Packages
llmcompressor optimizes large language models for efficient deployment by applying quantization, pruning, and other compression techniques compatible with vLLM inference.
Provides a high-performance Rust implementation of the Leiden community detection algorithm for graph analysis, exposed to Python via a native extension module.
Generates pseudorandom numbers using algebraic methods.
Provides numba-optimized window operations (rolling, expanding, seasonal) on numpy arrays as a faster alternative to pandas rolling and expanding methods, plus online classes for incremental updates.
Not recommended if you need active development, broad window function coverage, or integration with pandas DataFrames.
Provides basic mathematical statistics functions (mean, median, mode, standard deviation, variance) for Python 2.x, backported from Python 3.4's standard library statistics module.
Install only if you are maintaining legacy code on those versions that cannot be upgraded; otherwise, use Python 3's built-in statistics module.
MNN is a lightweight deep learning inference and training framework that runs neural network models efficiently on mobile devices, embedded systems, and servers with support for multiple hardware backends including CPU, GPU, and NPU.
Megatron Core provides GPU-optimized building blocks and parallelism strategies for training large transformer models at scale, including tensor parallelism, pipeline parallelism, and mixed precision support.
Not recommended for simple single-GPU training or inference-only use cases.
Packs a set of rectangles with fixed orientations into a minimal-area bounding box with no overlaps, returning the (x, y) coordinates of each rectangle's lower-left corner.
Install it if you need to pack rectangles efficiently.
A wrapper around scipy's optimize.minimize and optimize.root that adds progress bars, early stopping, flattened keyword arguments, and a unified callback API across different optimization methods.
TensorFlow Recommenders provides a Keras-based library for building recommender system models, covering the full workflow from data preparation through training, evaluation, and deployment.
Implements the TrueSkill rating algorithm to compute player skill ratings and match quality in competitive games.
Intel-optimized TensorFlow for Windows that accelerates numerical computation and machine learning workloads using oneDNN primitives for Intel architecture.
Install only if you are on Windows; the medium dependency footprint is standard for TensorFlow and the oneDNN environment variable must be set explicitly to activate…
Provides 2D discrete wavelet and dual-tree complex wavelet transforms (DWT and DTCWT) with their inverses in PyTorch, supporting gradient backpropagation and batched multichannel image processing.
However, verify that the 'Free To Use' license suits your use case, test compatibility with your PyTorch and NumPy versions, and be prepared to maintain or fork the…
Parses ULog files (a self-describing logging format) and provides command-line tools to extract, convert, and display logged data in multiple formats including CSV, KML, and ROS2 bag files.
Install it if you work with PX4 logs or need ULog parsing; skip it otherwise.
AntroPy computes entropy and fractal dimension measures from time-series data, optimized for speed via Numba JIT compilation and designed to work on both 1-D and N-D arrays.
Fxpmath provides fixed-point arithmetic with arbitrary word and fractional sizes, supporting signed and unsigned numbers, numpy arrays, and bitwise operations for DSP and FPGA applications.
Computes Hartigan's dip test statistic and p-value to detect multimodality in a sample, measuring how far an empirical distribution deviates from unimodality.
Flower is a framework for building federated learning systems where machine learning models are trained collaboratively across distributed clients without centralizing raw data.
Install it if you need to build federated learning systems or conduct distributed model training while preserving data privacy.
TensorFlow Transform preprocesses data with full-pass operations like normalization, vocabulary generation, and bucketing, exporting a reusable TensorFlow graph for consistent training and serving.
Openskill is a multiplayer rating system that calculates and updates player skill ratings in competitive games, supporting multiple teams, asymmetric team sizes, and various outcome types (win, draw, rank).
Adds type annotations for PyTorch tensor shape, dtype, layout, and dimension names with optional runtime checking via typeguard.
However, be aware that the package is aging and the author recommends jaxtyping for new projects due to better static type checker support.
