diff --git a/LLMFarm.xcodeproj/project.pbxproj b/LLMFarm.xcodeproj/project.pbxproj
index 90f4a1a..601c119 100644
--- a/LLMFarm.xcodeproj/project.pbxproj
+++ b/LLMFarm.xcodeproj/project.pbxproj
@@ -11,6 +11,8 @@
 		150CAD022A3CE7B30015CD66 /* GPTNeox.swift in Sources */ = {isa = PBXBuildFile; fileRef = 150CAD002A3CE7B30015CD66 /* GPTNeox.swift */; };
 		15141E162A6438AE0060E767 /* Starcoder.swift in Sources */ = {isa = PBXBuildFile; fileRef = 15141E152A6438AE0060E767 /* Starcoder.swift */; };
 		15141E172A6438AE0060E767 /* Starcoder.swift in Sources */ = {isa = PBXBuildFile; fileRef = 15141E152A6438AE0060E767 /* Starcoder.swift */; };
+		15141E5C2A651F8C0060E767 /* model_setting_templates in Resources */ = {isa = PBXBuildFile; fileRef = 15141E5B2A651F8C0060E767 /* model_setting_templates */; };
+		15141E712A65799D0060E767 /* ModelSettingsTemplate.swift in Sources */ = {isa = PBXBuildFile; fileRef = 15141E702A65799D0060E767 /* ModelSettingsTemplate.swift */; };
 		155FBB9F2A460566004DD5AE /* GPTBase.swift in Sources */ = {isa = PBXBuildFile; fileRef = 155FBB9E2A460566004DD5AE /* GPTBase.swift */; };
 		155FBBA02A460566004DD5AE /* GPTBase.swift in Sources */ = {isa = PBXBuildFile; fileRef = 155FBB9E2A460566004DD5AE /* GPTBase.swift */; };
 		155FBBA22A48B5C1004DD5AE /* Replit.swift in Sources */ = {isa = PBXBuildFile; fileRef = 155FBBA12A48B5C0004DD5AE /* Replit.swift */; };
@@ -73,6 +75,16 @@
 /* End PBXBuildFile section */
 
 /* Begin PBXCopyFilesBuildPhase section */
+		15141E6B2A6577160060E767 /* Embed Frameworks */ = {
+			isa = PBXCopyFilesBuildPhase;
+			buildActionMask = 2147483647;
+			dstPath = "";
+			dstSubfolderSpec = 10;
+			files = (
+			);
+			name = "Embed Frameworks";
+			runOnlyForDeploymentPostprocessing = 0;
+		};
 		15E415CF2A18A2970057A15E /* CopyFiles */ = {
 			isa = PBXCopyFilesBuildPhase;
 			buildActionMask = 12;
@@ -87,6 +99,18 @@
 /* Begin PBXFileReference section */
 		150CAD002A3CE7B30015CD66 /* GPTNeox.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = GPTNeox.swift; sourceTree = "<group>"; };
 		15141E152A6438AE0060E767 /* Starcoder.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = Starcoder.swift; sourceTree = "<group>"; };
+		15141E362A6488A60060E767 /* Replit 3B.json */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = text.json; path = "Replit 3B.json"; sourceTree = "<group>"; };
+		15141E372A6488A60060E767 /* dolly v2 3b.json */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = text.json; path = "dolly v2 3b.json"; sourceTree = "<group>"; };
+		15141E382A6488A60060E767 /* orca mini 3b.json */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = text.json; path = "orca mini 3b.json"; sourceTree = "<group>"; };
+		15141E392A6488A60060E767 /* openllama 3b.json */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = text.json; path = "openllama 3b.json"; sourceTree = "<group>"; };
+		15141E3A2A6488A60060E767 /* rp-incite-base-v1-3b.json */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = text.json; path = "rp-incite-base-v1-3b.json"; sourceTree = "<group>"; };
+		15141E3B2A6488A60060E767 /* santacoder.json */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = text.json; path = santacoder.json; sourceTree = "<group>"; };
+		15141E3C2A6488A60060E767 /* saiga 7B.json */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = text.json; path = "saiga 7B.json"; sourceTree = "<group>"; };
+		15141E3D2A6488A60060E767 /* cerebras 2.7b.json */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = text.json; path = "cerebras 2.7b.json"; sourceTree = "<group>"; };
+		15141E3F2A6488A70060E767 /* stablelm-tuned-alpha-3b.json */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = text.json; path = "stablelm-tuned-alpha-3b.json"; sourceTree = "<group>"; };
+		15141E402A6488A70060E767 /* AI Dungeon 2 Classic.json */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = text.json; path = "AI Dungeon 2 Classic.json"; sourceTree = "<group>"; };
+		15141E5B2A651F8C0060E767 /* model_setting_templates */ = {isa = PBXFileReference; lastKnownFileType = folder; name = model_setting_templates; path = LLMFarm/model_setting_templates; sourceTree = "<group>"; };
+		15141E702A65799D0060E767 /* ModelSettingsTemplate.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = ModelSettingsTemplate.swift; sourceTree = "<group>"; };
 		155FBB9E2A460566004DD5AE /* GPTBase.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = GPTBase.swift; sourceTree = "<group>"; };
 		155FBBA12A48B5C0004DD5AE /* Replit.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = Replit.swift; sourceTree = "<group>"; };
 		1560D6162A2D1A3D00918330 /* AddChatView.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = AddChatView.swift; sourceTree = "<group>"; };
@@ -171,6 +195,23 @@
 /* End PBXFrameworksBuildPhase section */
 
