From 5f5b983dc6a5079dc9b9d391511731f3dd99353c Mon Sep 17 00:00:00 2001
From: Tomislav Eric <tomislav.eric@arconsis.com>
Date: Thu, 15 Jan 2026 18:20:16 +0100
Subject: [PATCH] update to llama.cpp and fix deprecations

---
 .../xcschemes/LLamaSwift.xcscheme             | 79 +++++++++++++++++++
 Package.resolved                              | 13 +--
 Package.swift                                 | 17 ++--
 Sources/llama-cpp-swift/LLama.swift           |  9 ++-
 Sources/llama-cpp-swift/Model.swift           |  8 +-
 5 files changed, 100 insertions(+), 26 deletions(-)
 create mode 100644 .swiftpm/xcode/xcshareddata/xcschemes/LLamaSwift.xcscheme

diff --git a/.swiftpm/xcode/xcshareddata/xcschemes/LLamaSwift.xcscheme b/.swiftpm/xcode/xcshareddata/xcschemes/LLamaSwift.xcscheme
new file mode 100644
index 0000000..52fa605
--- /dev/null
+++ b/.swiftpm/xcode/xcshareddata/xcschemes/LLamaSwift.xcscheme
@@ -0,0 +1,79 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<Scheme
+   LastUpgradeVersion = "2610"
+   version = "1.7">
+   <BuildAction
+      parallelizeBuildables = "YES"
+      buildImplicitDependencies = "YES"
+      buildArchitectures = "Automatic">
+      <BuildActionEntries>
+         <BuildActionEntry
+            buildForTesting = "YES"
+            buildForRunning = "YES"
+            buildForProfiling = "YES"
+            buildForArchiving = "YES"
+            buildForAnalyzing = "YES">
+            <BuildableReference
+               BuildableIdentifier = "primary"
+               BlueprintIdentifier = "LLamaSwift"
+               BuildableName = "LLamaSwift"
+               BlueprintName = "LLamaSwift"
+               ReferencedContainer = "container:">
+            </BuildableReference>
+         </BuildActionEntry>
+      </BuildActionEntries>
+   </BuildAction>
+   <TestAction
+      buildConfiguration = "Debug"
+      selectedDebuggerIdentifier = "Xcode.DebuggerFoundation.Debugger.LLDB"
+      selectedLauncherIdentifier = "Xcode.DebuggerFoundation.Launcher.LLDB"
+      shouldUseLaunchSchemeArgsEnv = "YES"
+      shouldAutocreateTestPlan = "YES">
+      <Testables>
+         <TestableReference
+            skipped = "NO">
+            <BuildableReference
+               BuildableIdentifier = "primary"
+               BlueprintIdentifier = "llama-cpp-swiftTests"
+               BuildableName = "llama-cpp-swiftTests"
+               BlueprintName = "llama-cpp-swiftTests"
+               ReferencedContainer = "container:">
+            </BuildableReference>
+         </TestableReference>
+      </Testables>
+   </TestAction>
+   <LaunchAction
+      buildConfiguration = "Debug"
+      selectedDebuggerIdentifier = "Xcode.DebuggerFoundation.Debugger.LLDB"
+      selectedLauncherIdentifier = "Xcode.DebuggerFoundation.Launcher.LLDB"
+      launchStyle = "0"
+      useCustomWorkingDirectory = "NO"
+      ignoresPersistentStateOnLaunch = "NO"
+      debugDocumentVersioning = "YES"
+      debugServiceExtension = "internal"
+      allowLocationSimulation = "YES">
+   </LaunchAction>
+   <ProfileAction
+      buildConfiguration = "Release"
+      shouldUseLaunchSchemeArgsEnv = "YES"
+      savedToolIdentifier = ""
+      useCustomWorkingDirectory = "NO"
+      debugDocumentVersioning = "YES">
+      <MacroExpansion>
+         <BuildableReference
+            BuildableIdentifier = "primary"
+            BlueprintIdentifier = "LLamaSwift"
+            BuildableName = "LLamaSwift"
+            BlueprintName = "LLamaSwift"
+            ReferencedContainer = "container:">
+         </BuildableReference>
+      </MacroExpansion>
+   </ProfileAction>
+   <AnalyzeAction
+      buildConfiguration = "Debug">
+   </AnalyzeAction>
+   <ArchiveAction
+      buildConfiguration = "Release"
+      revealArchiveInOrganizer = "YES">
+   </ArchiveAction>
+</Scheme>
diff --git a/Package.resolved b/Package.resolved
index b783b41..2361209 100644
--- a/Package.resolved
+++ b/Package.resolved
@@ -1,21 +1,12 @@
 {
   "pins" : [
-    {
-      "identity" : "llama.cpp",
-      "kind" : "remoteSourceControl",
-      "location" : "https://github.com/ggerganov/llama.cpp",
-      "state" : {
-        "branch" : "master",
-        "revision" : "26a8406ba9198eb6fdd8329fa717555b4f77f05f"
-      }
-    },
     {
       "identity" : "swift-log",
       "kind" : "remoteSourceControl",
       "location" : "https://github.com/apple/swift-log.git",
       "state" : {
-        "revision" : "96a2f8a0fa41e9e09af4585e2724c4e825410b91",
-        "version" : "1.6.2"
+        "revision" : "7ee16e465622412764b0ff0c1301801dc71b8f61",
+        "version" : "1.9.0"
       }
     }
   ],
diff --git a/Package.swift b/Package.swift
index b403c5b..e5769b2 100644
--- a/Package.swift
+++ b/Package.swift
@@ -11,21 +11,24 @@ let package = Package(
         .visionOS(.v1)
     ],
     products: [
-        .library(
-            name: "LLamaSwift",
-            targets: ["LLamaSwift"]),
+        .library(name: "LLamaSwift", targets: ["LLamaSwift"]),
     ],
     dependencies: [
-        .package(url: "https://github.com/ggerganov/llama.cpp", branch: "master"),
         .package(url: "https://github.com/apple/swift-log.git", from: "1.6.1"),
     ],
     targets: [
+        .binaryTarget(
+            name: "llama",
+            url: "https://github.com/ggml-org/llama.cpp/releases/download/b7642/llama-b7642-xcframework.zip",
+            checksum: "59b372b1991470cf6e3b48b49052495ddf5e1f39551b595dabb011a24b08b4fc"
+        ),
         .target(
-            name: "LLamaSwift",        
+            name: "LLamaSwift",
             dependencies: [
-                .product(name: "llama", package: "llama.cpp"),
+                "llama",
                 .product(name: "Logging", package: "swift-log"),
-            ]
+            ],
+            path: "Sources/llama-cpp-swift"
         ),
         .testTarget(
             name: "llama-cpp-swiftTests",
diff --git a/Sources/llama-cpp-swift/LLama.swift b/Sources/llama-cpp-swift/LLama.swift
index 3536514..5a2fdd4 100644
--- a/Sources/llama-cpp-swift/LLama.swift
+++ b/Sources/llama-cpp-swift/LLama.swift
@@ -19,9 +19,10 @@ public actor LLama {
     // Initialize sampling
     let sparams = llama_sampler_chain_default_params()
     self.sampling = llama_sampler_chain_init(sparams)
-    llama_sampler_chain_add(self.sampling, llama_sampler_init_temp(0.8))
-    llama_sampler_chain_add(self.sampling, llama_sampler_init_softmax())
-    llama_sampler_chain_add(self.sampling, llama_sampler_init_dist(1234))
+      llama_sampler_chain_add(self.sampling, llama_sampler_init_top_k(40));
+      llama_sampler_chain_add(self.sampling, llama_sampler_init_top_p(0.9, 1));
+      llama_sampler_chain_add(self.sampling, llama_sampler_init_temp (0.4));
+      llama_sampler_chain_add(self.sampling, llama_sampler_init_dist (1234));
   }
 
