@dvai-bridge/ios 4.0.0 → 4.0.2

package/ios/Sources/DVAIBridge/Offload/OffloadConfig.swift

@@ -1,118 +1,118 @@
-import Foundation
-
-/// Per-instance offload configuration. Mirrors the TS-side
-/// `OffloadConfig` in `packages/dvai-bridge-core/src/offload/types.ts`.
-///
-/// Default state: `enabled = false` — offload is opt-in at v3.0.
-///
-/// Set on `StartOptions(offload:)` when calling
-/// `DVAIBridge.shared.start(_:)`.
-public struct OffloadConfig: Sendable {
-    /// Master switch. Default `false`; offload is opt-in at v3.0.
-    public var enabled: Bool
-    /// Run mDNS to discover LAN peers. Default `true` when enabled.
-    public var discoverLAN: Bool
-    /// Below this tok/s, look for a peer. Default 10.
-    public var minLocalCapability: Double
-    /// Optional rendezvous-server URL — enables internet path if set.
-    public var rendezvousUrl: URL?
-    /// Optional pre-known peers (skip discovery for these).
-    public var knownPeers: [MDNSPeer]
-    /// Pairing TTL in days. Default 30 — matches the JS-side default
-    /// in `PairingPolicy`.
-    public var expireAfterDays: Int
-
-    public init(
-        enabled: Bool = false,
-        discoverLAN: Bool = true,
-        minLocalCapability: Double = 10,
-        rendezvousUrl: URL? = nil,
-        knownPeers: [MDNSPeer] = [],
-        expireAfterDays: Int = 30
-    ) {
-        self.enabled = enabled
-        self.discoverLAN = discoverLAN
-        self.minLocalCapability = minLocalCapability
-        self.rendezvousUrl = rendezvousUrl
-        self.knownPeers = knownPeers
-        self.expireAfterDays = expireAfterDays
-    }
-}
-
-/// Modern start-options surface that wraps `DVAIBridgeConfig` and adds
-/// the `offload` knob. Both call sites are supported on
-/// `DVAIBridge.shared.start(...)`:
-///
-///     try await DVAIBridge.shared.start(.init(backend: .auto, modelPath: "/x"))
-///     try await DVAIBridge.shared.start(StartOptions(
-///         backend: .auto,
-///         modelPath: "/x",
-///         offload: OffloadConfig(enabled: true, discoverLAN: true)
-///     ))
-///
-/// Internally `StartOptions` decomposes into `(DVAIBridgeConfig, OffloadConfig?)`
-/// so all existing tests + call sites that take a config keep working.
-public struct StartOptions: Sendable {
-    public var config: DVAIBridgeConfig
-    public var offload: OffloadConfig?
-    /// v3.2.2 — explicit path to the license `.jwt` file. Overrides the
-    /// auto-discovery walk (Bundle resource, Documents, etc). The same
-    /// .jwt format works across iOS / Android / .NET / RN / JS SDKs.
-    public var licenseKeyPath: String?
-    /// v3.2.2 — inline JWT license. Useful when the host app fetches
-    /// the license over the network and wants to inject the result
-    /// without touching disk. Wins over `licenseKeyPath` if both are set.
-    public var licenseToken: String?
-
-    public init(
-        config: DVAIBridgeConfig,
-        offload: OffloadConfig? = nil,
-        licenseKeyPath: String? = nil,
-        licenseToken: String? = nil
-    ) {
-        self.config = config
-        self.offload = offload
-        self.licenseKeyPath = licenseKeyPath
-        self.licenseToken = licenseToken
-    }
-
-    /// Convenience initializer that mirrors the documented public
-    /// surface in `docs/migration/v2.4-to-v3.0.md`.