 /* Begin PBXGroup section */
+		15141E332A64875F0060E767 /* model_setting_templates */ = {
+			isa = PBXGroup;
+			children = (
+				15141E402A6488A70060E767 /* AI Dungeon 2 Classic.json */,
+				15141E3D2A6488A60060E767 /* cerebras 2.7b.json */,
+				15141E372A6488A60060E767 /* dolly v2 3b.json */,
+				15141E392A6488A60060E767 /* openllama 3b.json */,
+				15141E382A6488A60060E767 /* orca mini 3b.json */,
+				15141E362A6488A60060E767 /* Replit 3B.json */,
+				15141E3A2A6488A60060E767 /* rp-incite-base-v1-3b.json */,
+				15141E3C2A6488A60060E767 /* saiga 7B.json */,
+				15141E3B2A6488A60060E767 /* santacoder.json */,
+				15141E3F2A6488A70060E767 /* stablelm-tuned-alpha-3b.json */,
+			);
+			path = model_setting_templates;
+			sourceTree = "<group>";
+		};
 		1567138A2A5B13C4002BE054 /* metal */ = {
 			isa = PBXGroup;
 			children = (
@@ -252,6 +293,7 @@
 		15F2107F2A191B190021F414 /* LLMFarm */ = {
 			isa = PBXGroup;
 			children = (
+				15141E332A64875F0060E767 /* model_setting_templates */,
 				15A3EDC42A2121D90030AF5A /* Chats */,
 				15A3EDC12A2121D90030AF5A /* Contacts */,
 				15A3EDBD2A2121D90030AF5A /* Settings */,
@@ -269,6 +311,7 @@
 				15A3EDBB2A2121C70030AF5A /* ContentView.swift */,
 				15A3EDD02A21266D0030AF5A /* SearchView.swift */,
 				156A48D12A52B65C007DF519 /* Orientation_helper.swift */,
+				15141E702A65799D0060E767 /* ModelSettingsTemplate.swift */,
 			);
 			path = LLMFarm;
 			sourceTree = "<group>";
@@ -293,6 +336,7 @@
 		E624E1F52958ECC800996CD1 = {
 			isa = PBXGroup;
 			children = (
+				15141E5B2A651F8C0060E767 /* model_setting_templates */,
 				1567138A2A5B13C4002BE054 /* metal */,
 				E6024EC229EA7BEB001FE9B0 /* Packages */,
 				E624E2102958F21E00996CD1 /* Info.plist */,
@@ -359,6 +403,7 @@
 				15F2107A2A191B190021F414 /* Sources */,
 				15F2107B2A191B190021F414 /* Frameworks */,
 				15F2107C2A191B190021F414 /* Resources */,
+				15141E6B2A6577160060E767 /* Embed Frameworks */,
 			);
 			buildRules = (
 			);
@@ -418,6 +463,7 @@
 			isa = PBXResourcesBuildPhase;
 			buildActionMask = 2147483647;
 			files = (
+				15141E5C2A651F8C0060E767 /* model_setting_templates in Resources */,
 				1567138D2A5B155D002BE054 /* ggml-metal.metal in Resources */,
 				15F210892A191B1B0021F414 /* Preview Assets.xcassets in Resources */,
 				15F210852A191B1B0021F414 /* Assets.xcassets in Resources */,
@@ -470,6 +516,7 @@
 				15F210A32A1921E40021F414 /* Math.swift in Sources */,
 				155FBBA02A460566004DD5AE /* GPTBase.swift in Sources */,
 				15A3EDCB2A2121D90030AF5A /* ModelsView.swift in Sources */,
+				15141E712A65799D0060E767 /* ModelSettingsTemplate.swift in Sources */,
 				15A3EDC92A2121D90030AF5A /* SettingsItem.swift in Sources */,
 				15F210A42A1921E40021F414 /* Utils.swift in Sources */,
 				15A480002A40D2A300B72DC3 /* GPT2.swift in Sources */,
@@ -560,7 +607,7 @@
 				LD_RUNPATH_SEARCH_PATHS = "@executable_path/Frameworks";
 				"LD_RUNPATH_SEARCH_PATHS[sdk=macosx*]" = "@executable_path/../Frameworks";
 				MACOSX_DEPLOYMENT_TARGET = 13.0;
-				MARKETING_VERSION = 0.2.2;
+				MARKETING_VERSION = 0.3.0;
 				PRODUCT_BUNDLE_IDENTIFIER = com.guinmoon.LLMFarm;
 				PRODUCT_NAME = "$(TARGET_NAME)";
 				PROVISIONING_PROFILE_SPECIFIER = "";
@@ -607,7 +654,7 @@
 				LD_RUNPATH_SEARCH_PATHS = "@executable_path/Frameworks";
 				"LD_RUNPATH_SEARCH_PATHS[sdk=macosx*]" = "@executable_path/../Frameworks";
 				MACOSX_DEPLOYMENT_TARGET = 13.0;
-				MARKETING_VERSION = 0.2.2;
+				MARKETING_VERSION = 0.3.0;
 				PRODUCT_BUNDLE_IDENTIFIER = com.guinmoon.LLMFarm;
 				PRODUCT_NAME = "$(TARGET_NAME)";
 				PROVISIONING_PROFILE_SPECIFIER = "";
diff --git a/LLMFarm/AIChatModel.swift b/LLMFarm/AIChatModel.swift
index d556418..2acc14f 100644
--- a/LLMFarm/AIChatModel.swift
+++ b/LLMFarm/AIChatModel.swift
@@ -88,17 +88,17 @@ final class AIChatModel: ObservableObject {
                     if (chat_config!["use_metal"] != nil){
                         model_context_param.use_metal = chat_config!["use_metal"] as! Bool
                     }
-                    try? self.chat?.loadModel(ModelInference.LLamaInference,contextParams: model_context_param)
+                    self.chat?.loadModel(ModelInference.LLamaInference,contextParams: model_context_param)
                 }else if chat_config!["model_inference"] as! String == "gptneox" {
-                    try? self.chat?.loadModel(ModelInference.GPTNeoxInference,contextParams: model_context_param)
+                    self.chat?.loadModel(ModelInference.GPTNeoxInference,contextParams: model_context_param)
                 }else if chat_config!["model_inference"] as! String == "gpt2" {
-                    try? self.chat?.loadModel(ModelInference.GPT2,contextParams: model_context_param)
+                    self.chat?.loadModel(ModelInference.GPT2,contextParams: model_context_param)
                     self.chat?.model.stop_words.append("<|endoftext|>")
                 }else if chat_config!["model_inference"] as! String == "replit" {
-                    try? self.chat?.loadModel(ModelInference.Replit,contextParams: model_context_param)
+                    self.chat?.loadModel(ModelInference.Replit,contextParams: model_context_param)
                     self.chat?.model.stop_words.append("<|endoftext|>")
                 }else if chat_config!["model_inference"] as! String == "starcoder" {
-                    try? self.chat?.loadModel(ModelInference.Starcoder,contextParams: model_context_param)
+                    self.chat?.loadModel(ModelInference.Starcoder,contextParams: model_context_param)
                     self.chat?.model.stop_words.append("<|endoftext|>")
                 }
             }
@@ -107,9 +107,9 @@ final class AIChatModel: ObservableObject {
                     model_lowercase.contains("alpaca") ||
                     model_lowercase.contains("vic") ||
                     model_lowercase.contains("orca")){
-                    try? self.chat?.loadModel(ModelInference.LLamaInference)
+                    self.chat?.loadModel(ModelInference.LLamaInference)
                 }else{
-                    try? self.chat?.loadModel(ModelInference.GPTNeoxInference)
+                    self.chat?.loadModel(ModelInference.GPTNeoxInference)
                 }
             }
             if self.chat?.model.context == nil{
@@ -120,7 +120,8 @@ final class AIChatModel: ObservableObject {
             print(model_sample_param)
             print(model_context_param)
 //Set prompt model if in config or try to set promt format by filename
-            if (chat_config!["prompt_format"] != nil && chat_config!["prompt_format"]! as! String != "auto"){
+            if (chat_config!["prompt_format"] != nil && chat_config!["prompt_format"]! as! String != "auto"
+                && chat_config!["prompt_format"]! as! String != "{{prompt}}"){
                 self.chat?.model.custom_prompt_format = chat_config!["prompt_format"]! as! String
                 self.chat?.model.promptFormat = .Custom
             }
diff --git a/LLMFarm/FileHelper.swift b/LLMFarm/FileHelper.swift
index 195d684..990b851 100644
--- a/LLMFarm/FileHelper.swift
+++ b/LLMFarm/FileHelper.swift
@@ -23,10 +23,68 @@ func get_avalible_models() -> [String]?{
         return res
     } catch {
         // failed to read directory – bad permissions, perhaps?
+        print(error)
     }
     return res
 }
 
