entropy-flux/TorchSystem

TorchSystem is a Python library that provides a domain-driven, event-driven architecture for building and compiling PyTorch neural networks. It introduces a Compiler pipeline for model construction, a registry for metadata and serialization, and service components like Publisher/Subscriber and Producer/Consumer for decoupled logic. The system supports dependency injection and lifecycle management to facilitate complex training workflows and distributed scenarios.

Added new example applications for training a Multi-Layer Perceptron on MNIST and a Vision Transformer on CIFAR-10. These examples demonstrate the framework's training loop, model compilation, and checkpointing capabilities using PyTorch.

examples, examples/mnist-mlp · high confidence

The torchsystem package is introduced, providing a new domain model for building and compiling neural network aggregates. The library includes a Compiler class that manages a pipeline of functions to build and compile models, supporting dependency injection via a Provider mechanism. This allows for late binding and runtime behavior changes, facilitating complex initialization and distributed training scenarios.

torchsystem · high confidence

The torchsystem/registry module has been introduced to provide a centralized way to register PyTorch modules and capture their initialization arguments. This new feature allows users to decorate classes with @register to automatically store the arguments passed to \_\init\\_, as well as retrieve metadata such as the model's name, arguments, and a deterministic hash. The registry provides accessor functions (getname, getarguments, gethash, getmetadata) to query these stored properties, enabling easier model introspection and serialization workflows.

torchsystem/registry · high confidence

Added new files to the \torchsystem.domain\ package: \aggregate.py\ and \events.py\. The \Aggregate\ class provides a base for managing neural network components (models, optimizers, etc.) as a unified unit with lifecycle hooks (\onphase\, \epoch\) and phase management. The \Events\ class introduces a domain event system that allows enqueuing and committing domain events or exceptions, dispatching them to registered handlers. These changes establish the foundational domain logic for the library.

torchsystem/domain · high confidence

The \torchsystem.services\ package now exposes a new set of architectural building blocks for building domain-driven and event-driven applications. Users can create stateless business logic via the \Service\ class, which maps action names to handler functions with dependency injection. For messaging, the library provides a \Subscriber\ and \Publisher\ for topic-based pub/sub communication, as well as a \Consumer\ and \Producer\ for event-based processing. These components support runtime dependency overrides and automatic injection of dependencies, enabling flexible and testable service implementations.

torchsystem/services · high confidence

The CIFAR-ViT example now features a structured training and evaluation workflow. A new \Classifier\ class encapsulates the model, criterion, optimizer, and metrics. The \training\ module defines handlers for training and evaluation loops that dispatch events. A \compilation\ module manages model setup, device placement, and weight restoration. Additionally, \persistence\ handles saving model weights and logging metrics, while \metrics\ tracks loss and accuracy.

examples/cifar-vit · high confidence

Added comprehensive test coverage for core system components; Added tests for the torchsystem registry module.

The project's dependency graph has been updated via a new \poetry.lock\ file and an updated \pyproject.toml\. This includes adding \babel\ 2.16.0, \certifi\ 2024.12.14, \charset-normalizer\ 3.4.1, \click\ 8.1.8, \colorama\ 0.4.6, \ghp-import\ 2.1.0, and \griffe\ 1.5.5 to the documentation and test groups. The \pyproject.toml\ also specifies Python ^3.12 and updates test and documentation dependencies such as \pytest\, \mypy\, \mkdocs\, and \mkdocstrings\.

(dependencies) · medium confidence

Written by watchdog.canine.dev from the codebase's own history, inside the signed delivery this page is composed from.

• CAI 57 → 60 (+3.1)
• Rubric changed (rubric-2026.08.19 → rubric-2026.09.15) — scores are not directly comparable.

• Code Health 99 → 100 (+0.2)
• Architecture 100 → 99 (-1.1)
• Maturity 53 → 53 (+0.0)
• Readiness 50 → 55 (+5.2)
• Security 55 → 70 (+15.0)
• Domain Modelling 75 (new)

• Coverage not included — suite not readable by the collector
• Dependency hygiene not measured — dependency manifest found but not parsed for hygiene
• High CVE: [GHSA redacted] (poetry.lock)
• High: security finding (details withheld)
• High: security finding (details withheld)
• High: security finding (details withheld)
• High: security finding (details withheld)
• High: security finding (details withheld)
• LLM evaluation failed
• Medium CVE: [GHSA redacted] (poetry.lock)
• Medium CVE: [GHSA redacted] (poetry.lock)
• Medium CVE: [GHSA redacted] (poetry.lock)
• Medium CVE: [GHSA redacted] (poetry.lock)
• Medium CVE: [GHSA redacted] (poetry.lock)
• Medium CVE: [GHSA redacted] (poetry.lock)
• Medium CVE: PYSEC-2026-2132 (poetry.lock)
• Medium CVE: PYSEC-2026-2987 (poetry.lock)
• No exposed public API
• Test reliability not included

• Dependency hygiene PARTLY measured — Python dependencies read, no exact pin to grade for currency
• High CVE: [GHSA redacted] (poetry.lock)
• High CVE: [GHSA redacted] (poetry.lock)
• High: security finding (details withheld)
• High: security finding (details withheld)
• High: security finding (details withheld)
• High: security finding (details withheld)
• High: security finding (details withheld)
• High: security finding (details withheld)
• Medium CVE: [GHSA redacted] (poetry.lock)
• Medium CVE: [GHSA redacted] (poetry.lock)
• Medium CVE: [GHSA redacted] (poetry.lock)
• Medium CVE: [GHSA redacted] (poetry.lock)
• Medium CVE: [GHSA redacted] (poetry.lock)
• Medium CVE: [GHSA redacted] (poetry.lock)
• Medium CVE: PYSEC-2026-2132 (poetry.lock)
• Medium CVE: PYSEC-2026-2987 (poetry.lock)
• Medium: security finding (details withheld)
• No dependency advisory monitoring

Written by watchdog.canine.dev from the codebase's own history, inside the signed delivery this page is composed from.

entropy-flux/TorchSystem was measured the same way every project in this corpus was: the same rubric, at a pinned commit, with the result published in full. Point a surveyor at a repository you know and see whether you agree with it.