{"categories":[{"label":"Artificial Intelligence","url":"https://skillfed.io/packages/category/scientific-engineering-artificial-intelligence/9"}],"enrichment":{"capability":"Provides reproducible random number generation for PyTorch by seeding Python, NumPy, and all CUDA RNGs together, plus context managers to isolate or save RNG state.","skillfed_tags":["reproducibility","rng-seeding"],"use_cases":["Seed all RNGs at the start of a training script to ensure reproducible model initialization and data shuffling across runs.","Use SavedRNG context manager to run stochastic operations (e.g., data augmentation) without affecting the main training RNG stream.","Maintain separate RNG streams for different parts of an experiment (e.g., one for data generation, one for model dropout) that evolve independently.","Verify that two seeded runs produce identical random sequences to validate experiment reproducibility."],"what_it_does":"pytorch-seed is a lightweight utility for reproducible random number generation in PyTorch projects. It provides a single `seed()` function that simultaneously seeds Python's base RNG, NumPy's RNG, PyTorch's CPU RNG, and all CUDA RNGs\u2014ensuring deterministic behavior across the entire stack. It also offers a `SavedRNG` context manager that lets you temporarily isolate RNG state, either to run code without affecting the global RNG or to maintain independent RNG streams that resume from where they left off.\n\nThe package is designed for researchers and practitioners who need reproducible training runs or experiments. Its API mirrors PyTorch Lightning's seeding utilities, making it familiar to users of that ecosystem. Installation is straightforward with only torch and numpy as dependencies, but note that the project is abandoned as of 2023-02-15 with no active maintenance.","worth_installing":"Yes, if you need simple, unified RNG seeding for PyTorch and are comfortable with an abandoned package. The code is straightforward and unlikely to break with minor PyTorch updates, but you should test it with your specific PyTorch version and consider forking if you need future fixes. For active projects requiring long-term support, evaluate whether PyTorch Lightning's seed utilities or manual seeding better fits your needs."},"id":"pytorch-seed","links":{"html":"https://skillfed.io/packages/pytorch-seed","md":"https://skillfed.io/packages/pytorch-seed.md","pypi":"https://pypi.org/project/pytorch-seed/"},"maintenance":{"status":"abandoned"},"meta":{"latest_release":"2023-02-15","license_spdx":null,"license_treatment":"permissive","name":"pytorch-seed","python_support":"supports_current","summary":"RNG seeding and context management for pytorch"},"popularity":{"monthly_downloads":124120,"position":11885,"tier":"top_15000"},"security":{"n_vulnerabilities":0},"version":"0.2.0"}