+func parse_model_setting_template(template_path:String) -> ModelSettingsTemplate{
+    var tmp_template:ModelSettingsTemplate = ModelSettingsTemplate()
+    do{
+        let data = try Data(contentsOf: URL(fileURLWithPath: template_path), options: .mappedIfSafe)
+        let jsonResult = try JSONSerialization.jsonObject(with: data, options: .mutableLeaves)
+        let jsonResult_dict = jsonResult as? Dictionary<String, AnyObject>
+        if (jsonResult_dict!["template_name"] != nil){
+            tmp_template.template_name = jsonResult_dict!["template_name"] as! String
+        }else{
+            tmp_template.template_name = (template_path as NSString).lastPathComponent
+        }
+        if (jsonResult_dict!["model_inference"] != nil){
+            tmp_template.inference = jsonResult_dict!["model_inference"] as! String
+        }
+        if (jsonResult_dict!["prompt_format"] != nil){
+            tmp_template.prompt_format = jsonResult_dict!["prompt_format"] as! String
+        }
+        if (jsonResult_dict!["n_batch"] != nil){
+            tmp_template.n_batch = jsonResult_dict!["n_batch"] as! Int32
+        }
+        if (jsonResult_dict!["temp"] != nil){
+            tmp_template.temp = Float(jsonResult_dict!["temp"] as! Double)
+        }
+        if (jsonResult_dict!["top_k"] != nil){
+            tmp_template.top_k = jsonResult_dict!["top_k"] as! Int32
+        }
+        if (jsonResult_dict!["top_p"] != nil){
+            tmp_template.top_p = Float(jsonResult_dict!["top_p"] as! Double)
+        }
+        if (jsonResult_dict!["repeat_penalty"] != nil){
+            tmp_template.repeat_penalty = Float(jsonResult_dict!["repeat_penalty"] as! Double)
+        }
+        if (jsonResult_dict!["repeat_last_n"] != nil){
+            tmp_template.repeat_last_n = jsonResult_dict!["repeat_last_n"] as! Int32
+        }
+    }
+    catch {
+        print(error)
+    }
+    return tmp_template
+}
+
+func get_model_setting_templates() -> [ModelSettingsTemplate]{
+    var model_setting_templates: [ModelSettingsTemplate] = []
+    model_setting_templates.append(ModelSettingsTemplate())
+    do{
+        var templates_path=Bundle.main.resourcePath!.appending("/model_setting_templates")
+        let tenplate_files = try FileManager.default.contentsOfDirectory(atPath: templates_path)
+        for tenplate_file in tenplate_files {
+            model_setting_templates.append(parse_model_setting_template(template_path: templates_path+"/"+tenplate_file))
+        }
+    }
+    catch {
+        print(error)
+    }
+    return model_setting_templates
+}
 