-    public init(
-        backend: BackendKind = .auto,
-        modelPath: String? = nil,
-        mmprojPath: String? = nil,
-        tokenizerPath: String? = nil,
-        gpuLayers: Int = 99,
-        contextSize: Int = 2048,
-        threads: Int = 4,
-        embeddingMode: Bool = false,
-        httpBasePort: Int = 38883,
-        httpMaxPortAttempts: Int = 16,
-        corsOrigin: DVAIBridgeConfig.CORSOrigin = .wildcard,
-        autoUnloadOnLowMemory: Bool = false,
-        logLevel: String = "info",
-        offload: OffloadConfig? = nil,
-        licenseKeyPath: String? = nil,
-        licenseToken: String? = nil
-    ) {
-        self.config = DVAIBridgeConfig(
-            backend: backend,
-            modelPath: modelPath,
-            mmprojPath: mmprojPath,
-            tokenizerPath: tokenizerPath,
-            gpuLayers: gpuLayers,
-            contextSize: contextSize,
-            threads: threads,
-            embeddingMode: embeddingMode,
-            httpBasePort: httpBasePort,
-            httpMaxPortAttempts: httpMaxPortAttempts,
-            corsOrigin: corsOrigin,
-            autoUnloadOnLowMemory: autoUnloadOnLowMemory,
-            logLevel: logLevel
-        )
-        self.offload = offload
-        self.licenseKeyPath = licenseKeyPath
-        self.licenseToken = licenseToken
-    }
-}
+import Foundation
+
+/// Per-instance offload configuration. Mirrors the TS-side
+/// `OffloadConfig` in `packages/dvai-bridge-core/src/offload/types.ts`.
+///
+/// Default state: `enabled = false` — offload is opt-in at v3.0.
+///
+/// Set on `StartOptions(offload:)` when calling
+/// `DVAIBridge.shared.start(_:)`.
+public struct OffloadConfig: Sendable {
+    /// Master switch. Default `false`; offload is opt-in at v3.0.
+    public var enabled: Bool
+    /// Run mDNS to discover LAN peers. Default `true` when enabled.
+    public var discoverLAN: Bool
+    /// Below this tok/s, look for a peer. Default 10.
+    public var minLocalCapability: Double
+    /// Optional rendezvous-server URL — enables internet path if set.
+    public var rendezvousUrl: URL?
+    /// Optional pre-known peers (skip discovery for these).
+    public var knownPeers: [MDNSPeer]
+    /// Pairing TTL in days. Default 30 — matches the JS-side default
+    /// in `PairingPolicy`.
+    public var expireAfterDays: Int
+
+    public init(
+        enabled: Bool = false,
+        discoverLAN: Bool = true,
+        minLocalCapability: Double = 10,
+        rendezvousUrl: URL? = nil,
+        knownPeers: [MDNSPeer] = [],
+        expireAfterDays: Int = 30
+    ) {
+        self.enabled = enabled
+        self.discoverLAN = discoverLAN
+        self.minLocalCapability = minLocalCapability
+        self.rendezvousUrl = rendezvousUrl
+        self.knownPeers = knownPeers
+        self.expireAfterDays = expireAfterDays
+    }
+}
+
+/// Modern start-options surface that wraps `DVAIBridgeConfig` and adds
+/// the `offload` knob. Both call sites are supported on
+/// `DVAIBridge.shared.start(...)`:
+///
+///     try await DVAIBridge.shared.start(.init(backend: .auto, modelPath: "/x"))
+///     try await DVAIBridge.shared.start(StartOptions(
+///         backend: .auto,
+///         modelPath: "/x",
+///         offload: OffloadConfig(enabled: true, discoverLAN: true)
+///     ))
+///
+/// Internally `StartOptions` decomposes into `(DVAIBridgeConfig, OffloadConfig?)`
+/// so all existing tests + call sites that take a config keep working.
+public struct StartOptions: Sendable {
+    public var config: DVAIBridgeConfig
+    public var offload: OffloadConfig?
+    /// v3.2.2 — explicit path to the license `.jwt` file. Overrides the
+    /// auto-discovery walk (Bundle resource, Documents, etc). The same
+    /// .jwt format works across iOS / Android / .NET / RN / JS SDKs.
+    public var licenseKeyPath: String?
+    /// v3.2.2 — inline JWT license. Useful when the host app fetches
+    /// the license over the network and wants to inject the result
+    /// without touching disk. Wins over `licenseKeyPath` if both are set.
+    public var licenseToken: String?