   deinit {
@@ -197,7 +198,7 @@ public actor LLama {
     var newTokenID: llama_token = 0
     newTokenID = llama_sampler_sample(sampling, model.context, batch.n_tokens - 1)
 
-    if llama_token_is_eog(model.model, newTokenID) || nCur == nLen {
+    if llama_vocab_is_eog(model.model, newTokenID) || nCur == nLen {
       isDone = true
       let newTokenStr = String(
         decoding: Data(temporaryInvalidCChars.map { UInt8(bitPattern: $0) }), as: UTF8.self)
diff --git a/Sources/llama-cpp-swift/Model.swift b/Sources/llama-cpp-swift/Model.swift
index d2bc1a0..3c706db 100644
--- a/Sources/llama-cpp-swift/Model.swift
+++ b/Sources/llama-cpp-swift/Model.swift
@@ -16,7 +16,7 @@ public final class Model {
       logger.debug("Running on simulator, force use n_gpu_layers = 0")
     #endif
 
-    guard let model = llama_load_model_from_file(modelPath, modelParams) else {
+    guard let model = llama_model_load_from_file(modelPath, modelParams) else {
       llama_backend_free()
       throw InitializationError(message: "Failed to load model", code: .failedToLoadModel)
     }
@@ -31,8 +31,8 @@ public final class Model {
     ctxParams.n_threads = Int32(nThreads)
     ctxParams.n_threads_batch = Int32(nThreads)
 
-    guard let context = llama_new_context_with_model(model, ctxParams) else {
-      llama_free_model(model)
+    guard let context = llama_init_from_model(model, ctxParams) else {
+      llama_model_free(model)
       llama_backend_free()
       throw InitializationError(
         message: "Failed to initialize context", code: .failedToInitializeContext)
@@ -42,7 +42,7 @@ public final class Model {
 
   deinit {
     llama_free(context)
-    llama_free_model(model)
+    llama_model_free(model)
     llama_backend_free()
   }
 }