 public func get_chat_info(_ chat_fname:String) -> Dictionary<String, AnyObject>? {
     do {
@@ -39,7 +97,7 @@ public func get_chat_info(_ chat_fname:String) -> Dictionary<String, AnyObject>?
         let jsonResult_dict = jsonResult as? Dictionary<String, AnyObject>
         return jsonResult_dict
     } catch {
-        
+        print(error)
     }
     return nil
 }
diff --git a/LLMFarm/LLMFarmApp.swift b/LLMFarm/LLMFarmApp.swift
index c6e241a..e5d5af9 100644
--- a/LLMFarm/LLMFarmApp.swift
+++ b/LLMFarm/LLMFarmApp.swift
@@ -26,6 +26,7 @@ struct LLMFarmApp: App {
         aiChatModel.stop_predict()
     }
     
+    
     var body: some Scene {
         WindowGroup {
             
diff --git a/LLMFarm/ModelSettingsTemplate.swift b/LLMFarm/ModelSettingsTemplate.swift
index 74b4a30..0030b6d 100644
--- a/LLMFarm/ModelSettingsTemplate.swift
+++ b/LLMFarm/ModelSettingsTemplate.swift
@@ -1,8 +1,29 @@
 //
-//  model_settings_stemplate.swift
+//  ModelSettingsTemplate.swift
 //  LLMFarm
 //
 //  Created by guinmoon on 17.07.2023.
 //
 
 import Foundation
+
+struct ModelSettingsTemplate: Hashable {
+    var template_name: String = "Custom"
+    var inference = "llama"
+    var context: Int32 = 2048
+    var n_batch: Int32 = 512
+    var temp: Float = 0.9
+    var top_k: Int32 = 40
+    var top_p: Float = 0.95
+    var repeat_last_n: Int32 = 64
+    var repeat_penalty: Float = 1.174
+    var prompt_format: String = "{{prompt}}"
+    
+    func hash(into hasher: inout Hasher) {
+        hasher.combine(template_name)
+    }
+    
+    static func == (lhs: ModelSettingsTemplate, rhs: ModelSettingsTemplate) -> Bool {
+        return lhs.template_name == rhs.template_name 
+    }
+}
diff --git a/LLMFarm/Settings/AddChatView.swift b/LLMFarm/Settings/AddChatView.swift
index c83be96..3fe60d9 100644
--- a/LLMFarm/Settings/AddChatView.swift
+++ b/LLMFarm/Settings/AddChatView.swift
@@ -68,7 +68,7 @@ struct AddChatView: View {
     @State private var model_top_p: Float = 0.95
     @State private var model_repeat_last_n: Int32 = 64
     @State private var model_repeat_penalty: Float = 1.1
-    @State private var prompt_format: String = "auto"
+    @State private var prompt_format: String = "{{prompt}}"
     @State private var numberOfThreads: Int32 = 0
     @State private var use_metal: Bool = false
     @State private var isImporting: Bool = false
@@ -77,8 +77,11 @@ struct AddChatView: View {
     private var chat_name: String = ""
     let bin_type = UTType(tag: "bin", tagClass: .filenameExtension, conformingTo: nil)
     
-    @State private var model_inference = "auto"
-    let model_inferences = ["auto","gptneox", "llama", "gpt2", "replit", "starcoder"]
+    @State private var model_settings_template:ModelSettingsTemplate = ModelSettingsTemplate()
+    let model_setting_templates = get_model_setting_templates()
+    
+    @State private var model_inference = "llama"
+    let model_inferences = ["gptneox", "llama", "gpt2", "replit", "starcoder"]
     
     @State private var model_icon: String = "ava0"
     let model_icons = ["ava0","ava1","ava2","ava3","ava4","ava5","ava6","ava7"]
@@ -141,9 +144,21 @@ struct AddChatView: View {
         }
     }
     
+    func apply_setting_template(template:ModelSettingsTemplate){
+        model_inference = template.inference
+        prompt_format = template.prompt_format
+        model_context = template.context
+        model_n_batch = template.n_batch
+        model_temp = template.temp
+        model_top_k = template.top_k
+        model_top_p = template.top_p
+        model_repeat_penalty = template.repeat_penalty
+        model_repeat_last_n = template.repeat_last_n
+    }
+    
     var body: some View {
         ZStack{
-//            Color("color_bg").edgesIgnoringSafeArea(.all)
+            //            Color("color_bg").edgesIgnoringSafeArea(.all)
             VStack{
                 
                 HStack{
@@ -258,17 +273,34 @@ struct AddChatView: View {
                             .padding(.top, 8)
                         
                         HStack{
-                            VStack {
-                                Picker("inference", selection: $model_inference) {
-                                    ForEach(model_inferences, id: \.self) {
-                                        Text($0)
-                                    }
+                            Text("Settings template:")
+                                .frame(maxWidth: .infinity, alignment: .leading)
+                            Picker("", selection: $model_settings_template) {
+                                ForEach(model_setting_templates, id: \.self) { template in
+                                    Text(template.template_name).tag(template)
                                 }
-                                .pickerStyle(.menu)
                             }
-                            .frame(maxWidth: .infinity, alignment: .trailing)
+                            .onChange(of: model_settings_template) { tmpl in
+                                apply_setting_template(template:model_settings_template)
+                            }
+                            .pickerStyle(.menu)
                         }
-                        .padding()
+                        .padding(.horizontal)
+                        .padding(.top, 8)
+                        
+                        HStack{
+                            Text("Inference:")
+                                .frame(maxWidth: .infinity, alignment: .leading)
+                            Picker("", selection: $model_inference) {
+                                ForEach(model_inferences, id: \.self) {
+                                    Text($0)
+                                }
+                            }
+                            .pickerStyle(.menu)
+                            //
+                        }
+                        .padding(.horizontal)
+                        .padding(.top, 8)
                         
