fix: update reasoningTimePlugin and smoothReasoningPlugin for improved performance tracking

- Changed the invocation of `reasoningTimePlugin` to a direct reference in `ModernAiProvider`.
- Initialized `thinkingStartTime` with `performance.now()` in `reasoningTimePlugin` for accurate timing.
- Removed `thinking_millsec` from the enqueued chunks in `smoothReasoningPlugin` to streamline data handling.
- Added console logging for performance tracking in `reasoningTimePlugin` to aid in debugging.

src/renderer/src/aiCore/index_new.ts

@@ -133,7 +133,7 @@ export default class ModernAiProvider {
           delayInMs: 80,
           chunkingRegex: /([\u4E00-\u9FFF]{3})|\S+\s+/
         }),
-        reasoningTimePlugin()
+        reasoningTimePlugin
       )
     }
 

src/renderer/src/aiCore/plugins/reasoningTimePlugin.ts

@@ -1,9 +1,9 @@
 import { definePlugin } from '@cherrystudio/ai-core'
 
-export default definePlugin(() => ({
+export default definePlugin({
   name: 'reasoningTimePlugin',
   transformStream: () => () => {
-    let thinkingStartTime = 0
+    let thinkingStartTime = performance.now()
     let hasStartedThinking = false
     let accumulatedThinkingContent = ''
     return new TransformStream({
@@ -11,9 +11,10 @@ export default definePlugin(() => ({
         if (chunk.type === 'reasoning') {
           if (!hasStartedThinking) {
             hasStartedThinking = true
-            thinkingStartTime = performance.now()
+            // thinkingStartTime = performance.now()
           }
           accumulatedThinkingContent += chunk.textDelta
+          console.log('performance.now() - thinkingStartTime', performance.now() - thinkingStartTime)
           controller.enqueue({
             ...chunk,
             thinking_millsec: performance.now() - thinkingStartTime
@@ -34,4 +35,4 @@ export default definePlugin(() => ({
       }
     })
   }
-}))
+})

src/renderer/src/aiCore/plugins/smoothReasoningPlugin.ts

@@ -19,7 +19,7 @@ export default definePlugin(({ delayInMs, chunkingRegex }: { delayInMs: number;
       async transform(chunk, controller) {
         if (chunk.type !== 'reasoning') {
           if (buffer.length > 0) {
-            controller.enqueue({ type: 'reasoning', textDelta: buffer, thinking_millsec: chunk.thinking_millsec })
+            controller.enqueue({ type: 'reasoning', textDelta: buffer })
             buffer = ''
           }
 
@@ -31,7 +31,7 @@ export default definePlugin(({ delayInMs, chunkingRegex }: { delayInMs: number;
         let match
 
         while ((match = detectChunk(buffer)) != null) {
-          controller.enqueue({ type: 'reasoning', textDelta: match, thinking_millsec: chunk.thinking_millsec })
+          controller.enqueue({ type: 'reasoning', textDelta: match })
           buffer = buffer.slice(match.length)
 
           await delay(delayInMs)

src/renderer/src/services/ApiService.ts

@@ -306,12 +306,13 @@ export async function fetchChatCompletion({
     enableTools: isEnabledToolUse(assistant),
     requestOptions: options
   })
-
+  console.log('assistant.settings?.reasoning_effort', assistant.settings?.reasoning_effort)
   const middlewareConfig: AiSdkMiddlewareConfig = {
     streamOutput: assistant.settings?.streamOutput ?? true,
     onChunk: onChunkReceived,
     model: assistant.model,
     provider: provider,
+    // FIXME: 这里需要根据模型来判断是否启用推理
     enableReasoning: assistant.settings?.reasoning_effort !== undefined,
     enableTool: assistant.settings?.toolUseMode === 'prompt',
     mcpTools