Packages
tf-nightly-cpu provides a CPU-optimized nightly build of TensorFlow for numerical computation, machine learning, and deep learning across CPUs, GPUs, TPUs, and edge devices.
NVIDIA NeMo Agent Toolkit is a framework-agnostic library for building, composing, and deploying enterprise agents with integrated tools, data sources, and observability.
Integrates the Milvus vector database with LlamaIndex for storing and retrieving vector embeddings in LLM applications.
Install only if you have Milvus available or plan to deploy it; this is a connector, not a standalone vector store.
PIQ provides a collection of image quality metrics and measures—both full-reference (comparing two images) and no-reference (assessing a single image)—implemented as PyTorch modules that can be used as loss functions or evaluated independently.
vllm-cpu provides CPU-optimized inference for large language models with automatic detection of available CPU instruction sets (AVX2, AVX-512, AMX on x86; NEON, BF16, DOTPROD on ARM), packaged as a single unified wheel.
However, be aware it is community-maintained (not official vLLM), has 63 runtime dependencies creating medium install friction, and performance will be significantly…
FiftyOne is a Python framework for building, visualizing, and evaluating computer vision datasets and models, with integrated labeling, model evaluation, and data quality tools.
Maintains an exponential moving average (EMA) copy of a PyTorch model during training, with configurable decay, update frequency, and optional post-hoc synthesis across multiple EMA versions.
Install it if you are training a PyTorch model and want to leverage weight averaging for better generalization or if you need post-hoc EMA synthesis for research.
Integrates Neo4j graph databases with LangChain, providing wrappers for querying graphs, storing chat history, managing vector stores, and building knowledge graphs from text using LLMs.
Install only if you have a Neo4j instance available and a use case that benefits from graph storage or reasoning.
robosuite is a MuJoCo-powered simulation framework for building, benchmarking, and training robot control policies through reinforcement learning and imitation learning.
Integrates Chroma vector database with LlamaIndex for storing and retrieving embeddings in RAG and semantic search applications.
AgentScope is a production-ready framework for building multi-agent systems with LLMs, providing abstractions for agent composition, tool management, model integration, and service deployment.
TorchX is a universal job launcher that submits PyTorch training and ML pipeline workloads to multiple schedulers including Kubernetes, Slurm, AWS Batch, Docker, and local execution.
Shimmy converts external reinforcement learning environments to Gymnasium and PettingZoo APIs, letting you use popular RL simulators with a unified interface.
Provides PyTorch implementations of librosa audio feature extraction functions, enabling GPU-accelerated computation of spectrograms, mel-spectrograms, and STFT/ISTFT transforms with numerical compatibility to the original librosa.
However, verify that PyTorch is available in your environment, and be aware that no updates have shipped since 2023-02-21—if you encounter bugs or incompatibilities…
Integrates local Ollama language models with LlamaIndex, enabling text completion, chat, streaming, and structured JSON responses through a unified Python interface.
Integrates LangChain and LangGraph with NVIDIA's NeMo Agent Toolkit to build agentic AI applications using multiple language model providers and retrieval-augmented generation.
However, verify whether this subpackage can be used independently or requires the parent toolkit, and whether the 20 bundled dependencies align with your actual needs.
Integrates AWS Bedrock's Converse API with LlamaIndex, enabling LLM applications to call foundation models like Claude through Bedrock with support for streaming, function calling, and prompt caching.
Install it if you're already using LlamaIndex and want to run Bedrock models; skip it if you're calling Bedrock directly or using a different LLM framework.
Converts machine learning models to LiteRT format for deployment on mobile and embedded devices with low latency and minimal binary size.
Provides shared utilities for the TabPFN tabular foundation model ecosystem, including regression result handling, data serialization between pandas/NumPy/CSV formats, dataset loading, and optional telemetry.
However, it is very new (first release September 2025) with minimal external adoption signals.
GenAgent provides utilities for building applications with large language models, including integrations with OpenAI and Anthropic, agent memory management, chat sessions, and structured output generation.
