chore: try using arg --chat-template-kwargs

extensions/llamacpp-extension/src/index.ts

@@ -58,6 +58,7 @@ type LlamacppConfig = {
   ubatch_size: number
   device: string
   split_mode: string
+  chat_template_kwargs: string
   main_gpu: number
   flash_attn: boolean
   cont_batching: boolean
@@ -1393,6 +1394,9 @@ export default class llamacpp_extension extends AIEngine {
     if (!cfg.reasoning_budget) {
       args.push('--reasoning-budget', '0')
     }
+    if (cfg.chat_template_kwargs) {
+      args.push('--chat-template-kwargs', cfg.chat_template_kwargs)
+    }
     if (cfg.override_tensor_buffer_t)
       args.push('--override-tensor', cfg.override_tensor_buffer_t)
     // offload multimodal projector model to the GPU by default. if there is not enough memory

web-app/src/containers/ChatInput.tsx

@@ -721,7 +721,7 @@ const ChatInput = ({ model, className, initialMessage }: ChatInputProps) => {
                             <div className="h-7 mr-1 text-sm rounded-sm text-main-view-fg/70 cursor-pointer flex items-center gap-1">
                               <span className="capitalize">
                                 {selectedModel?.reasoning?.reasoning_effort ||
-                                  'auto'}
+                                  'medium'}
                               </span>
                               {reasoningEffortOpen ? (
                                 <IconChevronUp
@@ -738,23 +738,30 @@ const ChatInput = ({ model, className, initialMessage }: ChatInputProps) => {
                           </PopoverTrigger>
                           <PopoverContent className="w-24 p-0" align="start">
                             <div className="py-1">
-                              {['auto', 'low', 'medium', 'high'].map(
+                              {['low', 'medium', 'high'].map((effort) => (
-                                (effort) => (
+                                <div
-                                  <div
+                                  key={effort}
-                                    key={effort}
+                                  className="px-2 py-1.5 text-sm cursor-pointer hover:bg-main-view-fg/20 capitalize"
-                                    className="px-2 py-1.5 text-sm cursor-pointer hover:bg-main-view-fg/20 capitalize"
+                                  onClick={async () => {
-                                    onClick={async () => {
+                                    if (selectedModel?.reasoning) {
-                                      if (selectedModel?.reasoning) {
+                                      selectedModel.reasoning.reasoning_effort =
-                                        selectedModel.reasoning.reasoning_effort =
+                                        effort
-                                          effort
+                                      setReasoningEffortOpen(false)
-                                        setReasoningEffortOpen(false)
+                                      // Restart model with new reasoning effort
+                                      try {
+                                        await stopModel(selectedModel.id)
+                                      } catch (error) {
+                                        console.error(
+                                          'Error restarting model with new reasoning effort:',
+                                          error
+                                        )
                                       }
-                                    }}
+                                    }
-                                  >
+                                  }}
-                                    {effort}
+                                >
-                                  </div>
+                                  {effort}
-                                )
+                                </div>
-                              )}
+                              ))}
                             </div>
                           </PopoverContent>
                         </Popover>

web-app/src/hooks/useChat.ts

@@ -289,10 +289,6 @@ export const useChat = () => {
             {
               ...modelSettings,
               ...currentAssistant.parameters,
-              ...(selectedModel?.reasoning?.reasoning_budget &&
-                selectedModel?.reasoning?.reasoning_effort !== 'auto' && {
-                  reasoning_effort: selectedModel?.reasoning?.reasoning_effort,
-                }),
             } as unknown as Record<string, object>
           )
 

web-app/src/services/models.ts

@@ -427,6 +427,10 @@ export const startModel = async (
     ...(modelConfig?.reasoning?.reasoning_budget !== undefined && {
       reasoning_budget: modelConfig.reasoning.reasoning_budget,
     }),
+    ...(modelConfig?.reasoning?.reasoning_budget &&
+      modelConfig?.reasoning?.reasoning_effort !== undefined && {
+        chat_template_kwargs: `{"reasoning_effort":"${modelConfig.reasoning.reasoning_effort}"}`,
+      }),
   }
 
   return engine.load(model, settings).catch((error) => {

web-app/src/services/providers.ts

@@ -104,7 +104,7 @@ export const getProviders = async (): Promise<ModelProvider[]> => {
                       model.id,
                       'reasoning_effort'
                     ))
-                      ? { reasoning_effort: 'auto' }
+                      ? { reasoning_effort: 'medium' }
                       : {}),
                   },
                 }