How bghira/simpletuner works
SimpleTuner is a multimodal diffusion-model training toolkit with a Web UI, CLI, and multi-user worker platform. The principal workflow configures a training job, obtains and prepares data, trains a supported model, and monitors results; the Web UI also supports checkpoint inference and cloud-worker orchestration. The README documents broad capabilities, while sampled source shows selected server, process, inference, and model-pipeline behavior. Links for unsampled subsystems rely on README statements, not source inspection.
Main entry points: app.py Server application · __main__.py Training runtime · train.py CLI · cmd_args.py Configuration
Components 21
Training runtime
- CLI
simpletuner/cli/train.py - Configuration
simpletuner/helpers/configuration/cmd_args.py - Dataset backends
simpletuner/helpers/data_backend/factory.py - Embedding cache
simpletuner/helpers/caching/text_embeds.py - Training runtime
simpletuner/__main__.py - Distributed training
simpletuner/helpers/acceleration/backends.py - Logging
simpletuner/helpers/logging.py
Model support
- Model registry
simpletuner/helpers/models/all.py - Model pipelines
simpletuner/helpers/models/sdxl/pipeline.py - LoRA and LyCORIS
simpletuner/helpers/training/lycoris.py
Web interface
- Web UI
simpletuner/static/js/sse-manager.js - Server application
simpletuner/simpletuner_sdk/server/app.py - Config store
simpletuner/simpletuner_sdk/server/services/config_store.py - Checkpoint inference
simpletuner/simpletuner_sdk/server/services/checkpoint_inference_service.py - Process manager
simpletuner/simpletuner_sdk/process_keeper.py
Cloud platform
- Worker orchestration
simpletuner/simpletuner_sdk/server/services/worker_manager.py - Worker agent
simpletuner/worker_agent.py - Approval rules
simpletuner/simpletuner_sdk/server/services/cloud/approval/rules_engine.py - Cloud storage API
simpletuner/simpletuner_sdk/server/routes/cloud/storage.py - Storage backend
simpletuner/simpletuner_sdk/server/services/cloud/storage/base.py
Other components
- User
Connections 21
- User → CLI: configures job
README.md - CLI → Configuration: loads settings
- Configuration → Dataset backends: selects datasets
- Dataset backends → Embedding cache: provides data
- Embedding cache → Training runtime: supplies embeddings
- Training runtime → Model registry: selects architecture
- Training runtime → Distributed training: uses sharding
README.md - Training runtime → LoRA and LyCORIS: trains adapters
README.md - Training runtime → Logging: emits logs
- Model registry → Model pipelines: selects pipeline
- User → Web UI: manages training
README.md - Web UI → Server application: sends requests
- Server application → Config store: loads configs
simpletuner/simpletuner_sdk/server/app.py - Server application → Process manager: starts jobs
simpletuner/simpletuner_sdk/server/app.py - Server application → Checkpoint inference: routes inference
- Checkpoint inference → Process manager: sends commands
simpletuner/simpletuner_sdk/server/services/checkpoint_inference_service.py - Server application → Worker orchestration: manages workers
simpletuner/simpletuner_sdk/server/app.py - Worker orchestration → Worker agent: dispatches jobs
README.md - Server application → Approval rules: evaluates approvals
- Worker agent → Cloud storage API: uploads artifacts
README.md - Cloud storage API → Storage backend: stores objects