+    const { LOBE_DEFAULT_MODEL_LIST } = await import('@/config/aiModels');
+
     const url = `${this.baseURL}/v1/models`;
     const response = await fetch(url, {
       headers: {
@@ -128,13 +129,24 @@ export class LobeAnthropicAI implements LobeRuntimeAI {
 
     return modelList
       .map((model) => {
+        const knownModel = LOBE_DEFAULT_MODEL_LIST.find((m) => model.id.toLowerCase() === m.id.toLowerCase());
+
         return {
-          contextWindowTokens: LOBE_DEFAULT_MODEL_LIST.find((m) => model.id === m.id)?.contextWindowTokens ?? undefined,
+          contextWindowTokens: knownModel?.contextWindowTokens ?? undefined,
           displayName: model.display_name,
-          enabled: LOBE_DEFAULT_MODEL_LIST.find((m) => model.id === m.id)?.enabled || false,
-          functionCall: model.id.toLowerCase().includes('claude-3'),
+          enabled: knownModel?.enabled || false,
+          functionCall:
+            model.id.toLowerCase().includes('claude-3')
+            || knownModel?.abilities?.functionCall
+            || false,
           id: model.id,
-          vision: model.id.toLowerCase().includes('claude-3') && !model.id.toLowerCase().includes('claude-3-5-haiku'),
+          reasoning:
+            knownModel?.abilities?.reasoning
+            || false,
+          vision:
+            model.id.toLowerCase().includes('claude-3') && !model.id.toLowerCase().includes('claude-3-5-haiku')
+            || knownModel?.abilities?.vision
+            || false,
         };
       })
       .filter(Boolean) as ChatModelCard[];

diff --git a/src/libs/agent-runtime/baichuan/index.ts b/src/libs/agent-runtime/baichuan/index.ts
@@ -3,7 +3,6 @@ import OpenAI from 'openai';
 import { ChatStreamPayload, ModelProvider } from '../types';
 import { LobeOpenAICompatibleFactory } from '../utils/openaiCompatibleFactory';
 
-import { LOBE_DEFAULT_MODEL_LIST } from '@/config/aiModels';
 import type { ChatModelCard } from '@/types/llm';
 
 export interface BaichuanModelCard {
@@ -32,18 +31,28 @@ export const LobeBaichuanAI = LobeOpenAICompatibleFactory({
     chatCompletion: () => process.env.DEBUG_BAICHUAN_CHAT_COMPLETION === '1',
   },
   models: async ({ client }) => {
+    const { LOBE_DEFAULT_MODEL_LIST } = await import('@/config/aiModels');
+
     const modelsPage = await client.models.list() as any;
     const modelList: BaichuanModelCard[] = modelsPage.data;
 
     return modelList
       .map((model) => {
+        const knownModel = LOBE_DEFAULT_MODEL_LIST.find((m) => model.model.toLowerCase() === m.id.toLowerCase());
+
         return {
           contextWindowTokens: model.max_input_length,
           displayName: model.model_show_name,
-          enabled: LOBE_DEFAULT_MODEL_LIST.find((m) => model.model === m.id)?.enabled || false,
+          enabled: knownModel?.enabled || false,
           functionCall: model.function_call,
           id: model.model,
           maxTokens: model.max_tokens,
+          reasoning:
+            knownModel?.abilities?.reasoning
+            || false,
+          vision:
+            knownModel?.abilities?.vision
+            || false,
         };
       })
       .filter(Boolean) as ChatModelCard[];

diff --git a/src/libs/agent-runtime/cloudflare/index.ts b/src/libs/agent-runtime/cloudflare/index.ts
@@ -12,7 +12,6 @@ import { debugStream } from '../utils/debugStream';
 import { StreamingResponse } from '../utils/response';
 import { createCallbacksTransformer } from '../utils/streams';
 
