Merge pull request #13 from undreamai/release/v1.0.1

v1.0.1

Author: amakropoulos

.github/GameObject.png

@@ -0,0 +1,7 @@
+## v1.0.1
+- Fix running commands for projects with space in path
+  -  closes #8
+  -  closes #9
+- Fix sample scenes for different screen resolutions
+  -  closes #10
+- Allow parallel prompts

CHANGELOG.md

@@ -63,8 +63,10 @@ public override void OnInspectorGUI()
             ShowProgress(llmScript.modelCopyProgress, "Model Copying");
             if (llmScript.model != ""){
                 AddModelSettings(llmScriptSO, false);
-                AddChatSettings(llmScriptSO);
+            } else {
+                EditorGUILayout.Space();
             }
+            AddChatSettings(llmScriptSO);
             GUI.enabled = true;
 
             EditorGUI.EndChangeCheck();

CHANGELOG.md.meta

@@ -168,6 +168,8 @@ The server can be either a LLMUnity server or a standard [llama.cpp server](http
 
 ## Options
 
+- `Show/Hide Advanced Options` Toggle to show/hide advanced options from below
+
 #### :computer: Server Settings
 
 <div>
@@ -178,24 +180,27 @@ The server can be either a LLMUnity server or a standard [llama.cpp server](http
 - `Num GPU Layers` number of model layers to offload to the GPU.
 If set to 0 the GPU is not used. Use a large number i.e. >30 to utilise the GPU as much as possible.<br>
 If the user's GPU is not supported, the LLM will fall back to the CPU
-- `Debug` select to log the output of the model in the Unity Editor
-- `Port` port to run the server
 - `Stream` select to receive the reply from the model as it is produced (recommended!).<br>
 If it is not selected, the full reply from the model is received in one go
+- Advanced options:
+  - `Parallel Prompts` number of prompts that can happen in parallel (default: -1 = number of LLM/LLMClient objects)
+  - `Debug` select to log the output of the model in the Unity Editor
+  - `Port` port to run the server
 
 #### :hugs: Model Settings
 - `Download model` click to download the default model (Mistral 7B Instruct)
 - `Load model` click to load your own model in .gguf format
 - `Load lora` click to load a LORA model in .bin format
 - `Model` the model being used (inside the Assets/StreamingAssets folder)
 - `Lora` the LORA model being used (inside the Assets/StreamingAssets folder)
-- `Context Size` Size of the prompt context (0 = context size of the model)
-- `Batch Size` Batch size for prompt processing (default: 512)
-- `Seed` seed for reproducibility. For random results every time select -1
-- `Temperature` LLM temperature, lower values give more deterministic answers
-- `Top K` top-k sampling (default: 40, 0 = disabled)
-- `Top P` top-p sampling (default: 0.9, 1.0 = disabled)
-- `Num Predict` number of tokens to predict (default: 256, -1 = infinity, -2 = until context filled)
+- Advanced options:
+  - `Context Size` Size of the prompt context (0 = context size of the model)
+  - `Batch Size` Batch size for prompt processing (default: 512)
+  - `Seed` seed for reproducibility. For random results every time select -1
+  - `Temperature` LLM temperature, lower values give more deterministic answers
+  - `Top K` top-k sampling (default: 40, 0 = disabled)
+  - `Top P` top-p sampling (default: 0.9, 1.0 = disabled)
+  - `Num Predict` number of tokens to predict (default: 256, -1 = infinity, -2 = until context filled)
 
 #### :left_speech_bubble: Chat Settings
 - `Player Name` the name of the player

Editor/LLMEditor.cs

@@ -9,12 +9,14 @@
 
 namespace LLMUnity
 {
+    [DefaultExecutionOrder(-2)]
     public class LLM : LLMClient
     {
         [HideInInspector] public bool modelHide = true;
 
         [Server] public int numThreads = -1;
         [Server] public int numGPULayers = 0;
+        [ServerAdvanced] public int parallelPrompts = -1;
         [ServerAdvanced] public bool debug = false;
 