However, maintenance is aging (240 days since last release), the license treatment is unclear in the metadata, and you will need active API keys for OpenAI or Anthropic.
ncnn is a neural network inference framework that loads and runs pre-trained deep learning models on CPU and Vulkan GPU backends, with Python bindings for model execution.
Adds multi-agent delegation to Pydantic AI agents, letting you spawn and coordinate specialist subagents that run synchronously, asynchronously, or in auto-selected mode with built-in token tracking and task cancellation.
Tokamax provides custom accelerator kernels for JAX, including optimized implementations of attention, normalization, and mixture-of-experts operations for NVIDIA GPUs and Google TPUs, plus tooling to build and autotune custom kernels.
However, expect API changes and incomplete features—suitable for research and experimentation, less so for production systems requiring stability guarantees.
YOLOv5 is a packaged object detection model that runs inference on images and video to identify and localize objects, with integrated training, validation, export, and CLI support.
Integrates Astra DB (a serverless vector database built on Apache Cassandra) with LangChain, providing vector storage, chat history, caching, and document loading capabilities.
Install it if you are building LangChain applications on Astra DB or migrating away from deprecated community plugins.
ConfigSpace defines and manipulates configuration spaces for hyperparameter optimization and algorithm configuration, supporting continuous, discrete, and categorical parameters with sampling and constraint handling.
Install it if you are working on AutoML, hyperparameter tuning, or algorithm configuration tasks and need a structured way to define and sample parameter spaces.
Stirrup is a lightweight Python framework for building LLM-powered agents that can execute code, search the web, process files, and call external tools with minimal scaffolding.
Embeds imperceptible watermarks into audio files and detects them afterward, using neural network-based techniques that survive common audio transformations.
However, the aging maintenance status and lack of recent updates suggest you should verify that the neural network models and robustness claims meet your specific use…
AutoGluon Multimodal automates machine learning on image, text, and mixed-data tasks, training and deploying high-accuracy models with minimal code using deep learning and foundation models.
Pynini compiles grammar rules into weighted finite-state transducers and related automata structures, enabling programmatic construction and application of complex linguistic transformations.
Autonomous AI agent framework that executes plain-English tasks through a 5-role LLM pipeline, with access to 1,371 tools across 100 classes and support for 12 LLM providers including free NPMAI models, OpenAI, Groq, Anthropic, and Gemini.
Maintains a structured index of machine learning model metadata (parameters, results, training data, paper links) stored as YAML, JSON, or markdown annotations, with optional integration to Papers with Code.
Provides scalable datasets for software engineering research and development, integrating with cloud storage via Alibaba OSS and the Hugging Face datasets library.
Flair is a PyTorch-based NLP framework that applies pre-trained models for named entity recognition, sentiment analysis, part-of-speech tagging, and other sequence labeling tasks, plus text embedding and model training capabilities.
However, verify that the aging maintenance status (555 days since last release) does not conflict with your dependency update cycles, and test compatibility with your…
Python SDK for Baidu Qianfan large language model platform, enabling chat, text completion, embeddings, image generation, model training, and dataset management through a unified API.
However, note that maintenance is aging (553 days since last release), so verify that the current version supports your target models and features before committing…
LitData optimizes and streams large datasets for PyTorch training by chunking data into efficient formats and loading directly from cloud storage or local paths without full local downloads.
ColBERT is a BERT-based retrieval model that encodes passages and queries into token-level embedding matrices, then uses late interaction scoring to find relevant passages from large text collections in milliseconds.
However, the unclear license requires verification before commercial use, and the aging maintenance status (368 days since last release) means you should check…
pyjpt learns and reasons about joint probability distributions using tree-based partitioning, supporting both symbolic and subsymbolic variables in a single hybrid model with interpretable white-box inference.
pywhispercpp provides Python bindings for whisper.cpp, enabling speech-to-text transcription with support for multiple hardware accelerators and output formats.
Provides a real-time web dashboard for visualizing and monitoring Optuna hyperparameter optimization studies stored in a database backend.