Skip to content

model-manager.md

Latest commit

 

History

History
165 lines (103 loc) · 2.71 KB

model-manager.md

File metadata and controls

165 lines (103 loc) · 2.71 KB

ModelManager

Handles lifecycle of ML models on device: download, import, validation, listing, deletion, and model instantiation.

ROOT_DIR

filesDir/models/

downloadModelInternal

suspend fun downloadModelInternal(
    context: Context,
    modelInfo: ModelInfo,
    onProgress: ((Int) -> Unit)? = null
): Result<File>

Downloads a model into internal storage.

  • Streams file from network to local storage
  • Reports progress (0–100) if available
  • Uses temporary file then atomic rename
  • Returns final model file

downloadModelExternal

fun downloadModelExternal(context: Context, url: String)

Opens external browser for model download URL.

importModel

suspend fun importModel(context: Context, modelInfo: ModelInfo, uri: Uri)

Imports a model from a URI into internal storage.

  • Copies URI content into model directory
  • Replaces existing model if present

modelExists

fun modelExists(context: Context, model: ModelName): Boolean

Checks whether a model is fully available locally.

deleteModel

fun deleteModel(context: Context, modelInfo: ModelInfo): Boolean

Deletes a model from internal storage.

listModels

fun listModels(context: Context, type: ModelType? = null): List<ModelName>

Returns installed models, optionally filtered by type.

getModelFile

fun getModelFile(context: Context, modelInfo: ModelInfo): File

Resolves model path:

filesDir/models/<modelInfo.path>

downloadFileInternal

Core network download utility.

  • Streams HTTP response to local file
  • Writes to temp file first
  • Reports progress if content length is known
  • Replaces final file atomically

Model Providers

Text Embedders

fun getTextEmbedder(context: Context, modelName: ModelName): TextEmbeddingProvider

Returns initialized text embedding model.

Supported models:

  • ALL_MINILM_L6_V2
  • CLIP_VIT_B_32_TEXT

Throws ModelNotDownloaded if missing.

Image Embedders

fun getImageEmbedder(context: Context, modelName: ModelName): ImageEmbeddingProvider

Supported models:

  • DINOV2_SMALL
  • CLIP_VIT_B_32_IMAGE
  • INCEPTION_RESNET_V1

Throws ModelNotDownloaded if missing.

Object Detectors

fun getObjectDetector(context: Context, modelName: ModelName): DetectorProvider

Supported models:

  • ULTRA_LIGHT_FACE_DETECTOR

Throws ModelNotDownloaded if missing.

Design Notes

  • Internal storage only (filesDir/models)
  • Atomic file replacement for safety
  • Progress-aware streaming downloads
  • Model validation enforced before instantiation
  • Strict model-type matching for provider factories