wehub-resource-sync
120 lines
3.3 KiB
Swift
120 lines
3.3 KiB
Swift
// PR 12 — oQ Quantization DTOs.
|
|
//
|
|
// Mirrors omlx/admin/oq_manager.py:QuantTask.to_dict (line 79-98) and the
|
|
// /admin/api/oq/* responses defined in omlx/admin/routes.py:4881-4992.
|
|
//
|
|
// The client uses `convertFromSnakeCase` so snake_case server keys map to
|
|
// camelCase Swift fields automatically (e.g. `output_size_formatted` →
|
|
// `outputSizeFormatted`).
|
|
|
|
import Foundation
|
|
|
|
// MARK: - Model listing
|
|
|
|
/// One row in the source-model list. Quantizable rows come from
|
|
/// `source_models` (added `numLayers`, `numExperts`, `memoryStreaming`); the
|
|
/// sensitivity picker pulls from `allModels` (no extra fields needed).
|
|
struct OQModelInfo: Codable, Equatable, Sendable, Identifiable {
|
|
let path: String
|
|
let name: String
|
|
let size: Int64
|
|
let sizeFormatted: String
|
|
let modelType: String
|
|
let isQuantized: Bool
|
|
let isVlm: Bool
|
|
let hasMtpHeads: Bool
|
|
|
|
// Only populated on `source_models` rows.
|
|
let numLayers: Int?
|
|
let numExperts: Int?
|
|
let memoryStreaming: OQMemoryEstimate?
|
|
|
|
var id: String { path }
|
|
}
|
|
|
|
/// Streaming-memory estimate attached to each source model (oq.py:1427).
|
|
struct OQMemoryEstimate: Codable, Equatable, Sendable {
|
|
let peakBytes: Int64
|
|
let peakFormatted: String
|
|
}
|
|
|
|
struct OQModelsResponse: Codable, Sendable {
|
|
let models: [OQModelInfo]
|
|
let allModels: [OQModelInfo]
|
|
}
|
|
|
|
// MARK: - Estimate
|
|
|
|
/// Response from `GET /admin/api/oq/estimate`. The Python side may also
|
|
/// include `memory_streaming_formatted` so we surface it as optional.
|
|
struct OQEstimateResponse: Codable, Equatable, Sendable {
|
|
let effectiveBpw: Double
|
|
let outputSizeBytes: Int64
|
|
let outputSizeFormatted: String
|
|
let memoryStreamingFormatted: String?
|
|
}
|
|
|
|
// MARK: - Start quantization
|
|
|
|
/// Body for `POST /admin/api/oq/start`. Field names match
|
|
/// `OQStartRequest` (routes.py:288) and rely on the client's
|
|
/// `convertToSnakeCase` encoder.
|
|
struct OQStartRequest: Encodable, Sendable {
|
|
let modelPath: String
|
|
let oqLevel: Double
|
|
let groupSize: Int
|
|
let sensitivityModelPath: String
|
|
let textOnly: Bool
|
|
let dtype: String
|
|
let preserveMtp: Bool
|
|
}
|
|
|
|
struct OQStartResponse: Decodable, Sendable {
|
|
let success: Bool
|
|
let task: OQTaskDTO?
|
|
}
|
|
|
|
// MARK: - Tasks
|
|
|
|
/// Mirrors `QuantTask.to_dict` (oq_manager.py:79). The `status` field is a
|
|
/// string in the JSON; we expose `statusEnum` for callers that switch on it.
|
|
struct OQTaskDTO: Codable, Equatable, Sendable, Identifiable {
|
|
let taskId: String
|
|
let modelName: String
|
|
let modelPath: String
|
|
let oqLevel: Double
|
|
let outputName: String
|
|
let outputPath: String
|
|
let status: String
|
|
let progress: Double
|
|
let phase: String
|
|
let error: String
|
|
let createdAt: Double
|
|
let startedAt: Double
|
|
let completedAt: Double
|
|
let sourceSize: Int64
|
|
let outputSize: Int64
|
|
let dtype: String
|
|
|
|
var id: String { taskId }
|
|
|
|
enum Status: String {
|
|
case pending, loading, quantizing, saving, completed, failed, cancelled
|
|
}
|
|
|
|
var statusEnum: Status? { Status(rawValue: status) }
|
|
|
|
/// True while the task occupies the quantization slot (and the screen
|
|
/// should keep polling).
|
|
var isActive: Bool {
|
|
switch statusEnum {
|
|
case .pending, .loading, .quantizing, .saving: return true
|
|
default: return false
|
|
}
|
|
}
|
|
}
|
|
|
|
struct OQTasksResponse: Codable, Sendable {
|
|
let tasks: [OQTaskDTO]
|
|
}
|