-import { LOBE_DEFAULT_MODEL_LIST } from '@/config/aiModels';
 import { ChatModelCard } from '@/types/llm';
 
 export interface CloudflareModelCard {
@@ -113,6 +112,8 @@ export class LobeCloudflareAI implements LobeRuntimeAI {
   }
 
   async models(): Promise<ChatModelCard[]> {
+    const { LOBE_DEFAULT_MODEL_LIST } = await import('@/config/aiModels');
+
     const url = `${DEFAULT_BASE_URL_PREFIX}/client/v4/accounts/${this.accountID}/ai/models/search`;
     const response = await fetch(url, {
       headers: {
@@ -127,16 +128,30 @@ export class LobeCloudflareAI implements LobeRuntimeAI {
 
     return modelList
       .map((model) => {
+        const knownModel = LOBE_DEFAULT_MODEL_LIST.find((m) => model.name.toLowerCase() === m.id.toLowerCase());
+
         return {
           contextWindowTokens: model.properties?.max_total_tokens
             ? Number(model.properties.max_total_tokens)
-            : LOBE_DEFAULT_MODEL_LIST.find((m) => model.name === m.id)?.contextWindowTokens ?? undefined,
-          displayName: LOBE_DEFAULT_MODEL_LIST.find((m) => model.name === m.id)?.displayName ?? (model.properties?.["beta"] === "true" ? `${model.name} (Beta)` : undefined),
-          enabled: LOBE_DEFAULT_MODEL_LIST.find((m) => model.name === m.id)?.enabled || false,
-          functionCall: model.description.toLowerCase().includes('function call') || model.properties?.["function_calling"] === "true",
+            : knownModel?.contextWindowTokens ?? undefined,
+          displayName: knownModel?.displayName ?? (model.properties?.["beta"] === "true" ? `${model.name} (Beta)` : undefined),
+          enabled: knownModel?.enabled || false,
+          functionCall:
+            model.description.toLowerCase().includes('function call')
+            || model.properties?.["function_calling"] === "true"
+            || knownModel?.abilities?.functionCall
+            || false,
           id: model.name,
-          reasoning: model.name.toLowerCase().includes('deepseek-r1'),
-          vision: model.name.toLowerCase().includes('vision') || model.task?.name.toLowerCase().includes('image-to-text') || model.description.toLowerCase().includes('vision'),
+          reasoning:
+            model.name.toLowerCase().includes('deepseek-r1')
+            || knownModel?.abilities?.reasoning
+            || false,
+          vision:
+            model.name.toLowerCase().includes('vision')
+            || model.task?.name.toLowerCase().includes('image-to-text')
+            || model.description.toLowerCase().includes('vision')
+            || knownModel?.abilities?.vision
+            || false,
         };
       })
       .filter(Boolean) as ChatModelCard[];

diff --git a/src/libs/agent-runtime/deepseek/index.ts b/src/libs/agent-runtime/deepseek/index.ts
@@ -3,7 +3,7 @@ import OpenAI from 'openai';
 import { ChatStreamPayload, ModelProvider } from '../types';
 import { LobeOpenAICompatibleFactory } from '../utils/openaiCompatibleFactory';
 
-import { LOBE_DEFAULT_MODEL_LIST } from '@/config/aiModels';
+import type { ChatModelCard } from '@/types/llm';
 
 export interface DeepSeekModelCard {
   id: string;
@@ -59,19 +59,35 @@ export const LobeDeepSeekAI = LobeOpenAICompatibleFactory({
   debug: {
     chatCompletion: () => process.env.DEBUG_DEEPSEEK_CHAT_COMPLETION === '1',
   },
-  models: {
-    transformModel: (m) => {
-      const model = m as unknown as DeepSeekModelCard;
+  models: async ({ client }) => {
+    const { LOBE_DEFAULT_MODEL_LIST } = await import('@/config/aiModels');
 