PyMC-Marketing provides Bayesian statistical models for marketing analytics, including Marketing Mix Modeling (MMM), Customer Lifetime Value (CLV), Customer Choice, Bass Diffusion, and Predicted Incrementality by Experimentation (PIE).
LiteRT is a runtime for running trained machine learning models on mobile and embedded devices with low latency and minimal binary footprint.
However, avoid it for production deployments requiring stability—use a stable release instead.
Constructs isolines and isosurfaces of 2D and 3D scalar fields defined by functions, using a quadtree-based approach to efficiently sample only regions near the implicit surface.
However, note that maintenance is aging (last update February 2024, no commits in over 900 days)—suitable for stable use cases, but not for projects requiring active…
Theano-PyMC is an optimizing compiler for mathematical expressions on multi-dimensional arrays, enabling GPU acceleration and automatic differentiation for numerical computations.
PyOpenCL provides Python bindings to OpenCL, enabling you to write GPU and massively parallel compute code that runs on diverse hardware accelerators through a Pythonic interface.
A library of ready-to-use STEM equations implemented as functions, covering algebra, geometry, and statistics to avoid reimplementing common mathematical formulas.
Not recommended if you need a comprehensive math library.
Quantizes LiteRT models to reduce size and improve inference performance on edge devices, supporting dynamic, weight-only, and static quantization strategies with configurable recipes.
Theano is a Python library for defining, optimizing, and evaluating mathematical expressions on CPUs and GPUs, with symbolic differentiation and tight NumPy integration.
Install only if required to run existing legacy code, and expect significant friction on modern systems.
Converts PyTorch models to .tflite format for on-device deployment on Android, iOS, and IoT devices via LiteRT, with CPU coverage and initial GPU and NPU support.
Lightly provides self-supervised learning models and loss functions for computer vision, enabling you to train neural networks on unlabeled image data using methods like MoCo, SimCLR, BYOL, and DINO.
Install it if you need to train on unlabeled image data or experiment with SSL methods.
Solves sparse linear systems in JAX using the KLU algorithm from SuiteSparse, optimized for CPU arrays with float64 or complex128 precision.
torchax is a PyTorch backend that runs PyTorch code on Google Cloud TPUs via JAX, with bidirectional interoperability allowing you to call JAX functions from PyTorch and vice versa.
However, it is in alpha (Development Status :: 3), so expect incomplete operation coverage and potential API changes.
Computes financial return and risk metrics—including drawdown, alpha/beta, Value at Risk, Sharpe and Sortino ratios—from return arrays or pandas Series, with support for rolling window calculations and optional data fetching from Yahoo Finance or Fama-French sources.
Automates IBM Quantum backend connection, authentication, and selection by detecting your subscription plan and returning the least-busy available QPU resource for immediate use in Qiskit circuits.
However, note the aging maintenance status (219 days since last release) and verify that the connector's error handling and backend-selection logic align with your…
tfx_bsl provides shared libraries and utilities used internally by TensorFlow eXtended (TFX) components and standalone TFX libraries like TFDV, TFMA, and TFT.
TensorFlow CPU for AWS is a machine learning framework optimized for numerical computation on CPU-based AWS infrastructure, supporting deployment across diverse platforms and devices.
However, verify that this AWS-specific variant offers advantages over the standard TensorFlow CPU distribution, and check whether version 2.15.1 (released 2024-03-14)…
Provides utility functions and helpers for the Lightly self-supervised learning framework, supporting computer vision model pretraining with numpy and Pillow.
However, verify that this utility package is still the recommended approach for your Lightly version, as the last release was in 2021.
Performs shape-preserving piecewise rational cubic interpolation using the Delbourgo-Gregory algorithm, maintaining monotonicity and convexity of interpolated curves.
However, it is aging (no updates since initial release) and has minimal adoption; consider it only if standard cubic splines or other interpolation libraries do not…
Provides high-precision implementations of error functions (erf, erfc, erfcx) and normal distribution functions (PDF, CDF, inverse CDF) using W.J. Cody's rational Chebyshev approximations.