+
+    public init(
+        config: DVAIBridgeConfig,
+        offload: OffloadConfig? = nil,
+        licenseKeyPath: String? = nil,
+        licenseToken: String? = nil
+    ) {
+        self.config = config
+        self.offload = offload
+        self.licenseKeyPath = licenseKeyPath
+        self.licenseToken = licenseToken
+    }
+
+    /// Convenience initializer that mirrors the documented public
+    /// surface in `docs/migration/v2.4-to-v3.0.md`.
+    public init(
+        backend: BackendKind = .auto,
+        modelPath: String? = nil,
+        mmprojPath: String? = nil,
+        tokenizerPath: String? = nil,
+        gpuLayers: Int = 99,
+        contextSize: Int = 2048,
+        threads: Int = 4,
+        embeddingMode: Bool = false,
+        httpBasePort: Int = 38883,
+        httpMaxPortAttempts: Int = 16,
+        corsOrigin: DVAIBridgeConfig.CORSOrigin = .wildcard,
+        autoUnloadOnLowMemory: Bool = false,
+        logLevel: String = "info",
+        offload: OffloadConfig? = nil,
+        licenseKeyPath: String? = nil,
+        licenseToken: String? = nil
+    ) {
+        self.config = DVAIBridgeConfig(
+            backend: backend,
+            modelPath: modelPath,
+            mmprojPath: mmprojPath,
+            tokenizerPath: tokenizerPath,
+            gpuLayers: gpuLayers,
+            contextSize: contextSize,
+            threads: threads,
+            embeddingMode: embeddingMode,
+            httpBasePort: httpBasePort,
+            httpMaxPortAttempts: httpMaxPortAttempts,
+            corsOrigin: corsOrigin,
+            autoUnloadOnLowMemory: autoUnloadOnLowMemory,
+            logLevel: logLevel
+        )
+        self.offload = offload
+        self.licenseKeyPath = licenseKeyPath
+        self.licenseToken = licenseToken
+    }
+}

package/ios/Sources/DVAIBridge/ProgressEvent.swift

@@ -1,34 +1,34 @@
-import Foundation
-
-/// Lifecycle progress event emitted during start(), downloadModel(), and
-/// related long-running operations. Mirrors the existing TS / Capacitor
-/// `ProgressEvent` shape so the iOS SDK reads identically to the JS API.
-public struct ProgressEvent: Sendable, Equatable, Codable {
-    public enum Phase: String, Sendable, Codable {
-        case download
-        case verify
-        case load
-        case ready
-        case error
-    }
-
-    public let phase: Phase
-    public let bytesReceived: Int64?
-    public let bytesTotal: Int64?
-    public let percent: Double?
-    public let message: String?
-
-    public init(
-        phase: Phase,
-        bytesReceived: Int64? = nil,
-        bytesTotal: Int64? = nil,
-        percent: Double? = nil,
-        message: String? = nil
-    ) {
-        self.phase = phase
-        self.bytesReceived = bytesReceived
-        self.bytesTotal = bytesTotal
-        self.percent = percent
-        self.message = message
-    }
-}
+import Foundation
+
+/// Lifecycle progress event emitted during start(), downloadModel(), and
+/// related long-running operations. Mirrors the existing TS / Capacitor
+/// `ProgressEvent` shape so the iOS SDK reads identically to the JS API.
+public struct ProgressEvent: Sendable, Equatable, Codable {
+    public enum Phase: String, Sendable, Codable {
+        case download
+        case verify
+        case load
+        case ready
+        case error
+    }
+
+    public let phase: Phase
+    public let bytesReceived: Int64?
+    public let bytesTotal: Int64?
+    public let percent: Double?
+    public let message: String?
