fix: prompt token

extensions/llamacpp-extension/src/index.ts

@@ -332,12 +332,14 @@ export default class llamacpp_extension extends AIEngine {
           )
           // Clear the invalid stored preference
           this.clearStoredBackendType()
-          bestAvailableBackendString =
-            await this.determineBestBackend(version_backends)
+          bestAvailableBackendString = await this.determineBestBackend(
+            version_backends
+          )
         }
       } else {
-        bestAvailableBackendString =
-          await this.determineBestBackend(version_backends)
+        bestAvailableBackendString = await this.determineBestBackend(
+          version_backends
+        )
       }
 
       let settings = structuredClone(SETTINGS)
@@ -2151,7 +2153,12 @@ export default class llamacpp_extension extends AIEngine {
     if (mmprojPath && !this.isAbsolutePath(mmprojPath))
       mmprojPath = await joinPath([await getJanDataFolderPath(), path])
     try {
-      const result = await planModelLoadInternal(path, this.memoryMode, mmprojPath, requestedCtx)
+      const result = await planModelLoadInternal(
+        path,
+        this.memoryMode,
+        mmprojPath,
+        requestedCtx
+      )
       return result
     } catch (e) {
       throw new Error(String(e))
@@ -2279,12 +2286,18 @@ export default class llamacpp_extension extends AIEngine {
     }
 
     // Calculate text tokens
-    const messages = JSON.stringify({ messages: opts.messages })
+    // Use chat_template_kwargs from opts if provided, otherwise default to disable enable_thinking
+    const tokenizeRequest = {
+      messages: opts.messages,
+      chat_template_kwargs: opts.chat_template_kwargs || {
+        enable_thinking: false,
+      },
+    }
 
     let parseResponse = await fetch(`${baseUrl}/apply-template`, {
       method: 'POST',
       headers: headers,
-      body: messages,
+      body: JSON.stringify(tokenizeRequest),
     })
 
     if (!parseResponse.ok) {

web-app/src/services/models/default.ts

@@ -31,8 +31,8 @@
   }
 
   async getModel(modelId: string): Promise<modelInfo | undefined> {
     return this.getEngine()?.get(modelId)
   }
 
   async fetchModels(): Promise<modelInfo[]> {
     return this.getEngine()?.list() ?? []
@@ -78,9 +78,9 @@
         `https://huggingface.co/api/models/${cleanRepoId}?blobs=true&files_metadata=true`,
         {
           headers: hfToken
             ? {
                 Authorization: `Bearer ${hfToken}`,
               }
             : {},
         }
       )
@@ -139,13 +139,13 @@
 
     // Convert mmproj files to mmproj_models format
     const mmprojModels = mmprojFiles.map((file) => {
       const modelId = file.rfilename.replace(/\.gguf$/i, '')
 
       return {
         model_id: sanitizeModelId(modelId),
         path: `https://huggingface.co/${repo.modelId}/resolve/main/${file.rfilename}`,
         file_size: formatFileSize(file.size),
       }
     })
 
     return {
@@ -193,30 +193,30 @@
   }
 
   async pullModelWithMetadata(
     id: string,
     modelPath: string,
     mmprojPath?: string,
     hfToken?: string
   ): Promise<void> {
     let modelSha256: string | undefined
     let modelSize: number | undefined
     let mmprojSha256: string | undefined
     let mmprojSize: number | undefined
 
     // Extract repo ID from model URL
     // URL format: https://huggingface.co/{repo}/resolve/main/{filename}
     const modelUrlMatch = modelPath.match(
       /https:\/\/huggingface\.co\/([^/]+\/[^/]+)\/resolve\/main\/(.+)/
     )
 
     if (modelUrlMatch) {
       const [, repoId, modelFilename] = modelUrlMatch
 
       try {
         // Fetch real-time metadata from HuggingFace
         const repoInfo = await this.fetchHuggingFaceRepo(repoId, hfToken)
 
         if (repoInfo?.siblings) {
           // Find the specific model file
           const modelFile = repoInfo.siblings.find(
             (file) => file.rfilename === modelFilename
@@ -578,6 +578,9 @@
                   }
                 }>
           }>
+          chat_template_kwargs?: {
+            enable_thinking: boolean
+          }
         }) => Promise<number>
       }
 
@@ -654,6 +657,9 @@
         return await engine.getTokensCount({
           model: modelId,
           messages: transformedMessages,
+          chat_template_kwargs: {
+            enable_thinking: false,
+          },
         })
       }