altic-dev · mvanhorn · Jun 28, 2026
diff --git a/Fluid.xcodeproj/project.pbxproj b/Fluid.xcodeproj/project.pbxproj
@@ -15,6 +15,7 @@
 		7CDB0A2D2F3C4D5600FB7CAD /* DictationE2ETests.swift in Sources */ = {isa = PBXBuildFile; fileRef = 7CDB0A292F3C4D5600FB7CAD /* DictationE2ETests.swift */; };
 		7CDB0A2E2F3C4D5600FB7CAD /* AudioFixtureLoader.swift in Sources */ = {isa = PBXBuildFile; fileRef = 7CDB0A2A2F3C4D5600FB7CAD /* AudioFixtureLoader.swift */; };
 		86CAA2D4EF18433096185602 /* LLMClientRequestBodyTests.swift in Sources */ = {isa = PBXBuildFile; fileRef = 343B29013F4441D6A797D12D /* LLMClientRequestBodyTests.swift */; };
+		7CAA000A2F99000000000001 /* NemotronCacheRecoveryTests.swift in Sources */ = {isa = PBXBuildFile; fileRef = 7CAA000B2F99000000000002 /* NemotronCacheRecoveryTests.swift */; };
 		7CDB0A2F2F3C4D5600FB7CAD /* dictation_fixture.wav in Resources */ = {isa = PBXBuildFile; fileRef = 7CDB0A2B2F3C4D5600FB7CAD /* dictation_fixture.wav */; };
 		7CDB0A302F3C4D5600FB7CAD /* XCTest.framework in Frameworks */ = {isa = PBXBuildFile; fileRef = 7CDB0A2C2F3C4D5600FB7CAD /* XCTest.framework */; };
 		7CE006BD2E80EBE600DDCCD6 /* AppUpdater in Frameworks */ = {isa = PBXBuildFile; productRef = 7CE006BC2E80EBE600DDCCD6 /* AppUpdater */; };
@@ -36,6 +37,7 @@
 		7C91B0022F42AA0100C0DEF0 /* HotkeyShortcutTests.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = HotkeyShortcutTests.swift; sourceTree = "<group>"; };
 		7CDB0A292F3C4D5600FB7CAD /* DictationE2ETests.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = DictationE2ETests.swift; sourceTree = "<group>"; };
 		343B29013F4441D6A797D12D /* LLMClientRequestBodyTests.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = LLMClientRequestBodyTests.swift; sourceTree = "<group>"; };
+		7CAA000B2F99000000000002 /* NemotronCacheRecoveryTests.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = NemotronCacheRecoveryTests.swift; sourceTree = "<group>"; };
 		7CDB0A2A2F3C4D5600FB7CAD /* AudioFixtureLoader.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = AudioFixtureLoader.swift; sourceTree = "<group>"; };
 		7CDB0A2B2F3C4D5600FB7CAD /* dictation_fixture.wav */ = {isa = PBXFileReference; lastKnownFileType = audio.wav; path = dictation_fixture.wav; sourceTree = "<group>"; };
 		7CDB0A2C2F3C4D5600FB7CAD /* XCTest.framework */ = {isa = PBXFileReference; lastKnownFileType = wrapper.framework; name = XCTest.framework; path = Platforms/MacOSX.platform/Developer/Library/Frameworks/XCTest.framework; sourceTree = DEVELOPER_DIR; };
@@ -107,6 +109,7 @@
 				7CDB0A292F3C4D5600FB7CAD /* DictationE2ETests.swift */,
 				7C91B0022F42AA0100C0DEF0 /* HotkeyShortcutTests.swift */,
 				343B29013F4441D6A797D12D /* LLMClientRequestBodyTests.swift */,
+				7CAA000B2F99000000000002 /* NemotronCacheRecoveryTests.swift */,
 			);
 			path = FluidDictationIntegrationTests;
 			sourceTree = "<group>";
@@ -262,6 +265,7 @@
 				7CDB0A2D2F3C4D5600FB7CAD /* DictationE2ETests.swift in Sources */,
 				7C91B0012F42AA0100C0DEF0 /* HotkeyShortcutTests.swift in Sources */,
 				86CAA2D4EF18433096185602 /* LLMClientRequestBodyTests.swift in Sources */,
+				7CAA000A2F99000000000001 /* NemotronCacheRecoveryTests.swift in Sources */,
 			);
 			runOnlyForDeploymentPostprocessing = 0;
 		};