+
+    public init(
+        phase: Phase,
+        bytesReceived: Int64? = nil,
+        bytesTotal: Int64? = nil,
+        percent: Double? = nil,
+        message: String? = nil
+    ) {
+        self.phase = phase
+        self.bytesReceived = bytesReceived
+        self.bytesTotal = bytesTotal
+        self.percent = percent
+        self.message = message
+    }
+}

package/ios/Sources/DVAICoreMLCore/CoreMLBackendError.swift

@@ -1,19 +1,19 @@
-import Foundation
-
-public enum CoreMLBackendError: Error, LocalizedError, Sendable {
-    case modelLoadFailed(reason: String)
-    case tokenizerLoadFailed(reason: String)
-    case stateInitFailed(reason: String)
-    case generationFailed(reason: String)
-    case unsupportedModelFormat(reason: String)
-
-    public var errorDescription: String? {
-        switch self {
-        case .modelLoadFailed(let r): return "CoreML model load failed: \(r)"
-        case .tokenizerLoadFailed(let r): return "Tokenizer load failed: \(r)"
-        case .stateInitFailed(let r): return "MLState init failed: \(r)"
-        case .generationFailed(let r): return "Generation failed: \(r)"
-        case .unsupportedModelFormat(let r): return "Unsupported model format: \(r)"
-        }
-    }
-}
+import Foundation
+
+public enum CoreMLBackendError: Error, LocalizedError, Sendable {
+    case modelLoadFailed(reason: String)
+    case tokenizerLoadFailed(reason: String)
+    case stateInitFailed(reason: String)
+    case generationFailed(reason: String)
+    case unsupportedModelFormat(reason: String)
+
+    public var errorDescription: String? {
+        switch self {
+        case .modelLoadFailed(let r): return "CoreML model load failed: \(r)"
+        case .tokenizerLoadFailed(let r): return "Tokenizer load failed: \(r)"
+        case .stateInitFailed(let r): return "MLState init failed: \(r)"
+        case .generationFailed(let r): return "Generation failed: \(r)"
+        case .unsupportedModelFormat(let r): return "Unsupported model format: \(r)"
+        }
+    }
+}

package/ios/Sources/DVAICoreMLCore/CoreMLHandlers.swift

@@ -1,123 +1,123 @@
-import Foundation
-#if !COCOAPODS
-import DVAISharedCore
-#endif
-
-/// `DVAIHandlers` conformer for the CoreML backend.
-/// Translates OpenAI-compatible HTTP requests into CoreMLGenerator calls and
-/// formats the results as OpenAI JSON / SSE responses.
-@available(iOS 18.0, macOS 15.0, *)
-public final class CoreMLHandlers: DVAIHandlers {
-    private let generator: CoreMLGenerator
-    private let modelId: String
-
-    // Internal init — `CoreMLGenerator` is an implementation detail of
-    // DVAICoreMLCore and stays internal. The only construction site is
-    // `CoreMLPluginState.start()` inside the same module.
-    internal init(generator: CoreMLGenerator, modelId: String) {
-        self.generator = generator
-        self.modelId = modelId
-    }
-
-    public func handleChatCompletion(body: [String: Any], ctx: HandlerContext) async throws -> HandlerResponse {
-        guard let messages = body["messages"] as? [[String: String]] else {
-            return .error(400, "messages array is required")
-        }
-        let stream = (body["stream"] as? Bool) ?? false
-        let temperature = (body["temperature"] as? Double).map(Float.init) ?? 0.0
-        let topP = (body["top_p"] as? Double).map(Float.init) ?? 1.0
-        let maxTokens = (body["max_tokens"] as? Int) ?? 512
-
-        // Build a generator with the per-request sampling params.