-      return {
-        contextWindowTokens: LOBE_DEFAULT_MODEL_LIST.find((m) => model.id === m.id)?.contextWindowTokens ?? undefined,
-        displayName: LOBE_DEFAULT_MODEL_LIST.find((m) => model.id === m.id)?.displayName ?? undefined,
-        enabled: LOBE_DEFAULT_MODEL_LIST.find((m) => model.id === m.id)?.enabled || false,
-        functionCall: !model.id.toLowerCase().includes('deepseek-reasoner'),
-        id: model.id,
-        reasoning: model.id.toLowerCase().includes('deepseek-reasoner'),
-      };
-    },
+    const modelsPage = await client.models.list() as any;
+    const modelList: DeepSeekModelCard[] = modelsPage.data;
+
+    return modelList
+      .map((model) => {
+        const knownModel = LOBE_DEFAULT_MODEL_LIST.find((m) => model.id.toLowerCase() === m.id.toLowerCase());
+
+        return {
+          contextWindowTokens: knownModel?.contextWindowTokens ?? undefined,
+          displayName: knownModel?.displayName ?? undefined,
+          enabled: knownModel?.enabled || false,
+          functionCall:
+            !model.id.toLowerCase().includes('reasoner')
+            || knownModel?.abilities?.functionCall
+            || false,
+          id: model.id,
+          reasoning:
+            model.id.toLowerCase().includes('reasoner')
+            || knownModel?.abilities?.reasoning
+            || false,
+          vision:
+            knownModel?.abilities?.vision
+            || false,
+        };
+      })
+      .filter(Boolean) as ChatModelCard[];
   },
   provider: ModelProvider.DeepSeek,
 });
diff --git a/src/libs/agent-runtime/fireworksai/index.ts b/src/libs/agent-runtime/fireworksai/index.ts
@@ -1,7 +1,7 @@
 import { ModelProvider } from '../types';
 import { LobeOpenAICompatibleFactory } from '../utils/openaiCompatibleFactory';
 
-import { LOBE_DEFAULT_MODEL_LIST } from '@/config/aiModels';
+import type { ChatModelCard } from '@/types/llm';
 
 export interface FireworksAIModelCard {
   context_length: number;
@@ -15,25 +15,37 @@ export const LobeFireworksAI = LobeOpenAICompatibleFactory({
   debug: {
     chatCompletion: () => process.env.DEBUG_FIREWORKSAI_CHAT_COMPLETION === '1',
   },
-  models: {
-    transformModel: (m) => {
-      const reasoningKeywords = [
-        'deepseek-r1',
-        'qwq',
-      ];
+  models: async ({ client }) => {
+    const { LOBE_DEFAULT_MODEL_LIST } = await import('@/config/aiModels');
 
-      const model = m as unknown as FireworksAIModelCard;
+    const reasoningKeywords = [
+      'deepseek-r1',
+      'qwq',
+    ];
 
-      return {
-        contextWindowTokens: model.context_length,
-        displayName: LOBE_DEFAULT_MODEL_LIST.find((m) => model.id === m.id)?.displayName ?? undefined,
-        enabled: LOBE_DEFAULT_MODEL_LIST.find((m) => model.id === m.id)?.enabled || false,
-        functionCall: model.supports_tools || model.id.toLowerCase().includes('function'),
-        id: model.id,
-        reasoning: reasoningKeywords.some(keyword => model.id.toLowerCase().includes(keyword)),
-        vision: model.supports_image_input,
-      };
-    },
+    const modelsPage = await client.models.list() as any;
+    const modelList: FireworksAIModelCard[] = modelsPage.data;
+
+    return modelList
+      .map((model) => {
+        const knownModel = LOBE_DEFAULT_MODEL_LIST.find((m) => model.id.toLowerCase() === m.id.toLowerCase());
+
+        return {
+          contextWindowTokens: model.context_length,
+          displayName: knownModel?.displayName ?? undefined,
+          enabled: knownModel?.enabled || false,
+          functionCall:
+            model.supports_tools
+            || model.id.toLowerCase().includes('function'),
+          id: model.id,
+          reasoning:
+            reasoningKeywords.some(keyword => model.id.toLowerCase().includes(keyword))
+            || knownModel?.abilities?.reasoning
+            || false,
+          vision: model.supports_image_input,
+        };
+      })
+      .filter(Boolean) as ChatModelCard[];
   },
   provider: ModelProvider.FireworksAI,
 });