                         HStack {
                             Toggle("Use Metal", isOn: $use_metal)
@@ -279,12 +311,16 @@ struct AddChatView: View {
                         VStack {
                             Text("Prompt format:")
                                 .frame(maxWidth: .infinity, alignment: .leading)
-                            TextField("size..", text: $prompt_format)
+                            TextField("size..", text: $prompt_format, axis: .vertical)
+                                .lineLimit(2)
+
+                                            .textFieldStyle(.roundedBorder)
                                 .frame( alignment: .leading)
-                                .multilineTextAlignment(.trailing)
-                                .textFieldStyle(.plain)
+                            //                                .multilineTextAlignment(.trailing)
+                            //                                .textFieldStyle(.plain)
                         }
-                        .padding()
+                        .padding(.horizontal)
+                        .padding(.top, 8)
                         
                         Divider()
                             .padding(.top, 8)
@@ -369,7 +405,7 @@ struct AddChatView: View {
                             .padding(.horizontal)
                             
                             HStack {
-                                Text("Repean last N:")
+                                Text("Repeat last N:")
                                     .frame(maxWidth: 75, alignment: .leading)
                                 TextField("count..", value: $model_repeat_last_n, format:.number)
                                     .frame( alignment: .leading)
@@ -395,8 +431,10 @@ struct AddChatView: View {
                             .padding(.horizontal)
                             
                             HStack{
+                                Text("Icon:")
+                                    .frame(maxWidth: .infinity, alignment: .leading)
                                 VStack {
-                                    Picker("icon", selection: $model_icon) {
+                                    Picker("", selection: $model_icon) {
                                         ForEach(model_icons, id: \.self) {
                                             Text($0)
                                         }
@@ -417,9 +455,9 @@ struct AddChatView: View {
     