-        let requestSampler = CoreMLSampler(temperature: temperature, topP: topP, topK: 0)
-        let requestGenerator = CoreMLGenerator(
-            engine: generator.engine,
-            tokenizer: generator.tokenizer,
-            sampler: requestSampler,
-            maxNewTokens: maxTokens
-        )
-
-        let promptTokens: [Int]
-        do {
-            promptTokens = try generator.tokenizer.applyChatTemplate(messages: messages)
-        } catch {
-            return .error(400, "tokenizer chat-template failed: \(error.localizedDescription)")
-        }
-
-        if stream {
-            let sse = requestGenerator.generateStream(promptTokens: promptTokens)
-            let streamId = UUID().uuidString
-            let mappedStream = AsyncStream<String> { cont in
-                Task {
-                    do {
-                        for try await chunk in sse {
-                            let evt = "data: {\"id\":\"\(streamId)\",\"object\":\"chat.completion.chunk\",\"created\":\(Int(Date().timeIntervalSince1970)),\"model\":\"\(modelId)\",\"choices\":[{\"index\":0,\"delta\":{\"content\":\(jsonString(chunk))},\"finish_reason\":null}]}\n\n"
-                            cont.yield(evt)
-                        }
-                        cont.yield("data: [DONE]\n\n")
-                        cont.finish()
-                    } catch {
-                        cont.yield("data: {\"error\":\"\(error.localizedDescription)\"}\n\n")
-                        cont.finish()
-                    }
-                }
-            }
-            return .sse(mappedStream)
-        }
-
-        let text: String
-        do {
-            text = try await requestGenerator.generate(promptTokens: promptTokens)
-        } catch {
-            return .error(500, "generation failed: \(error.localizedDescription)")
-        }
-        let responseJSON: [String: Any] = [
-            "id": UUID().uuidString,
-            "object": "chat.completion",
-            "created": Int(Date().timeIntervalSince1970),
-            "model": modelId,
-            "choices": [[
-                "index": 0,
-                "message": ["role": "assistant", "content": text],
-                "finish_reason": "stop"
-            ]],
-            "usage": [
-                "prompt_tokens": promptTokens.count,
-                "completion_tokens": -1,  // CoreML decoding doesn't track this per checkpoint
-                "total_tokens": -1
-            ]
-        ]
-        return .json(200, responseJSON)
-    }
-
-    public func handleCompletion(body: [String: Any], ctx: HandlerContext) async throws -> HandlerResponse {
-        let prompt = body["prompt"] as? String ?? ""
-        let chatBody: [String: Any] = [
-            "messages": [["role": "user", "content": prompt]],
-            "stream": body["stream"] as? Bool ?? false,
-            "temperature": body["temperature"] as? Double ?? 0.0,
-            "top_p": body["top_p"] as? Double ?? 1.0,
-            "max_tokens": body["max_tokens"] as? Int ?? 512,
-        ]
-        return try await handleChatCompletion(body: chatBody, ctx: ctx)
-    }
-
-    public func handleEmbeddings(body: [String: Any], ctx: HandlerContext) async throws -> HandlerResponse {
-        return .error(501, "embeddings not yet supported by the CoreML backend")
-    }
-
-    public func handleModels(ctx: HandlerContext) async throws -> HandlerResponse {
-        return .json(200, [
-            "object": "list",
-            "data": [["id": modelId, "object": "model", "owned_by": "dvai-bridge"]]
-        ])
-    }
-
-    /// JSON-encode a single string value (produces a quoted JSON string).
-    private func jsonString(_ s: String) -> String {
-        let data = (try? JSONSerialization.data(withJSONObject: [s], options: [])) ?? Data()
-        let str = String(data: data, encoding: .utf8) ?? "[\"\"]"
-        // Strip the surrounding array brackets — leaves the quoted string value.
-        return String(str.dropFirst().dropLast())
-    }
-}
+import Foundation
+#if !COCOAPODS
+import DVAISharedCore
+#endif
+
+/// `DVAIHandlers` conformer for the CoreML backend.
+/// Translates OpenAI-compatible HTTP requests into CoreMLGenerator calls and
+/// formats the results as OpenAI JSON / SSE responses.
+@available(iOS 18.0, macOS 15.0, *)
+public final class CoreMLHandlers: DVAIHandlers {
+    private let generator: CoreMLGenerator
+    private let modelId: String
+
+    // Internal init — `CoreMLGenerator` is an implementation detail of
+    // DVAICoreMLCore and stays internal. The only construction site is
+    // `CoreMLPluginState.start()` inside the same module.
+    internal init(generator: CoreMLGenerator, modelId: String) {
+        self.generator = generator
+        self.modelId = modelId
+    }
+
+    public func handleChatCompletion(body: [String: Any], ctx: HandlerContext) async throws -> HandlerResponse {
+        guard let messages = body["messages"] as? [[String: String]] else {
+            return .error(400, "messages array is required")
+        }
+        let stream = (body["stream"] as? Bool) ?? false
+        let temperature = (body["temperature"] as? Double).map(Float.init) ?? 0.0
+        let topP = (body["top_p"] as? Double).map(Float.init) ?? 1.0
+        let maxTokens = (body["max_tokens"] as? Int) ?? 512
+
+        // Build a generator with the per-request sampling params.
