Skip to content

Planned Modules Roadmap¤

This page is a checked status surface over the current package tree. It is not a freehand completeness matrix, and it intentionally avoids summary totals that drift away from the live runtime inventory.

Status Legend¤

Status Meaning
Shipped Importable and part of the retained runtime surface
Experimental Importable, but intentionally narrow, unstable, or placeholder-heavy
Roadmap-only Not importable today, or intentionally not presented as current support

Current Runtime Status¤

Generative Models¤

Family Status Runtime evidence Notes
VAE Shipped artifex.generative_models.models.vae Retained VAE owners and typed configs
GAN Shipped artifex.generative_models.models.gan Retained GAN owners such as DCGAN and PatchGAN
Diffusion Shipped artifex.generative_models.models.diffusion Retained diffusion owners such as DDPM and DiT
Flow Shipped artifex.generative_models.models.flow Retained flow owners include RealNVP, Glow, MAF, IAF, NeuralSplineFlow, and ConditionalRealNVP
Autoregressive Shipped artifex.generative_models.models.autoregressive Retained autoregressive owners
Energy-Based Shipped artifex.generative_models.models.energy Retained EBM owners
Geometric Shipped artifex.generative_models.models.geometric Retained point-cloud, graph, mesh, voxel, and protein owners
Audio models Experimental artifex.generative_models.models.audio Importable, but narrower and less settled than the core image/model families
Backbones Shipped artifex.generative_models.models.backbones Shared retained backbones used by multiple families

Core Framework¤

Surface Status Runtime evidence Notes
Base protocols and classes Shipped artifex.generative_models.models.base, artifex.generative_models.core.base Current public protocol split includes GenerativeModelProtocol and TrainableGenerativeModelProtocol
Losses, distributions, and sampling Shipped artifex.generative_models.core.losses, core.distributions, core.sampling Retained runtime owners
Layers and configuration Shipped artifex.generative_models.core.layers, core.configuration Typed-config runtime surface
Evaluation package Experimental artifex.generative_models.core.evaluation Current top-level exports are only benchmarks and metrics
Device management Shipped artifex.generative_models.core.device_manager Retained runtime owner

Training¤

Surface Status Runtime evidence Notes
Trainers and training loops Shipped artifex.generative_models.training Retained trainer and loop surface
Callbacks Shipped artifex.generative_models.training.callbacks Current callback surface
Distributed data parallel helper Shipped artifex.generative_models.training.distributed.data_parallel Narrow retained distributed helper

Extensions¤

Surface Status Runtime evidence Notes
Protein extensions Shipped artifex.generative_models.extensions.protein Retained protein-extension bundle

Modalities¤

Family Status Runtime evidence Notes
Image Shipped artifex.generative_models.modalities ImageModality
Text Shipped artifex.generative_models.modalities TextModality
Protein Shipped artifex.generative_models.modalities ProteinModality
Timeseries Shipped artifex.generative_models.modalities TimeseriesModality
Audio Experimental artifex.generative_models.modalities AudioModality remains narrower than the core retained image/text surface
Molecular Experimental artifex.generative_models.modalities MolecularModality
Tabular Experimental artifex.generative_models.modalities TabularModality
Multi-modal helper Experimental artifex.generative_models.modalities.multi_modal Importable helper package, but not registry-backed

Inference Pipeline¤

Surface Status Runtime evidence Notes
Inference package namespace Experimental artifex.generative_models.inference Live package namespace; the retained runtime surface is narrow
Production optimization pocket Experimental artifex.generative_models.inference.optimization.production ProductionOptimizer plus OptimizationTarget, OptimizationResult, and MonitoringMetrics; only jit_compilation is a real applied optimization today

Benchmarks¤

Surface Status Runtime evidence Notes
Benchmark package Experimental artifex.benchmarks Registry plus retained suites and model adapters
Retained suites Experimental GeometricBenchmarkSuite, ProteinLigandBenchmarkSuite, MultiBetaVAEBenchmarkSuite Useful checked-in coverage, but not a complete stable benchmark matrix

Roadmap-Only Surfaces¤

These entries are still coming soon and should not be read as part of the current importable runtime.

Surface Status Evidence Notes
Video modality Roadmap-only artifex.generative_models.modalities.video is not importable Keep video in roadmap pages until a real module exists
Neural ODE flow owners Roadmap-only No Neural ODE owner is exported from artifex.generative_models.models.flow Do not publish as current support
CNF flow owners Roadmap-only No CNF owner is exported from artifex.generative_models.models.flow Do not publish as current support
Broader inference stacks Roadmap-only Adaptation, batching, conversion, serving, and per-family generator layers are not a checked-in supported inference package Keep these in roadmap language until real modules land
Complete benchmark matrix Roadmap-only Current benchmark layer is partial and experimental Additional datasets, metrics, suites, and reporting remain roadmap work
StyleGAN, standalone UNet, conditioning package pages Roadmap-only The corresponding standalone runtime modules are not currently shipped Keep them as roadmap references only

Fine-Tuning¤

There is no standalone artifex.fine_tuning runtime package today. Current RL trainers live under artifex.generative_models.training, while LoRA, prefix tuning, prompt tuning, distillation, few-shot transfer, and RLHF remain roadmap-only topics.

Data Pipeline¤

A broad internal data package is still roadmap-only. Current example and tutorial flows use direct arrays plus datarax helpers where appropriate. Dataset, loader, preprocessing, tokenizer, and streaming families beyond that retained surface remain planned work.

CLI¤

The supported CLI surface is the retained config command entrypoint under artifex.core.cli.config_commands. Broader train/generate/serve/evaluate CLI families remain roadmap-only.

Utilities¤

The retained utility surface is small. artifex.utils.file_utils, artifex.generative_models.utils.jax.device, the logging helpers, code-analysis helpers, and protein visualization compatibility layers are live. Most other utility families remain coming soon and should stay on roadmap-only pages until real importable modules exist.

Configuration¤

The retained configuration runtime is the frozen-dataclass-based surface under artifex.generative_models.core.configuration plus the checked-in config loader and template management helpers. Older freehand utility families such as config merge or conversion helpers remain roadmap-only unless they are reintroduced as real importable owners.

Maintenance Notes¤

  • Keep current-runtime rows tied to importable modules or exported owners.
  • Keep missing families in the roadmap-only section until a real runtime module exists.
  • Do not reintroduce summary-count tables unless they are mechanically derived from the checked-in inventory.