 }
 //
-//struct AddChatView_Previews: PreviewProvider {
-//    static var previews: some View {
-//        AddChatView(add_chat_dialog: .constant(true),edit_chat_dialog:.constant(false))
-//            .preferredColorScheme(.dark)
-//    }
-//}
+struct AddChatView_Previews: PreviewProvider {
+    static var previews: some View {
+        AddChatView(add_chat_dialog: .constant(true),edit_chat_dialog:.constant(false),renew_chat_list: .constant({}))
+            .preferredColorScheme(.dark)
+    }
+}
diff --git a/LLMFarm/model_setting_templates/AI Dungeon 2 Classic.json b/LLMFarm/model_setting_templates/AI Dungeon 2 Classic.json
new file mode 100644
index 0000000..4178012
--- /dev/null
+++ b/LLMFarm/model_setting_templates/AI Dungeon 2 Classic.json	
@@ -0,0 +1,11 @@
+{
+  "model_inference" : "gpt2",
+  "temp" : 0.80000001192092896,
+  "prompt_format" : "{{prompt}}",
+  "top_k" : 100,
+  "icon" : "ava6",
+  "n_batch" : 1024,
+  "context" : 1024,
+  "top_p" : 0.94999998807907104,
+  "template_name" : "AI Dungeon 2 Classic"
+}
diff --git a/LLMFarm/model_setting_templates/Replit 3B.json b/LLMFarm/model_setting_templates/Replit 3B.json
new file mode 100644
index 0000000..a2b84f2
--- /dev/null
+++ b/LLMFarm/model_setting_templates/Replit 3B.json	
@@ -0,0 +1,12 @@
+{
+  "context" : 2048,
+  "top_k" : 1,
+  "n_batch" : 512,
+  "temp" : 0.8,
+  "template_name" : "Replit 3B",
+  "top_p" : 0.80000001192092896,
+  "prompt_format" : "{{prompt}}",
+  "repeat_last_n" : 128,
+  "repeat_penalty" : 1.174,
+  "use_metal" : false
+}
diff --git a/LLMFarm/model_setting_templates/cerebras 2.7b.json b/LLMFarm/model_setting_templates/cerebras 2.7b.json
new file mode 100644
index 0000000..a850e95
--- /dev/null
+++ b/LLMFarm/model_setting_templates/cerebras 2.7b.json	
@@ -0,0 +1,13 @@
+{
+  "top_k" : 40,
+  "n_batch" : 512,
+  "context" : 2048,
+  "top_p" : 0.94999998807907104,
+  "prompt_format" : "{{prompt}}",
+  "repeat_last_n" : 64,
+  "repeat_penalty" : 1.1000000238418579,
+  "temp" : 0.89999997615814209,
+  "model_inference" : "gpt2",
+  "use_metal" : false,
+  "template_name" : "cerebras 2.7B"
+}
diff --git a/LLMFarm/model_setting_templates/dolly v2 3b.json b/LLMFarm/model_setting_templates/dolly v2 3b.json
new file mode 100644
index 0000000..6cf874a
--- /dev/null
+++ b/LLMFarm/model_setting_templates/dolly v2 3b.json	
@@ -0,0 +1,10 @@
+{
+  "context" : 1024,
+  "temp" : 0.89999997615814209,
+  "top_p" : 0.94999998807907104,
+  "prompt_format" : "Below is an instruction that describes a task. Write a response that appropriately completes the request.### Instruction:{{prompt}}### Response:",
+  "top_k" : 40,
+  "model_inference" : "gptneox",
+  "n_batch" : 512,
+  "template_name" : "Dolly v2 3B"
+}
diff --git a/LLMFarm/model_setting_templates/openllama 3b.json b/LLMFarm/model_setting_templates/openllama 3b.json
new file mode 100644
index 0000000..6f04646
--- /dev/null
+++ b/LLMFarm/model_setting_templates/openllama 3b.json	
@@ -0,0 +1,11 @@
+{
+  "temp" : 0.8,
+  "context" : 1024,
+  "top_p" : 0.89999997615814209,
+  "top_k" : 80,
+  "model_inference" : "llama",
+  "n_batch" : 512,
+  "template_name" : "OpenLLaMA 3B 1T",
+  "prompt_format" : "Q: {{prompt}}\\nA:",
+  "icon" : "ava5"
+}
diff --git a/LLMFarm/model_setting_templates/orca mini 3b.json b/LLMFarm/model_setting_templates/orca mini 3b.json
new file mode 100644
index 0000000..cfef726
--- /dev/null
+++ b/LLMFarm/model_setting_templates/orca mini 3b.json	
@@ -0,0 +1,12 @@
+{
+  "context" : 1024,
+  "prompt_format" : "### User:\\n{{prompt}}\\n\\n### Response:\\n",
+  "template_name" : "ORCA mini 3B",
+  "temp" : 0.89999997615814209,
+  "model_inference" : "llama",
+  "top_p" : 0.94999998807907104,
+  "icon" : "ava0",
+  "n_batch" : 512,
+  "top_k" : 40,
+  "use_metal" : true
+}
diff --git a/LLMFarm/model_setting_templates/rp-incite-base-v1-3b.json b/LLMFarm/model_setting_templates/rp-incite-base-v1-3b.json
new file mode 100644
index 0000000..ee5060c
--- /dev/null
+++ b/LLMFarm/model_setting_templates/rp-incite-base-v1-3b.json
@@ -0,0 +1,12 @@
+{
+  "use_metal" : false,
+  "model_inference" : "gptneox",
+  "template_name" : "RedPajama incite base v1 3B",
+  "prompt_format" : "{{prompt}}",
+  "icon" : "ava3",
+  "temp" : 0.80000001192092896,
+  "n_batch" : 512,
+  "top_p" : 0.94999998807907104,
+  "context" : 1024,
+  "top_k" : 40
+}
diff --git a/LLMFarm/model_setting_templates/saiga 7B.json b/LLMFarm/model_setting_templates/saiga 7B.json
new file mode 100644
index 0000000..a9eba35
--- /dev/null
+++ b/LLMFarm/model_setting_templates/saiga 7B.json	
@@ -0,0 +1,14 @@
+{
+  "repeat_penalty" : 1.1000000238418579,
+  "use_metal" : false,
+  "icon" : "ava2",
+  "n_batch" : 512,
+  "prompt_format" : "Ты — Сайга, русскоязычный автоматический ассистент. Ты разговариваешь с людьми и помогаешь им.\\n<s>user:{{prompt}}<\/s>",
+  "top_k" : 40,
+  "context" : 2048,
+  "temp" : 0.40000000596046448,
+  "model_inference" : "llama",
+  "repeat_last_n" : 64,
+  "prompt_format" : "Saiga 7B",
+  "top_p" : 0.94999998807907104
+}
diff --git a/LLMFarm/model_setting_templates/santacoder.json b/LLMFarm/model_setting_templates/santacoder.json
new file mode 100644
index 0000000..0d36c34
--- /dev/null
+++ b/LLMFarm/model_setting_templates/santacoder.json
@@ -0,0 +1,13 @@
+{
+  "model_inference" : "starcoder",
+  "template_name" : "Santacoder",
+  "context" : 2048,
+  "n_batch" : 512,
+  "top_k" : 1,
+  "prompt_format" : "{{prompt}}",
+  "temp" : 0.89999997615814209,
+  "top_p" : 0.94999998807907104,
+  "use_metal" : false,
+  "repeat_last_n" : 64,
+  "repeat_penalty" : 1.1759999990463257
+}
diff --git a/LLMFarm/model_setting_templates/stablelm-tuned-alpha-3b.json b/LLMFarm/model_setting_templates/stablelm-tuned-alpha-3b.json
new file mode 100644
index 0000000..3df93fd
--- /dev/null
+++ b/LLMFarm/model_setting_templates/stablelm-tuned-alpha-3b.json
@@ -0,0 +1,11 @@
+{
+  "icon" : "ava0",
+  "top_k" : 40,
+  "temp" : 0.89999997615814209,
+  "n_batch" : 512,
+  "model_inference" : "gptneox",
+  "template_name" : "StableLM tuned 3B",
+  "context" : 1024,
+  "prompt_format" : "{{prompt}}",
+  "top_p" : 0.94999998807907104
+}
diff --git a/LLMFarm_core/AI.swift b/LLMFarm_core/AI.swift
index baba19c..14d745d 100644
--- a/LLMFarm_core/AI.swift
+++ b/LLMFarm_core/AI.swift
@@ -35,17 +35,22 @@ class AI {
     
     func loadModel(_ aiModel: ModelInference, contextParams: ModelContextParams = .default) {
         print("AI init")
-        switch aiModel {
-        case .LLamaInference:
-            model = try? LLaMa(path: self.modelPath, contextParams: contextParams)
-        case .GPTNeoxInference:
-            model = try? GPTNeoX(path: self.modelPath, contextParams: contextParams)
-        case .GPT2:
-            model = try? GPT2(path: self.modelPath, contextParams: contextParams)
-        case .Replit:
-            model = try? Replit(path: self.modelPath, contextParams: contextParams)
-        case .Starcoder:
-            model = try? Starcoder(path: self.modelPath, contextParams: contextParams)
+        do{
+            switch aiModel {
+            case .LLamaInference:
+                model = try LLaMa(path: self.modelPath, contextParams: contextParams)
+            case .GPTNeoxInference:
+                model = try GPTNeoX(path: self.modelPath, contextParams: contextParams)
+            case .GPT2:
+                model = try GPT2(path: self.modelPath, contextParams: contextParams)
+            case .Replit:
+                model = try Replit(path: self.modelPath, contextParams: contextParams)
+            case .Starcoder:
+                model = try Starcoder(path: self.modelPath, contextParams: contextParams)
+            }
+        }
+        catch {
+            print(error)
         }
     }
     