diff --git a/Sources/Fluid/Services/NemotronProvider.swift b/Sources/Fluid/Services/NemotronProvider.swift
@@ -74,6 +74,12 @@ final class NemotronProvider: TranscriptionProvider {
             .appendingPathComponent(self.folderHint, isDirectory: true)
     }
 
+    static var compiledModelCacheDirectory: URL? {
+        FileManager.default.urls(for: .cachesDirectory, in: .userDomainMask).first?
+            .appendingPathComponent("FluidAudio", isDirectory: true)
+            .appendingPathComponent("CompiledNemotronModels", isDirectory: true)
+    }
+
     func modelsExistOnDisk() -> Bool {
         guard let dir = self.cacheDirectory else { return false }
         return Self.artifactsAreComplete(at: dir)
@@ -174,7 +180,16 @@ final class NemotronProvider: TranscriptionProvider {
                 "Nemotron: ANE model load failed; retrying with cpuAndGPU fallback [error=\(error.localizedDescription)]",
                 source: "Nemotron"
             )
-            manager = try await self.loadManager(modelDirectory: dir, computeUnits: .cpuAndGPU)
+            self.clearCompiledModelCache()
+            do {
+                manager = try await self.loadManager(modelDirectory: dir, computeUnits: .cpuAndGPU)
+            } catch {
+                let compiledCachePath = Self.compiledModelCacheDirectory?.path ?? "~/Library/Caches/FluidAudio/CompiledNemotronModels"
+                throw Self.makeError(
+                    "Nemotron model failed to load even after clearing the compiled CoreML cache at \(compiledCachePath). If this continues, manually delete ~/Library/Caches/FluidAudio/CompiledNemotronModels and try again.",
+                    underlyingError: error
+                )
+            }
         }
         try await self.applySelectedLanguage(to: manager)
         try Task.checkCancellation()
@@ -337,6 +352,26 @@ final class NemotronProvider: TranscriptionProvider {
         self.activeLanguageCode = nil
     }
 
+    private func clearCompiledModelCache() {
+        Self.clearCompiledModelCache(at: Self.compiledModelCacheDirectory)
+    }
+
+    static func clearCompiledModelCache(at directory: URL?) {
+        guard let directory, FileManager.default.fileExists(atPath: directory.path) else { return }
+        DebugLogger.shared.warning(
+            "Nemotron: clearing compiled CoreML cache at \(directory.path)",
+            source: "Nemotron"
+        )
+        do {
+            try FileManager.default.removeItem(at: directory)
+        } catch {
+            DebugLogger.shared.warning(
+                "Nemotron: failed to clear compiled CoreML cache at \(directory.path): \(error.localizedDescription)",
+                source: "Nemotron"
+            )
+        }
+    }
+
     private func transcribeBatched(_ samples: [Float]) async throws -> ASRTranscriptionResult {
         guard samples.isEmpty == false else { return ASRTranscriptionResult(text: "", confidence: 0) }
 
@@ -597,7 +632,7 @@ final class NemotronProvider: TranscriptionProvider {
         return nil
     }
 
-    private static func shouldRetryWithoutNeuralEngine(_ error: Error) -> Bool {
+    static func shouldRetryWithoutNeuralEngine(_ error: Error) -> Bool {
         if self.hasNeuralEngineRetryCode(error as NSError) {
             return true
         }
@@ -636,8 +671,12 @@ final class NemotronProvider: TranscriptionProvider {
         }
     }
 