         [Model] public string model = "";
@@ -36,7 +38,7 @@ public class LLM : LLMClient
         private static float binariesDone = 0;
         private Process process;
         private bool serverListening = false;
-        private static ManualResetEvent serverStarted = new ManualResetEvent(false);
+        public ManualResetEvent serverStarted = new ManualResetEvent(false);
 
         private static string GetAssetPath(string relPath=""){
             // Path to store llm server binaries and models
@@ -95,7 +97,7 @@ public async void SetLora(string path){
         }
         #endif
 
-        new void OnEnable()
+        new public void OnEnable()
         {
             // start the llm server and run the OnEnable of the client
             StartLLMServer();
@@ -156,16 +158,17 @@ private void StartLLMServer()
                 if (!File.Exists(loraPath)) throw new System.Exception($"File {loraPath} not found!");
             }
 
+            int slots = parallelPrompts == -1? FindObjectsOfType<LLMClient>().Length: parallelPrompts;
             string binary = server;
-            string arguments = $" --port {port} -m {modelPath} -c {contextSize} -b {batchSize} --log-disable --nobrowser";
+            string arguments = $" --port {port} -m \"{modelPath}\" -c {contextSize} -b {batchSize} --log-disable --nobrowser -np {slots}";
             if (numThreads > 0) arguments += $" -t {numThreads}";
             if (numGPULayers > 0) arguments += $" -ngl {numGPULayers}";
-            if (loraPath != "") arguments += $" --lora {loraPath}";
+            if (loraPath != "") arguments += $" --lora \"{loraPath}\"";
             List<(string, string)> environment = null;
 
             if (Application.platform != RuntimePlatform.WindowsEditor && Application.platform != RuntimePlatform.WindowsPlayer){
                 // use APE binary directly if not on Windows
-                arguments = $"{binary} {arguments}";
+                arguments = $"\"{binary}\" {arguments}";
                 binary = SelectApeBinary();
                 if (numGPULayers <= 0){
                     // prevent nvcc building if not using GPU
@@ -175,7 +178,7 @@ private void StartLLMServer()
             Debug.Log($"Server command: {binary} {arguments}");
             process = LLMUnitySetup.CreateProcess(binary, arguments, CheckIfListening, DebugLogError, environment);
             // wait for at most 2'
-            serverStarted.WaitOne(120000);
+            serverStarted.WaitOne(60000);
         }
 
         public void StopProcess()

README.md

@@ -13,8 +13,9 @@ public class ClientAdvancedAttribute : PropertyAttribute {}
     public class ServerAdvancedAttribute : PropertyAttribute {}
     public class ModelAdvancedAttribute : PropertyAttribute {}
 
+    [DefaultExecutionOrder(-1)]
     public class LLMClient : MonoBehaviour
-    {   
+    {
         [HideInInspector] public bool advancedOptions = false;
 
         [ClientAdvanced] public string host = "localhost";

Runtime/LLM.cs

@@ -110,9 +110,10 @@ public static async Task DownloadFile(
 
                 if (executable && Application.platform != RuntimePlatform.WindowsEditor && Application.platform != RuntimePlatform.WindowsPlayer){
                     // macOS/Linux: Set executable permissions using chmod
-                    RunProcess("chmod", "+x " + savePath);
+                    RunProcess("chmod", $"+x \"{savePath}\"");
                 }
                 AssetDatabase.StopAssetEditing();
+                Debug.Log($"Download complete!");
             }
             progresscallback(1f);
             callback?.Invoke(savePath);

Runtime/LLMClient.cs

@@ -78,12 +78,15 @@ void SetBubblePosition(RectTransform bubbleRectTransform, RectTransform imageRec
             bubbleRectTransform.pivot = new Vector2(bubbleUI.leftPosition, bubbleUI.bottomPosition);
             bubbleRectTransform.anchorMin = new Vector2(bubbleUI.leftPosition, bubbleUI.bottomPosition);
             bubbleRectTransform.anchorMax = new Vector2(bubbleUI.leftPosition, bubbleUI.bottomPosition);
+            bubbleRectTransform.localScale = Vector3.one;
             Vector2 anchoredPosition = new Vector2(bubbleUI.bubbleOffset + bubbleUI.textPadding, bubbleUI.bubbleOffset + bubbleUI.textPadding);
             if (bubbleUI.leftPosition == 1) anchoredPosition.x *= -1;
             if (bubbleUI.bottomPosition == 1) anchoredPosition.y *= -1;
             bubbleRectTransform.anchoredPosition = anchoredPosition;
 
