update default models gpt4o

packages/grafana-llm-app/README.md

@@ -224,7 +224,7 @@ const MyComponent = (): JSX.Element => {
     // Stream the completions. Each element is the next stream chunk.
     const stream = llms.openai
       .streamChatCompletions({
-        model: 'gpt-3.5-turbo',
+        model: llms.openai.Model.BASE,
         messages: [
           { role: 'system', content: 'You are a cynical assistant.' },
           { role: 'user', content: message },

packages/grafana-llm-app/llmclient/llmclient.go

@@ -23,9 +23,9 @@ const (
 type Model string
 
 const (
-	// ModelBase is the base model, for efficient and high-throughput tasks. OpenAI default: gpt-3.5-turbo
+	// ModelBase is the base model, for efficient and high-throughput tasks. OpenAI default: gpt-4o-mini
 	ModelBase = "base"
-	// ModelLarge is the large model, for more advanced tasks with longer context windows. OpenAI default: gpt-4-turbo
+	// ModelLarge is the large model, for more advanced tasks with longer context windows. OpenAI default: gpt-4o
 	ModelLarge = "large"
 )
 

packages/grafana-llm-app/pkg/plugin/llm_provider.go

@@ -25,9 +25,9 @@ const (
 // compatability, or the new abstract model names.
 func ModelFromString(m string) (Model, error) {
 	switch {
-	case m == ModelLarge || strings.HasPrefix(m, "gpt-4"):
+	case m == ModelLarge || (strings.HasPrefix(m, "gpt-4") && !strings.Contains(m, "-mini")):
 		return ModelLarge, nil
-	case m == ModelBase || strings.HasPrefix(m, "gpt-3.5"):
+	case m == ModelBase || strings.HasPrefix(m, "gpt-3.5") || strings.Contains(m, "-mini"):
 		return ModelBase, nil
 	}
 	// TODO: Give users the ability to specify a default model abstraction in settings, and use that here.
@@ -54,9 +54,9 @@ func (m Model) toOpenAI(modelSettings *ModelSettings) string {
 	if modelSettings == nil || len(modelSettings.Mapping) == 0 {
 		switch m {
 		case ModelBase:
-			return "gpt-3.5-turbo"
+			return "gpt-4o-mini"
 		case ModelLarge:
-			return "gpt-4-turbo"
+			return "gpt-4o"
 		}
 		panic(fmt.Sprintf("unrecognized model: %s", m))
 	}

packages/grafana-llm-app/pkg/plugin/llm_provider_test.go

@@ -49,6 +49,16 @@ func TestModelFromString(t *testing.T) {
 			expected: ModelBase,
 			wantErr:  false,
 		},
+		{
+			input:    "gpt-4o-mini",
+			expected: ModelBase,
+			wantErr:  false,
+		},
+		{
+			input:    "gpt-4o-mini-2024-07-18",
+			expected: ModelBase,
+			wantErr:  false,
+		},
 		{
 			input:    "gpt-4-turbo",
 			expected: ModelLarge,
@@ -64,6 +74,11 @@ func TestModelFromString(t *testing.T) {
 			expected: ModelLarge,
 			wantErr:  false,
 		},
+		{
+			input:    "gpt-4o",
+			expected: ModelLarge,
+			wantErr:  false,
+		},
 		{
 			input:    "gpt-4-32k-0613",
 			expected: ModelLarge,

packages/grafana-llm-app/pkg/plugin/settings.go

@@ -100,8 +100,8 @@ func (c ModelSettings) getModel(model Model) string {
 var DEFAULT_MODEL_SETTINGS = &ModelSettings{
 	Default: ModelBase,
 	Mapping: map[Model]string{
-		ModelBase:  "gpt-3.5-turbo",
-		ModelLarge: "gpt-4-turbo",
+		ModelBase:  "gpt-4o-mini",
+		ModelLarge: "gpt-4o",
 	},
 }
 

packages/grafana-llm-app/src/README.md

@@ -78,7 +78,7 @@ const MyComponent = (): JSX.Element => {
     // Stream the completions. Each element is the next stream chunk.
     const stream = llms.openai
       .streamChatCompletions({
-        model: 'gpt-3.5-turbo',
+        model: llms.openai.Model.BASE
         messages: [
           { role: 'system', content: 'You are a cynical assistant.' },
           { role: 'user', content: message },

packages/grafana-llm-app/src/components/AppConfig/ModelConfig.tsx

@@ -22,13 +22,13 @@ const DEFAULT_MODEL_ID = openai.Model.BASE;
 const MODEL_MAPPING_CONFIG: ModelMappingConfig[] = [
   {
     id: openai.Model.BASE,
-    name: 'gpt-3.5-turbo',
+    name: 'gpt-4o-mini',
     label: 'Base',
     description: 'A fast and cost-effective model for efficient, high-throughput tasks.',
   },
   {
     id: openai.Model.LARGE,
-    name: 'gpt-4-turbo',
+    name: 'gpt-4o',
     label: 'Large',
     description: 'A larger, higher cost model for more advanced tasks with longer context windows.',
   },