Something went wrong. Try again.
Routing embedding stack for Qwen3-Embedding-0.6B on Apple Silicon: orchestrator + MLX + llama.cpp backends
Something went wrong. Try again.
1.1 kB · 43 lines
Swift
at main
1234567891011121314151617181920212223242526272829303132333435363738394041424344import Foundation
enum InputKind: String, Codable, Sendable { case document case query}
enum ModelSize: String, Codable, Sendable, CaseIterable { case s06 = "0.6B" case s4 = "4B" case s8 = "8B"
/// Official Qwen org F16 weights. mlx-community publishes only quantized /// variants for these sizes; the F16 repos are the canonical correctness /// oracle. mlx-swift-lm's `Qwen3Model.sanitize` consumes them as-is. var repoId: String { switch self { case .s06: return "Qwen/Qwen3-Embedding-0.6B" case .s4: return "Qwen/Qwen3-Embedding-4B" case .s8: return "Qwen/Qwen3-Embedding-8B" } }
var label: String { switch self { case .s06: return "qwen3-embedding-0.6B" case .s4: return "qwen3-embedding-4B" case .s8: return "qwen3-embedding-8B" } }}
struct EmbedRequest: Codable, Sendable { let texts: [String] let input_type: InputKind let size: ModelSize?}
struct EmbedResponse: Codable, Sendable { let vectors: [[Float]] let dim: Int let model: String}