apogee-ai-training¶
Fine-tuning recipes (LoRA/QLoRA), distillation, RLHF/DPO entrypoints.
This page is being expanded
The tables below are generated from the package source, so they are accurate. Narrative guides and worked examples are being written module by module.
Public API¶
Everything below is exported from the package root, so from apogee_ai_training import X works for each name.
| Layer | Symbols |
|---|---|
| Application · DTOs | BenchDTO, ConvertDTO, PrepareDTO, StartDTO |
| Application · Use cases | BenchPrepUseCase, ConvertFormatUseCase, GetJobUseCase, PrepareDatasetUseCase, StartJobUseCase, ValidateDatasetUseCase |
| Domain | DatasetFormat, JobStatus, TrainingConfig, TrainingDataset, TrainingExample, TrainingJob |
| Domain · Enums | TrainerKind, TrainingKind |
| Domain · Exceptions | DatasetValidationError, JobNotFoundException, TrainerNotAvailableException, TrainingError |
| Domain · Protocols (ports) | IFormatConverter, IJobRepository, ITrainer |
| Infrastructure | AlpacaConverter, ChatMLConverter, DryRunTrainer, HfDpoConverter, InMemoryJobRepository, JsonlDatasetLoader, OpenAISftConverter, OpenAITrainer, write_jsonl |
Install¶
The base install has no required third-party dependency. Each extra unlocks one adapter.
| Extra | Pulls in |
|---|---|
openai |
openai>=1.30 |
huggingface |
huggingface_hub>=0.23 |
CLI¶
Installed as a console script, and registered as an apogee plugin so the same subcommands work under the global binary.
Read next¶
- Installation
- Reference — every public symbol, generated from the source
- Apogee Framework — how the packages fit together