Research, software, data and methods. Inspect the context, compare your options, and take a useful next step.
Upstream project repository
Parameterization and execution of notebooks for repeatable computational workflows.
Investigate Does an isolated parameterized notebook run preserve supplied parameters and produce the independently expected deterministic output?
Upstream project repository
Parameter-efficient model adaptation methods for testing trainable-parameter budgets and adapter composition.
Investigate Does a low-rank adapter preserve the frozen base parameters and produce the same outputs before and after an explicitly supported merge?
Upstream project repository
Reading, modifying, and writing DICOM medical-imaging metadata and datasets in Python.
Investigate Which explicitly selected metadata elements survive a synthetic DICOM write/read round trip?
Upstream project repository
Tensor expression graphs, automatic differentiation, and compilation for numerical computation.
Investigate Do automatic gradients match analytic derivatives across a fixed grid of nonsingular inputs?
Upstream project repository
Tensor computation and automatic differentiation framework for constructing controlled learning and gradient experiments.
Investigate Do automatic gradients agree with analytical derivatives for a small well-conditioned scalar objective?
Upstream project repository
Training-loop framework for studying checkpoint restoration, reproducibility and separation of model logic from orchestration.
Investigate Does resuming a tiny deterministic run from a checkpoint match its uninterrupted reference?
Upstream project repository
Distributed execution and machine-learning infrastructure for studying task scheduling, tuning and fault handling.
Investigate Does a small synthetic parameter sweep retain each trial's identity and failure state under a declared retry policy?
Upstream project repository
Experiment organization and observation tooling for capturing configurations, dependencies and run outcomes.
Investigate Does a controlled failed run preserve its configuration and failure outcome as clearly as a successful run?
Upstream project repository
Sparse-autoencoder research tooling for examining reconstruction, sparsity and feature interventions on neural activations.
Investigate How does a fixed sparsity setting trade reconstruction error against active features on an authored low-dimensional activation set?
Upstream project repository
Tensor serialization tooling for studying shape, dtype and byte-level preservation without a pickle-style object format.
Investigate Do small authored tensors round-trip with exact dtype, shape and payload bytes through the selected serializer?
Upstream project repository
Model-serving and structured-generation tooling for evaluating runtime behavior on controlled language-model workloads.
Investigate Do constrained outputs satisfy a prespecified small grammar under the selected serving backend?
Executable Science local pilot drafts
Provide Dijkstra and Bellman–Ford implementations with hand-enumerated reference graphs, explicit negative-edge rejection, and a negative-cycle control.
Investigate Do the implementations respect their input assumptions and return the expected distances on the supplied controls?
Recorded license evidence is scoped to each source; it is not a blanket permission or a verification result.