-            bubbleRectTransform.sizeDelta = new Vector2(600 - 2*bubbleUI.textPadding, bubbleRectTransform.sizeDelta.y - 2*bubbleUI.textPadding);
+            float width = bubbleUI.bubbleWidth == -1? bubbleRectTransform.sizeDelta.x: bubbleUI.bubbleWidth;
+            float height = bubbleUI.bubbleHeight == -1? bubbleRectTransform.sizeDelta.y: bubbleUI.bubbleHeight;
+            bubbleRectTransform.sizeDelta = new Vector2(width-2*bubbleUI.textPadding, height-2*bubbleUI.textPadding);
             SyncParentRectTransform(imageRectTransform);
             imageRectTransform.offsetMin = new Vector2(-bubbleUI.textPadding, -bubbleUI.textPadding);
             imageRectTransform.offsetMax = new Vector2(bubbleUI.textPadding, bubbleUI.textPadding);
@@ -154,6 +157,7 @@ GameObject CreatePlaceholderObject(Transform parent, RectTransform textRectTrans
             RectTransform placeholderRectTransform = placeholderObject.GetComponent<RectTransform>();
             placeholderRectTransform.sizeDelta = textRectTransform.sizeDelta;
             placeholderRectTransform.anchoredPosition = textRectTransform.anchoredPosition;
+            placeholderRectTransform.localScale = Vector3.one;
             SyncParentRectTransform(placeholderRectTransform);
             return placeholderObject;
         }
@@ -168,7 +172,9 @@ GameObject CreateInputFieldObject(Transform parent, Text textObject, Text placeh
             inputField.lineType = InputField.LineType.MultiLineSubmit;
             inputField.shouldHideMobileInput = false;
             inputField.shouldActivateOnSelect = true;
-            SyncParentRectTransform(inputFieldObject.GetComponent<RectTransform>());
+            RectTransform inputFieldRect = inputFieldObject.GetComponent<RectTransform>();
+            inputFieldRect.localScale = Vector3.one;
+            SyncParentRectTransform(inputFieldRect);
             return inputFieldObject;
         }
 

Runtime/LLMUnitySetup.cs

@@ -2,7 +2,6 @@
 using System.Collections.Generic;
 using System.Threading.Tasks;
 using LLMUnity;
-using TMPro;
 
 namespace LLMUnitySamples
 {
@@ -117,7 +116,7 @@ public void UpdateBubblePositions()
             for (int i = chatBubbles.Count - 1; i >= 0; i--) {
                 Bubble bubble = chatBubbles[i];
                 RectTransform childRect = bubble.GetRectTransform();
-                childRect.position = new Vector2(childRect.position.x, y);
+                childRect.anchoredPosition = new Vector2(childRect.anchoredPosition.x, y);
 
                 // last bubble outside the container
                 if (y > containerHeight && lastBubbleOutsideFOV == -1){

Samples~/ChatBot/Bubble.cs

@@ -444,6 +444,7 @@ MonoBehaviour:
   m_Script: {fileID: 11500000, guid: a50e3140c3ecaaf1c848dbf141cc2074, type: 3}
   m_Name: 
   m_EditorClassIdentifier: 
+  advancedOptions: 0
   host: localhost
   port: 13333
   stream: 1
@@ -459,6 +460,7 @@ MonoBehaviour:
   modelHide: 1
   numThreads: -1
   numGPULayers: 0
+  parallelPrompts: -1
   debug: 0
   model: 
   lora: 
@@ -834,10 +836,10 @@ MonoBehaviour:
   m_Script: {fileID: 11500000, guid: 0cd44c1031e13a943bb63640046fad76, type: 3}
   m_Name: 
   m_EditorClassIdentifier: 
-  m_UiScaleMode: 0
+  m_UiScaleMode: 1
   m_ReferencePixelsPerUnit: 100
   m_ScaleFactor: 1
-  m_ReferenceResolution: {x: 1550, y: 600}
+  m_ReferenceResolution: {x: 1280, y: 720}
   m_ScreenMatchMode: 0
   m_MatchWidthOrHeight: 1
   m_PhysicalUnit: 3