v2: add EmbeddingServerApp, fix Sendable issues (WIP)
This commit is contained in:
@@ -20,16 +20,36 @@ struct SimpleServerApp {
|
||||
print("")
|
||||
|
||||
let engine = try MarkBaseEngine(autoCompile: true)
|
||||
let model = try E4BModel(modelDir: modelPath, engine: engine, maxContextLength: 512)
|
||||
|
||||
// Detect model type
|
||||
let isEmbeddingGemma = modelName.contains("embeddinggemma") || modelName.contains("embedding")
|
||||
let isE4B = !isEmbeddingGemma
|
||||
|
||||
var e4bModel: E4BModel? = nil
|
||||
var embedModel: EmbeddingGemmaModel? = nil
|
||||
let tokenizer = try TokenizerFactory.load(modelDir: modelPath)
|
||||
let generator = StreamingGenerator(model: model, tokenizer: tokenizer, engine: engine)
|
||||
let embeddingModel = try TextEmbeddingModel(modelDir: modelPath, engine: engine, config: TextEmbeddingConfig())
|
||||
|
||||
print("✓ E4B loaded (\(model.numHiddenLayers) layers)")
|
||||
var generator: StreamingGenerator? = nil
|
||||
|
||||
if isE4B {
|
||||
e4bModel = try E4BModel(modelDir: modelPath, engine: engine, maxContextLength: 512)
|
||||
generator = StreamingGenerator(model: e4bModel!, tokenizer: tokenizer, engine: engine)
|
||||
print("✓ E4B loaded (\(e4bModel!.numHiddenLayers) layers)")
|
||||
} else {
|
||||
embedModel = try EmbeddingGemmaModel(modelDir: modelPath, engine: engine)
|
||||
print("✓ EmbeddingGemma loaded (\(embedModel!.config.numHiddenLayers) layers, hidden=\(embedModel!.config.hiddenSize))")
|
||||
}
|
||||
|
||||
// Embedding model (use EmbeddingGemma if available, fallback to E4B)
|
||||
let textEmbedModel: TextEmbeddingModel?
|
||||
if isE4B {
|
||||
textEmbedModel = try TextEmbeddingModel(modelDir: modelPath, engine: engine, config: TextEmbeddingConfig())
|
||||
} else {
|
||||
textEmbedModel = nil
|
||||
}
|
||||
|
||||
let router = Router()
|
||||
|
||||
let layers = model.numHiddenLayers
|
||||
let layers = isE4B ? e4bModel!.numHiddenLayers : embedModel!.config.numHiddenLayers
|
||||
|
||||
@Sendable func helpJSON() -> String {
|
||||
return """
|
||||
@@ -184,7 +204,8 @@ struct SimpleServerApp {
|
||||
}
|
||||
|
||||
@Sendable func healthResponse() -> String {
|
||||
return "{\"status\":\"healthy\",\"model\":\"e4b\",\"layers\":\(layers)}"
|
||||
let m = isEmbeddingGemma ? "embeddinggemma-300m" : "e4b"
|
||||
return "{\"status\":\"healthy\",\"model\":\"\(m)\",\"layers\":\(layers)}"
|
||||
}
|
||||
|
||||
router.get("/") { _, _ in
|
||||
@@ -229,7 +250,7 @@ struct SimpleServerApp {
|
||||
topP: topP
|
||||
)
|
||||
|
||||
let generatedTokens = try generator.generateTokens(promptTokens: promptTokens, config: config)
|
||||
let generatedTokens = try generator!.generateTokens(promptTokens: promptTokens, config: config)
|
||||
|
||||
let id = UUID().uuidString
|
||||
let ts = Int(Date().timeIntervalSince1970)
|
||||
@@ -313,7 +334,14 @@ struct SimpleServerApp {
|
||||
var embeddings: [[String: Any]] = []
|
||||
for (i, text) in inputs.enumerated() {
|
||||
let t0 = Date()
|
||||
let embedding = try embeddingModel.embed(text: text)
|
||||
let embedding: [Float]
|
||||
if let eg = embedModelCopy {
|
||||
embedding = try eg.embed(text: text)
|
||||
} else if let tem = textEmbedModelCopy {
|
||||
embedding = try tem.embed(text: text)
|
||||
} else {
|
||||
return "{\"error\":\"no embedding model available\",\"type\":\"server_error\",\"code\":500}"
|
||||
}
|
||||
let duration = Date().timeIntervalSince(t0)
|
||||
|
||||
let embeddingData: [String: Any]
|
||||
@@ -338,7 +366,7 @@ struct SimpleServerApp {
|
||||
|
||||
let id = UUID().uuidString
|
||||
let ts = Int(Date().timeIntervalSince1970)
|
||||
let totalTokens = inputs.reduce(0) { $0 + tokenizer.encode(text: $1).count }
|
||||
let totalTokens = inputs.reduce(0) { $0 + $1.components(separatedBy: .whitespaces).count }
|
||||
|
||||
let response: [String: Any] = [
|
||||
"id": id,
|
||||
|
||||
Reference in New Issue
Block a user