+        let requestSampler = CoreMLSampler(temperature: temperature, topP: topP, topK: 0)
+        let requestGenerator = CoreMLGenerator(
+            engine: generator.engine,
+            tokenizer: generator.tokenizer,
+            sampler: requestSampler,
+            maxNewTokens: maxTokens
+        )
+
+        let promptTokens: [Int]
+        do {
+            promptTokens = try generator.tokenizer.applyChatTemplate(messages: messages)
+        } catch {
+            return .error(400, "tokenizer chat-template failed: \(error.localizedDescription)")
+        }
+
+        if stream {
+            let sse = requestGenerator.generateStream(promptTokens: promptTokens)
+            let streamId = UUID().uuidString
+            let mappedStream = AsyncStream<String> { cont in
+                Task {
+                    do {
+                        for try await chunk in sse {
+                            let evt = "data: {\"id\":\"\(streamId)\",\"object\":\"chat.completion.chunk\",\"created\":\(Int(Date().timeIntervalSince1970)),\"model\":\"\(modelId)\",\"choices\":[{\"index\":0,\"delta\":{\"content\":\(jsonString(chunk))},\"finish_reason\":null}]}\n\n"
+                            cont.yield(evt)
+                        }
+                        cont.yield("data: [DONE]\n\n")
+                        cont.finish()
+                    } catch {
+                        cont.yield("data: {\"error\":\"\(error.localizedDescription)\"}\n\n")
+                        cont.finish()
+                    }
+                }
+            }
+            return .sse(mappedStream)
+        }
+
+        let text: String
+        do {
+            text = try await requestGenerator.generate(promptTokens: promptTokens)
+        } catch {
+            return .error(500, "generation failed: \(error.localizedDescription)")
+        }
+        let responseJSON: [String: Any] = [
+            "id": UUID().uuidString,
+            "object": "chat.completion",
+            "created": Int(Date().timeIntervalSince1970),
+            "model": modelId,
+            "choices": [[
+                "index": 0,
+                "message": ["role": "assistant", "content": text],
+                "finish_reason": "stop"
+            ]],
+            "usage": [
+                "prompt_tokens": promptTokens.count,
+                "completion_tokens": -1,  // CoreML decoding doesn't track this per checkpoint
+                "total_tokens": -1
+            ]
+        ]
+        return .json(200, responseJSON)
+    }
+
+    public func handleCompletion(body: [String: Any], ctx: HandlerContext) async throws -> HandlerResponse {
+        let prompt = body["prompt"] as? String ?? ""
+        let chatBody: [String: Any] = [
+            "messages": [["role": "user", "content": prompt]],
+            "stream": body["stream"] as? Bool ?? false,
+            "temperature": body["temperature"] as? Double ?? 0.0,
+            "top_p": body["top_p"] as? Double ?? 1.0,
+            "max_tokens": body["max_tokens"] as? Int ?? 512,
+        ]
+        return try await handleChatCompletion(body: chatBody, ctx: ctx)
+    }
+
+    public func handleEmbeddings(body: [String: Any], ctx: HandlerContext) async throws -> HandlerResponse {
+        return .error(501, "embeddings not yet supported by the CoreML backend")
+    }
+
+    public func handleModels(ctx: HandlerContext) async throws -> HandlerResponse {
+        return .json(200, [
+            "object": "list",
+            "data": [["id": modelId, "object": "model", "owned_by": "dvai-bridge"]]
+        ])
+    }
+
+    /// JSON-encode a single string value (produces a quoted JSON string).
+    private func jsonString(_ s: String) -> String {
+        let data = (try? JSONSerialization.data(withJSONObject: [s], options: [])) ?? Data()
+        let str = String(data: data, encoding: .utf8) ?? "[\"\"]"
+        // Strip the surrounding array brackets — leaves the quoted string value.
+        return String(str.dropFirst().dropLast())
+    }
+}