LLava Async Loading

LLama.Examples/Examples/LlavaInteractiveModeExecute.cs

@@ -24,9 +24,9 @@ public static async Task Run()
             using var context = model.CreateContext(parameters);
 
             // Llava Init
-            using var clipModel = LLavaWeights.LoadFromFile(multiModalProj);
+            using var clipModel = await LLavaWeights.LoadFromFileAsync(multiModalProj);
 
-            var ex = new InteractiveExecutor(context, clipModel );
+            var ex = new InteractiveExecutor(context, clipModel);
 
             Console.ForegroundColor = ConsoleColor.Yellow;
             Console.WriteLine("The executor has been enabled. In this example, the prompt is printed, the maximum tokens is set to {0} and the context size is {1}.", maxTokens, parameters.ContextSize );

LLama/LLavaWeights.cs

@@ -1,5 +1,7 @@
 
 using System;
+using System.Threading;
+using System.Threading.Tasks;
 using LLama.Native;
 
 namespace LLama;
@@ -13,9 +15,9 @@ public sealed class LLavaWeights : IDisposable
     /// The native handle, which is used in the native APIs
     /// </summary>
     /// <remarks>Be careful how you use this!</remarks>
-    public SafeLlavaModelHandle NativeHandle { get; }   
-    
-    internal LLavaWeights(SafeLlavaModelHandle weights)
+    public SafeLlavaModelHandle NativeHandle { get; }
+
+    private LLavaWeights(SafeLlavaModelHandle weights)
     {
         NativeHandle = weights;
     }
@@ -31,6 +33,17 @@ public static LLavaWeights LoadFromFile(string mmProject)
         return new LLavaWeights(weights);
     }
 
+    /// <summary>
+    /// Load weights into memory
+    /// </summary>
+    /// <param name="mmProject">path to the "mmproj" model file</param>
+    /// <param name="token"></param>
+    /// <returns></returns>
+    public static Task<LLavaWeights> LoadFromFileAsync(string mmProject, CancellationToken token = default)
+    {
+        return Task.Run(() => LoadFromFile(mmProject), token);
+    }
+
     /// <summary>
     /// Create the Image Embeddings from the bytes of an image.
     /// </summary>

LLama/Native/SafeLlavaModelHandle.cs

@@ -39,8 +39,11 @@ public static SafeLlavaModelHandle LoadFromFile(string modelPath, int verbosity
                 if (!fs.CanRead)
                     throw new InvalidOperationException($"Llava MMP Model file '{modelPath}' is not readable");
 
-            return clip_model_load(modelPath, verbosity)
-                ?? throw new LoadWeightsFailedException(modelPath);
+            var handle = clip_model_load(modelPath, verbosity);
+            if (handle.IsInvalid)
+                throw new LoadWeightsFailedException(modelPath);
+
+            return handle;
         }
 
         /// <summary>