diff --git a/ModelTest/main.swift b/ModelTest/main.swift
index 9b1e233..d0155c0 100644
--- a/ModelTest/main.swift
+++ b/ModelTest/main.swift
@@ -37,11 +37,13 @@ func prompt_for_generation(_ instruction:String) -> String{
 
 func main(){
     print("Hello.")
-    var input_text = "State the meaning of life."
+    var input_text = "State the meaning of life. And tell about Stavropol."
+    
 //    let ai = AI(_modelPath: "/Users/guinmoon/Library/Containers/com.guinmoon.LLMFarm/Data/Documents/models/dolly-v2-3b-q5_1.bin",_chatName: "chat")
 //    try? ai.loadModel(ModelInference.GPTNeoxInference)
+//    ai.model.custom_prompt_format = "Below is an instruction that describes a task. Write a response that appropriately completes the request.### Instruction:{{prompt}}### Response:"
+//    ai.model.promptFormat = .Custom
 //    ai.model.promptFormat = .Dolly_b3
-    
 //    let ai = AI(_modelPath: "/Users/guinmoon/Library/Containers/com.guinmoon.LLMFarm/Data/Documents/models/AI-Dungeon-2-Classic.bin",_chatName: "chat")
 //    try? ai.loadModel(ModelInference.GPT2)
 //    ai.model.promptFormat = .None
@@ -50,17 +52,17 @@ func main(){
 //    try? ai.loadModel(ModelInference.Replit)
 //    ai.model.promptFormat = .None
     
-//    let ai = AI(_modelPath: "/Users/guinmoon/dev/alpaca_llama_etc/orca-mini-3b.ggmlv3.q4_0.bin",_chatName: "chat")
+    let ai = AI(_modelPath: "/Users/guinmoon/dev/alpaca_llama_etc/orca-mini-7b.ggmlv3.q2_K.bin",_chatName: "chat")
 //    let ai = AI(_modelPath: "/Users/guinmoon/dev/alpaca_llama_etc/orca-mini-7b.ggmlv3.q3_K_M.bin",_chatName: "chat")
-//    var params:ModelContextParams = .default
-//    params.use_metal = true
-//    try? ai.loadModel(ModelInference.LLamaInference,contextParams: params)
-//    ai.model.promptFormat = .LLaMa
+    var params:ModelContextParams = .default
+    params.use_metal = true
+    try? ai.loadModel(ModelInference.LLamaInference,contextParams: params)
+    ai.model.promptFormat = .LLaMa
     
-    let ai = AI(_modelPath: "/Users/guinmoon/dev/alpaca_llama_etc/santacoder-q8_0.bin",_chatName: "chat")
-    try? ai.loadModel(ModelInference.Starcoder)
-    ai.model.promptFormat = .None
-    input_text = "def quicksort"
+//    let ai = AI(_modelPath: "/Users/guinmoon/dev/alpaca_llama_etc/santacoder-q8_0.bin",_chatName: "chat")
+//    try? ai.loadModel(ModelInference.Starcoder)
+//    ai.model.promptFormat = .None
+//    input_text = "def quicksort"
     
     ai.model.contextParams.seed = 0;
 //    ai.model.promptStyle = .StableLM_Tuned
diff --git a/README.md b/README.md
index 6abe450..3f4f4dc 100644
--- a/README.md
+++ b/README.md
@@ -21,7 +21,6 @@ Also, when creating the application, the source codes from the repository [byron
 - [x] Metal for llama inference (MacOS and iOS)
 - [ ] Metal for other inference
 - [ ] Restore context state (now only chat history) 
-- [ ] Direct embedding input (LLaVA, etc.)
 - [ ] Model setting templates
 
 ## Inferences
@@ -30,7 +29,7 @@ Also, when creating the application, the source codes from the repository [byron
 - [x] [GPTNeoX](https://huggingface.co/docs/transformers/model_doc/gpt_neox)
 - [x] [Replit](https://huggingface.co/replit/replit-code-v1-3b)
 - [x] [GPT2](https://huggingface.co/docs/transformers/model_doc/gpt2) + [Cerebras](https://arxiv.org/abs/2304.03208)
-- [ ] [Starcoder(Santacoder)](https://huggingface.co/bigcode/santacoder)
+- [x] [Starcoder(Santacoder)](https://huggingface.co/bigcode/santacoder)
 - [ ] [RWKV](https://huggingface.co/docs/transformers/model_doc/rwkv)
 
 ## Getting Started
diff --git a/llmfarm_core.swift/Sources/llmfarm_core/gptneox/gptneox.cpp b/llmfarm_core.swift/Sources/llmfarm_core/gptneox/gptneox.cpp
index 2dc9187..6dd90d5 100644
--- a/llmfarm_core.swift/Sources/llmfarm_core/gptneox/gptneox.cpp
+++ b/llmfarm_core.swift/Sources/llmfarm_core/gptneox/gptneox.cpp
@@ -63,7 +63,7 @@ struct gpt_neox_layer {
 
 struct gpt_neox_hparams:gpt_base_hparams {
     int32_t n_vocab = 50257;
-    int32_t n_ctx   = 1024;
+    int32_t n_ctx   = 4096;
     int32_t n_embd  = 1024;
     int32_t n_head  = 32;
     int32_t n_layer = 16;
@@ -72,27 +72,11 @@ struct gpt_neox_hparams:gpt_base_hparams {
     int32_t ftype   = 1;
 };
 
