@lobehub/chat 1.97.11 → 1.97.13

package/src/database/migrations/meta/_journal.json

@@ -182,6 +182,13 @@
       "when": 1749309388370,
       "tag": "0025_add_provider_config",
       "breakpoints": true
+    },
+    {
+      "idx": 26,
+      "version": "7",
+      "when": 1752212281564,
+      "tag": "0026_add_autovacuum_tuning",
+      "breakpoints": true
     }
   ],
   "version": "6"

package/src/features/ChatInput/ActionBar/Model/ControlsForm.tsx

@@ -14,6 +14,7 @@ import ContextCachingSwitch from './ContextCachingSwitch';
 import ReasoningEffortSlider from './ReasoningEffortSlider';
 import ReasoningTokenSlider from './ReasoningTokenSlider';
 import ThinkingBudgetSlider from './ThinkingBudgetSlider';
+import ThinkingSlider from './ThinkingSlider';
 
 const ControlsForm = memo(() => {
   const { t } = useTranslation('chat');
@@ -105,6 +106,16 @@ const ControlsForm = memo(() => {
       },
       tag: 'thinkingBudget',
     },
+    {
+      children: <ThinkingSlider />,
+      label: t('extendParams.thinking.title'),
+      layout: 'horizontal',
+      minWidth: undefined,
+      name: 'thinking',
+      style: {
+        paddingBottom: 0,
+      },
+    },
   ].filter(Boolean) as FormItemProps[];
 
   return (

package/src/features/ChatInput/ActionBar/Model/ThinkingSlider.tsx

@@ -0,0 +1,57 @@
+import { Slider } from 'antd';
+import { memo, useCallback } from 'react';
+import { Flexbox } from 'react-layout-kit';
+
+import { useAgentStore } from '@/store/agent';
+import { agentChatConfigSelectors } from '@/store/agent/selectors';
+
+const ThinkingSlider = memo(() => {
+  const [config, updateAgentChatConfig] = useAgentStore((s) => [
+    agentChatConfigSelectors.currentChatConfig(s),
+    s.updateAgentChatConfig,
+  ]);
+
+  const thinking = config.thinking || 'auto'; // Default to 'auto' if not set
+
+  const marks = {
+    0: 'OFF',
+    1: 'Auto',
+    2: 'ON',
+  };
+
+  const thinkingValues = ['disabled', 'auto', 'enabled'];
+  const indexValue = thinkingValues.indexOf(thinking);
+  const currentValue = indexValue === -1 ? 1 : indexValue;
+
+  const updateThinking = useCallback(
+    (value: number) => {
+      const thinkingMode = thinkingValues[value] as 'disabled' | 'auto' | 'enabled';
+      updateAgentChatConfig({ thinking: thinkingMode });
+    },
+    [updateAgentChatConfig],
+  );
+
+  return (
+    <Flexbox
+      align={'center'}
+      gap={12}
+      horizontal
+      paddingInline={'0 20px'}
+      style={{ minWidth: 200, width: '100%' }}
+    >
+      <Flexbox flex={1}>
+        <Slider
+          marks={marks}
+          max={2}
+          min={0}
+          onChange={updateThinking}
+          step={1}
+          tooltip={{ open: false }}
+          value={currentValue}
+        />
+      </Flexbox>
+    </Flexbox>
+  );
+});
+
+export default ThinkingSlider;

package/src/libs/model-runtime/bedrock/index.ts

@@ -3,7 +3,6 @@ import {
   InvokeModelCommand,
   InvokeModelWithResponseStreamCommand,
 } from '@aws-sdk/client-bedrock-runtime';
-import { experimental_buildLlama2Prompt } from 'ai/prompts';
 
 import { LobeRuntimeAI } from '../BaseAI';
 import { AgentRuntimeErrorType } from '../error';
@@ -25,6 +24,38 @@ import {
   createBedrockStream,
 } from '../utils/streams';
 
+/**
+ * A prompt constructor for HuggingFace LLama 2 chat models.
+ * Does not support `function` messages.
+ * @see https://huggingface.co/meta-llama/Llama-2-70b-chat-hf and https://huggingface.co/blog/llama2#how-to-prompt-llama-2
+ */
+export function experimental_buildLlama2Prompt(messages: { content: string; role: string }[]) {
+  const startPrompt = `<s>[INST] `;
+  const endPrompt = ` [/INST]`;
+  const conversation = messages.map(({ content, role }, index) => {
+    switch (role) {
+      case 'user': {
+        return content.trim();
+      }
+      case 'assistant': {
+        return ` [/INST] ${content}</s><s>[INST] `;
+      }
+      case 'function': {
+        throw new Error('Llama 2 does not support function calls.');
+      }
+      default: {
+        if (role === 'system' && index === 0) {
+          return `<<SYS>>\n${content}\n<</SYS>>\n\n`;
+        } else {
+          throw new Error(`Invalid message role: ${role}`);
+        }
+      }
+    }
+  });
+
+  return startPrompt + conversation.join('') + endPrompt;
+}
+
 export interface LobeBedrockAIParams {
   accessKeyId?: string;
   accessKeySecret?: string;

package/src/libs/model-runtime/utils/streams/bedrock/common.ts

@@ -2,7 +2,8 @@ import {
   InvokeModelWithResponseStreamResponse,
   ResponseStream,
 } from '@aws-sdk/client-bedrock-runtime';
-import { readableFromAsyncIterable } from 'ai';
+
+import { readableFromAsyncIterable } from '../protocol';
 
 const chatStreamable = async function* (stream: AsyncIterable<ResponseStream>) {
   for await (const response of stream) {

package/src/libs/model-runtime/utils/streams/protocol.ts

@@ -108,6 +108,22 @@ const chatStreamable = async function* <T>(stream: AsyncIterable<T>) {
 };
 
