-
feat(api): implement local training MVP with adapter pattern
liumaolin
- Add adapter base classes (TaskQueue, Progress, Storage, Database)
- Implement local adapters (AsyncTrainingManager, SQLite, LocalStorage)
- Add Pydantic schemas for tasks, experiments, files, and stages
- Implement service layer (TaskService, ExperimentService, FileService)
- Add API endpoints for Quick Mode (/tasks) and Advanced Mode (/experiments)
- Add adapter factory with dependency injection support
- Update architecture design document with implementation status
-
feat(api): implement AsyncTrainingManager MVP with SQLite persistence
liumaolin
- Add api_server module with adapter pattern architecture
- Implement AsyncTrainingManager using asyncio.subprocess + SQLite
- Add TaskQueueAdapter abstract base class for future server mode
- Create domain models: Task, TaskStatus, ProgressInfo
- Add run_pipeline.py wrapper script for subprocess execution
- Create config module for centralized environment variables
- Add aiosqlite dependency to pyproject.toml
- Include test config files for pipeline validation
The AsyncTrainingManager provides:
- Async task queue with SQLite persistence
- Real-time progress tracking via stdout JSON parsing
- Task cancellation and status querying
- Progress subscription for SSE streaming
- Application restart recovery support