-struct gpt_neox_model {
+struct gpt_neox_model:gpt_base_model {
     gpt_neox_hparams hparams;
-
-    // normalization
-    struct ggml_tensor * ln_f_g;
-    struct ggml_tensor * ln_f_b;
-
-    struct ggml_tensor * wte; // position embedding
-
     struct ggml_tensor * lmh_g; // language model head
     //struct ggml_tensor * lmh_b; // language model bias
-
     std::vector<gpt_neox_layer> layers;
-
-    // key + value memory
-    struct ggml_tensor * memory_k;
-    struct ggml_tensor * memory_v;
-
-    //
-    struct ggml_context * ctx;
-    std::map<std::string, struct ggml_tensor *> tensors;
 };
 
 
@@ -204,10 +188,10 @@ bool gpt_neox_model_load(const std::string & fname, gpt_neox_model & model, gpt_
     {
         const auto & hparams = model.hparams;
 
-        const int n_embd  = hparams.n_embd;
-        const int n_layer = hparams.n_layer;
-        const int n_ctx   = hparams.n_ctx;
-        const int n_vocab = hparams.n_vocab;
+        const size_t n_embd  = hparams.n_embd;
+        const size_t n_layer = hparams.n_layer;
+        const size_t n_ctx   = hparams.n_ctx;
+        const size_t n_vocab = hparams.n_vocab;
 
         ctx_size += n_embd*ggml_type_sizef(GGML_TYPE_F32); // ln_f_g
         ctx_size += n_embd*ggml_type_sizef(GGML_TYPE_F32); // ln_f_b
@@ -238,7 +222,8 @@ bool gpt_neox_model_load(const std::string & fname, gpt_neox_model & model, gpt_
         ctx_size += n_ctx*n_layer*n_embd*ggml_type_sizef(GGML_TYPE_F32); // memory_k
         ctx_size += n_ctx*n_layer*n_embd*ggml_type_sizef(GGML_TYPE_F32); // memory_v
 
-        ctx_size += (6 + 16*n_layer)*512; // object overhead
+        size_t overhead =ggml_tensor_overhead();
+        ctx_size += (6 + 16*n_layer)*1024; // object overhead
 
         printf("%s: ggml ctx size = %6.2f MB\n", __func__, ctx_size/(1024.0*1024.0));
     }
@@ -489,6 +474,7 @@ bool gpt_neox_eval(
     const int n_rot   = hparams.n_rot;
 
     static size_t buf_size = 256u*1024*1024;
+//    static size_t buf_size = 256u*1024*ggml_tensor_overhead();
     static void * buf = malloc(buf_size);
 
     // use 2 scratch buffers
diff --git a/llmfarm_core.swift/Sources/llmfarm_core/replit/replit.cpp b/llmfarm_core.swift/Sources/llmfarm_core/replit/replit.cpp
index 4ba5b0c..c4fdfc8 100644
--- a/llmfarm_core.swift/Sources/llmfarm_core/replit/replit.cpp
+++ b/llmfarm_core.swift/Sources/llmfarm_core/replit/replit.cpp
@@ -166,37 +166,14 @@ struct replit_layer {
     struct ggml_tensor * ffn_down_proj;
 };
 
-struct replit_model {
+struct replit_model:gpt_base_model {
     replit_hparams hparams;
-
-    struct gpt_kv_cache kv_self;
-    
     struct ggml_tensor * wte_weight;    // position embedding
     struct ggml_tensor * norm_f_weight; // language model head
-
     std::vector<replit_layer> layers;
-
-    // key + value memory
-    struct ggml_tensor * memory_k;
-    struct ggml_tensor * memory_v;
-
-    struct ggml_context * ctx;
-    std::map<std::string, struct ggml_tensor *> tensors;
-     ~replit_model() {
-        if (ctx) {
-            ggml_free(ctx);
-        }
-    }
 };
 
 
-
-
-//struct replit_model:gpt_base_model {
-//    mpt_hparams hparams;
-//    std::vector<replit_layer> layers;
-//};
-
 struct replit_context:gpt_base_context {
     replit_model model;
     replit_tokenizer vocab;
diff --git a/llmfarm_core.swift/Sources/llmfarm_core/starcoder/starcoder.cpp b/llmfarm_core.swift/Sources/llmfarm_core/starcoder/starcoder.cpp
index 7ac708d..36bcbe7 100644
--- a/llmfarm_core.swift/Sources/llmfarm_core/starcoder/starcoder.cpp
+++ b/llmfarm_core.swift/Sources/llmfarm_core/starcoder/starcoder.cpp
@@ -54,26 +54,9 @@ struct starcoder_layer {
     struct ggml_tensor * c_mlp_proj_b;
 };
 
-struct starcoder_model {
+struct starcoder_model:gpt_base_model {
     starcoder_hparams hparams;
-
-    // normalization
-    struct ggml_tensor * ln_f_g;
-    struct ggml_tensor * ln_f_b;
-
-    struct ggml_tensor * wte;     // position embedding
-    struct ggml_tensor * wpe;     //    token embedding
-    struct ggml_tensor * lm_head; // language model head
-
     std::vector<starcoder_layer> layers;
-
-    // key + value memory
-    struct ggml_tensor * memory_k;
-    struct ggml_tensor * memory_v;
-
-    //
-    struct ggml_context * ctx;
-    std::map<std::string, struct ggml_tensor *> tensors;
 };