 const ERROR_CHUNK_PREFIX = '%FIRST_CHUNK_ERROR%: ';
+
+export function readableFromAsyncIterable<T>(iterable: AsyncIterable<T>) {
+  let it = iterable[Symbol.asyncIterator]();
+  return new ReadableStream<T>({
+    async cancel(reason) {
+      await it.return?.(reason);
+    },
+
+    async pull(controller) {
+      const { done, value } = await it.next();
+      if (done) controller.close();
+      else controller.enqueue(value);
+    },
+  });
+}
+
 // make the response to the streamable format
 export const convertIterableToStream = <T>(stream: AsyncIterable<T>) => {
   const iterable = chatStreamable(stream);

package/src/libs/model-runtime/utils/streams/qwen.ts

@@ -1,6 +1,8 @@
-import { ChatCompletionContentPartText } from 'ai/prompts';
 import OpenAI from 'openai';
-import { ChatCompletionContentPart } from 'openai/resources/index.mjs';
+import {
+  ChatCompletionContentPart,
+  ChatCompletionContentPartText,
+} from 'openai/resources/index.mjs';
 import type { Stream } from 'openai/streaming';
 
 import { ChatStreamCallbacks } from '../../types';

package/src/libs/model-runtime/volcengine/index.ts

@@ -2,6 +2,13 @@ import { ModelProvider } from '../types';
 import { createOpenAICompatibleRuntime } from '../utils/openaiCompatibleFactory';
 import { MODEL_LIST_CONFIGS, processModelList } from '../utils/modelParse';
 
+const THINKING_MODELS = [
+  'thinking-vision-pro',
+  'thinking-pro-m', 
+  'doubao-seed-1-6',
+  'doubao-1-5-ui-tars'
+];
+
 export interface VolcengineModelCard {
   id: string;
 }
@@ -15,12 +22,9 @@ export const LobeVolcengineAI = createOpenAICompatibleRuntime({
       return {
         ...rest,
         model,
-        ...(['thinking-vision-pro'].some((keyword) => model.toLowerCase().includes(keyword))
+        ...(THINKING_MODELS.some((keyword) => model.toLowerCase().includes(keyword))
           ? {
-              thinking:
-                thinking !== undefined && thinking.type === 'enabled'
-                  ? { type: 'enabled' }
-                  : { type: 'disabled' },
+              thinking: { type: thinking?.type }
             }
           : {}),
       } as any;

package/src/libs/model-runtime/xai/index.ts

@@ -9,9 +9,12 @@ export interface XAIModelCard {
 
 export const GrokReasoningModels = new Set([
   'grok-3-mini',
-  'grok-4-0709',
+  'grok-4',
 ]);
 
+export const isGrokReasoningModel = (model: string) =>
+  Array.from(GrokReasoningModels).some((id) => model.includes(id));
+
 export const LobeXAI = createOpenAICompatibleRuntime({
   baseURL: 'https://api.x.ai/v1',
   chatCompletion: {
@@ -20,9 +23,9 @@ export const LobeXAI = createOpenAICompatibleRuntime({
 
       return {
         ...rest,
-        frequency_penalty: GrokReasoningModels.has(model) ? undefined : frequency_penalty,
+        frequency_penalty: isGrokReasoningModel(model) ? undefined : frequency_penalty,
         model,
-        presence_penalty: GrokReasoningModels.has(model) ? undefined : presence_penalty,
+        presence_penalty: isGrokReasoningModel(model) ? undefined : presence_penalty,
         stream: true,
         ...(enabledSearch && {
           search_parameters: {

package/src/locales/default/chat.ts

@@ -47,6 +47,9 @@ export default {
     reasoningEffort: {
       title: '推理强度',
     },
+    thinking: {
+      title: '深度思考开关',
+    },
     title: '模型扩展功能',
   },
   history: {

package/src/services/chat.ts

@@ -264,6 +264,10 @@ class ChatService {
         extendParams.reasoning_effort = chatConfig.reasoningEffort;
       }
 
+      if (modelExtendParams!.includes('thinking') && chatConfig.thinking) {
+        extendParams.thinking = { type: chatConfig.thinking };
+      }
+
       if (
         modelExtendParams!.includes('thinkingBudget') &&
         chatConfig.thinkingBudget !== undefined

package/src/types/agent/chatConfig.ts

@@ -26,6 +26,7 @@ export interface LobeAgentChatConfig {
   enableReasoningEffort?: boolean;
   reasoningBudgetToken?: number;
   reasoningEffort?: 'low' | 'medium' | 'high';
+  thinking?: 'disabled' | 'auto' | 'enabled';
   thinkingBudget?: number;
   /**
    * 禁用上下文缓存

package/src/types/aiModel.ts

@@ -121,6 +121,7 @@ export interface AIBaseModelCard {
    * whether model is legacy (deprecated but not removed yet)
    */
   legacy?: boolean;
+  maxOutput?: number;
   /**
    * who create this model
    */
@@ -148,6 +149,7 @@ export type ExtendParamsType =
   | 'enableReasoning'
   | 'disableContextCaching'
   | 'reasoningEffort'
+  | 'thinking'
   | 'thinkingBudget';
 
 export interface AiModelSettings {
@@ -207,6 +209,7 @@ export interface AITTSModelCard extends AIBaseModelCard {
      * the input pricing, e.g. $1 / 1M tokens
      */
     input?: number;
+    output?: number;
   };
   type: 'tts';
 }
@@ -222,6 +225,7 @@ export interface AISTTModelCard extends AIBaseModelCard {
      * the input pricing, e.g. $1 / 1M tokens
      */
     input?: number;
+    output?: number;
   };
   type: 'stt';
 }