-    private static func makeError(_ description: String) -> NSError {
-        NSError(domain: "NemotronProvider", code: -1, userInfo: [NSLocalizedDescriptionKey: description])
+    private static func makeError(_ description: String, underlyingError: Error? = nil) -> NSError {
+        var userInfo: [String: Any] = [NSLocalizedDescriptionKey: description]
+        if let underlyingError {
+            userInfo[NSUnderlyingErrorKey] = underlyingError
+        }
+        return NSError(domain: "NemotronProvider", code: -1, userInfo: userInfo)
     }
 }
 #else

diff --git a/Tests/FluidDictationIntegrationTests/NemotronCacheRecoveryTests.swift b/Tests/FluidDictationIntegrationTests/NemotronCacheRecoveryTests.swift
@@ -0,0 +1,76 @@
+@testable import FluidVoice_Debug
+import XCTest
+
+#if arch(arm64)
+@MainActor
+@available(macOS 14.0, *)
+final class NemotronCacheRecoveryTests: XCTestCase {
+
+    func testShouldRetryWithoutNeuralEngine_matchesNestedErrorCodeMinus14() {
+        let underlying = NSError(domain: "CoreML", code: -14)
+        let error = NSError(
+            domain: "FluidVoiceTests",
+            code: 1,
+            userInfo: [NSUnderlyingErrorKey: underlying]
+        )
+
+        XCTAssertTrue(NemotronProvider.shouldRetryWithoutNeuralEngine(error))
+    }
+
+    func testShouldRetryWithoutNeuralEngine_matchesCoreMLExecutionPlanDescription() {
+        let error = NSError(
+            domain: "CoreML",
+            code: 1,
+            userInfo: [
+                NSLocalizedDescriptionKey: "Failed to build the model execution plan with error code: -14.",
+            ]
+        )
+
+        XCTAssertTrue(NemotronProvider.shouldRetryWithoutNeuralEngine(error))
+    }
+
+    func testShouldRetryWithoutNeuralEngine_ignoresUnrelatedErrors() {
+        let error = NSError(
+            domain: NSCocoaErrorDomain,
+            code: -999,
+            userInfo: [NSLocalizedDescriptionKey: "The operation was cancelled."]
+        )
+
+        XCTAssertFalse(NemotronProvider.shouldRetryWithoutNeuralEngine(error))
+    }
+
+    func testCompiledModelCacheDirectoryUsesFluidAudioCompiledNemotronCache() throws {
+        let cachesDirectory = try XCTUnwrap(
+            FileManager.default.urls(for: .cachesDirectory, in: .userDomainMask).first
+        ).standardizedFileURL
+        let compiledCacheDirectory = try XCTUnwrap(NemotronProvider.compiledModelCacheDirectory)
+            .standardizedFileURL
+
+        XCTAssertTrue(compiledCacheDirectory.path.hasPrefix(cachesDirectory.path + "/"))
+        XCTAssertEqual(
+            Array(compiledCacheDirectory.pathComponents.suffix(2)),
+            ["FluidAudio", "CompiledNemotronModels"]
+        )
+    }
+
+    func testClearCompiledModelCacheRemovesDirectoryAndIgnoresMissingDirectory() throws {
+        let tempRoot = FileManager.default.temporaryDirectory
+            .appendingPathComponent("NemotronCacheRecoveryTests-\(UUID().uuidString)", isDirectory: true)
+        let compiledCacheDirectory = tempRoot
+            .appendingPathComponent("FluidAudio", isDirectory: true)
+            .appendingPathComponent("CompiledNemotronModels", isDirectory: true)
+        defer {
+            try? FileManager.default.removeItem(at: tempRoot)
+        }
+
+        try FileManager.default.createDirectory(at: compiledCacheDirectory, withIntermediateDirectories: true)
+        try Data("cache".utf8).write(to: compiledCacheDirectory.appendingPathComponent("marker.txt"))
+
+        NemotronProvider.clearCompiledModelCache(at: compiledCacheDirectory)
+
+        XCTAssertFalse(FileManager.default.fileExists(atPath: compiledCacheDirectory.path))
+        NemotronProvider.clearCompiledModelCache(at: compiledCacheDirectory)
+        XCTAssertFalse(FileManager.default.fileExists(atPath: compiledCacheDirectory.path))
+    }
+}
+#endif