Add DeepSeek VL 7B support from replicate (#112)

Author: lzell

README.md

@@ -2564,6 +2564,51 @@ See the full range of controls for generating an image by viewing `ReplicateSDXL
 
 See the full range of controls for generating an image by viewing `ReplicateSDXLFreshInkInputSchema.swift`
 
+### How to call DeepSeek's 7B vision model on replicate
+
+Add a file called 'my-image.jpg' to Xcode app assets. Then run this snippet:
+
+```swift
+    import AIProxy
+
+    /* Uncomment for BYOK use cases */
+    // let replicateService = AIProxy.replicateDirectService(
+    //     unprotectedAPIKey: "your-replicate-key"
+    // )
+
+    /* Uncomment for all other production use cases */
+    // let replicateService = AIProxy.replicateService(
+    //     partialKey: "partial-key-from-your-developer-dashboard",
+    //     serviceURL: "service-url-from-your-developer-dashboard"
+    // )
+
+    guard let image = NSImage(named: "my-image") else {
+        print("Could not find an image named 'my-image' in your app assets")
+        return
+    }
+
+    guard let imageURL = AIProxy.encodeImageAsURL(image: image, compressionQuality: 0.4) else {
+        print("Could not encode image as a data URI")
+        return
+    }
+
+    do {
+        let input = ReplicateDeepSeekVL7BInputSchema(
+            image: imageURL,
+            prompt: "What are the colors in this pic"
+        )
+        let description = try await replicateService.runDeepSeekVL7B(input: input, secondsToWait: 300)
+        print("Done getting descriptions from DeepSeekVL7B: ", description)
+    } catch AIProxyError.unsuccessfulRequest(let statusCode, let responseBody) {
+        print("Received \(statusCode) status code with response body: \(responseBody)")
+    } catch {
+        // You may want to catch additional Foundation errors and pop the appropriate UI
+        // to the user. See "How to catch Foundation errors for specific conditions" here:
+        // https://www.aiproxy.com/docs/integration-options.html
+        print("Could not use deepseek vision on replicate: \(error.localizedDescription)")
+    }
+```
+
 
 ### How to call your own models on Replicate.
 

Sources/AIProxy/AIProxy.swift

@@ -8,7 +8,7 @@ import UIKit
 public struct AIProxy {
 
     /// The current sdk version
-    public static let sdkVersion = "0.74.0"
+    public static let sdkVersion = "0.75.0"
 
     /// - Parameters:
     ///   - partialKey: Your partial key is displayed in the AIProxy dashboard when you submit your provider's key.

Sources/AIProxy/OpenAI/OpenAIRealtimeSession.swift

@@ -23,7 +23,7 @@ open class OpenAIRealtimeSession {
         self.sessionConfiguration = sessionConfiguration
 
         Task {
-            try await self.sendMessage(OpenAIRealtimeSessionUpdate(session: self.sessionConfiguration))
+            await self.sendMessage(OpenAIRealtimeSessionUpdate(session: self.sessionConfiguration))
         }
         self.webSocketTask.resume()
         self.receiveMessage()

Sources/AIProxy/Replicate/ReplicateDeepSeekVL7BInputSchema.swift

@@ -0,0 +1,46 @@
+//
+//  ReplicateDeepSeekVL7BInputSchema.swift
+//  AIProxy
+//
+//  Created by Lou Zell on 3/5/25.
+//
+
+import Foundation
+
+/// https://replicate.com/deepseek-ai/deepseek-vl-7b-base?output=json
+public struct ReplicateDeepSeekVL7BInputSchema: Encodable {
+    // Required
+
+    /// Input image
+    public let image: URL
+
+    // Optional
+
+    /// Maximum number of tokens to generate
+    /// Default: 512
+    public let maxNewTokens: Int?
+
+    /// Input prompt
+    /// Default: "Describe this image"
+    public let prompt: String?
+
+
+    private enum CodingKeys: String, CodingKey {
+        case image
+        case maxNewTokens = "max_new_tokens"
+        case prompt
+    }
+
+    // This memberwise initializer is autogenerated.
+    // To regenerate, use `cmd-shift-a` > Generate Memberwise Initializer
+    // To format, place the cursor in the initializer's parameter list and use `ctrl-m`
+    public init(
+        image: URL,
+        maxNewTokens: Int? = nil,
+        prompt: String? = nil
+    ) {
+        self.image = image
+        self.maxNewTokens = maxNewTokens
+        self.prompt = prompt
+    }
+}

Sources/AIProxy/Replicate/ReplicateService+Convenience.swift

@@ -211,6 +211,30 @@ extension ReplicateService {
         return try await self.getPredictionOutput(prediction)
     }
 
+    /// Convenience method for running the DeepSeek 7B vision model:
+    /// https://replicate.com/deepseek-ai/deepseek-vl-7b-base
+    ///
+    /// In my testing, vision requests were completing in less than 2 seconds once the model was warm.
+    /// Note that the result can take several minutes if the model is cold.
+    ///
+    /// - Parameters:
+    ///   - input: The input containing the image and prompt that you want to have DeepSeek VL inspect
+    ///   - secondsToWait: Seconds to wait before raising a timeout error
+    ///
+    /// - Returns: The generated content
+    public func runDeepSeekVL7B(
+        input: ReplicateDeepSeekVL7BInputSchema,
+        version: String = "d1823e6f68cd3d57f2d315b9357dfa85f53817120ae0de8d2b95fbc8e93a1385",
+        secondsToWait: UInt
+    ) async throws -> String {
+        let prediction: ReplicatePrediction<[String]> = try await self.runCommunityModel(
+            version: version,
+            input: input,
+            secondsToWait: secondsToWait
+        )
+        return (try await self.getPredictionOutput(prediction)).joined(separator: "")
+    }
+
     // MARK: - Deprecated
     @available(*, deprecated, message: "Please use createFluxSchnellImages")
     public func createFluxSchnellImageURLs(