fix(ollama): increase timeout and add default max_tokens for large models

- Increase Ollama timeout to 5m for larger models (e.g. gemma4)
- Set default max_tokens to 4096 for common Ollama models
- Expand stream scanner buffer to 10MB to prevent truncation
- Improve model routing and prefix stripping in server

internal/providers/ollama.go

@@ -21,8 +21,8 @@ type OllamaProvider struct {
 
 func NewOllamaProvider(cfg config.OllamaConfig) *OllamaProvider {
 	client := resty.New()
-	// Set reasonable timeouts for local Ollama server
-	client.SetTimeout(30 * time.Second)
+	// Set reasonable timeouts for local Ollama server (longer for larger models)
+	client.SetTimeout(5 * time.Minute)
 	client.SetRetryCount(2)
 	client.SetRetryWaitTime(1 * time.Second)
 	
@@ -118,9 +118,15 @@ func BuildOllamaBody(request *models.UnifiedRequest, messagesJSON []interface{},
 	if request.Temperature != nil {
 		body["temperature"] = *request.Temperature
 	}
+	
 	if request.MaxTokens != nil {
 		body["max_tokens"] = *request.MaxTokens
+	} else if strings.Contains(request.Model, "gemma") || strings.Contains(request.Model, "llama") || strings.Contains(request.Model, "mistral") || strings.Contains(request.Model, "qwen") {
+		// Default to 4096 for common Ollama models if not specified, 
+		// as Ollama's compatibility layer sometimes defaults to 128
+		body["max_tokens"] = 4096
 	}
+
 	if request.TopP != nil {
 		body["top_p"] = *request.TopP
 	}
@@ -196,6 +202,11 @@ func ParseOllamaStreamChunk(line string) (*models.ChatCompletionStreamResponse,
 func StreamOllama(ctx io.ReadCloser, ch chan<- *models.ChatCompletionStreamResponse, model string) error {
 	defer ctx.Close()
 	scanner := bufio.NewScanner(ctx)
+	// Set a larger buffer for scanning to handle large chunks if they occur
+	const maxCapacity = 10 * 1024 * 1024 // 10MB
+	buf := make([]byte, 64*1024)
+	scanner.Buffer(buf, maxCapacity)
+
 	for scanner.Scan() {
 		line := scanner.Text()
 		chunk, done, err := ParseOllamaStreamChunk(line)