8a66b9086a
- Started from ac75faa (initial E4B-MarkBase integration)
- Kept Sources/ (all engine code) + Package.swift + .gitignore
- Removed all ad-hoc tests, documentation, scripts, Python files
- Added Tests/00_Unit/ (MathTest, TokenizerTest, SamplerTest)
- Added .gitea/workflows/ci.yaml (build + unit tests + lint)
- Added Scripts/check_resources.sh (memory-aware test runner)
- Added Tests/Manifest.json (resource requirements for all tests)
- Focus: 4-bit quantized models only
45 lines
1.4 KiB
Swift
45 lines
1.4 KiB
Swift
import Foundation
|
|
|
|
public struct VisionConfig: Codable {
|
|
public let hiddenSize: Int
|
|
public let numAttentionHeads: Int
|
|
public let numHiddenLayers: Int
|
|
public let headDim: Int
|
|
public let globalHeadDim: Int
|
|
public let intermediateSize: Int
|
|
public let hiddenAct: String
|
|
public let rmsNormEps: Float
|
|
public let outputProjDims: Int
|
|
public let patchSize: Int
|
|
public let imageSize: Int
|
|
|
|
public init(
|
|
hiddenSize: Int = 768,
|
|
numAttentionHeads: Int = 12,
|
|
numHiddenLayers: Int = 12,
|
|
headDim: Int = 64,
|
|
globalHeadDim: Int = 64,
|
|
intermediateSize: Int = 3072,
|
|
hiddenAct: String = "gelu_pytorch_tanh",
|
|
rmsNormEps: Float = 1e-6,
|
|
outputProjDims: Int = 1536,
|
|
patchSize: Int = 14,
|
|
imageSize: Int = 224
|
|
) {
|
|
self.hiddenSize = hiddenSize
|
|
self.numAttentionHeads = numAttentionHeads
|
|
self.numHiddenLayers = numHiddenLayers
|
|
self.headDim = headDim
|
|
self.globalHeadDim = globalHeadDim
|
|
self.intermediateSize = intermediateSize
|
|
self.hiddenAct = hiddenAct
|
|
self.rmsNormEps = rmsNormEps
|
|
self.outputProjDims = outputProjDims
|
|
self.patchSize = patchSize
|
|
self.imageSize = imageSize
|
|
}
|
|
|
|
public var numPatches: Int {
|
|
(imageSize / patchSize) * (imageSize / patchSize)
|
|
}
|
|
} |