Packages
YACS is a lightweight configuration management system that lets you define, serialize, and override experimental configurations using YAML files and Python code, designed for reproducible machine learning and scientific workflows.
Decompile and inspect the internal bytecode and optimization graphs generated by PyTorch's torch.compile, converting compiled code back to readable Python source and intermediate representations.
Install only if you have PyTorch>=2.2.0 available.
Integrates MLflow with Amazon SageMaker by signing requests with AWS IAM credentials and enabling model registration to the SageMaker Model Registry.
Evaluates mathematical expressions from LLM outputs by parsing LaTeX and plain expressions, converting them to a common symbolic form, and comparing them against gold-standard answers for equivalence.
kornia-rs provides low-level computer vision operations—image I/O, resizing, color conversion, and video capture—implemented in Rust with Python bindings for efficient, thread-safe processing.
OpenCLIP provides open-source implementations of CLIP (Contrastive Language-Image Pre-training) models that encode images and text into a shared embedding space for zero-shot classification, retrieval, and multimodal tasks.
A Python client library for calling ML models deployed on fal.ai, supporting both synchronous and asynchronous requests, file uploads, and status polling.
However, verify the license terms first—the metadata does not specify one—and confirm that fal.ai's pricing and availability align with your use case.
MTEB evaluates text and multimodal embeddings against a suite of standardized tasks and benchmarks, providing scores and rankings for embedding models and retrieval systems.
Install it if you need to benchmark embeddings, compare models, or contribute results to the public leaderboard.
SWE-ReX provides a runtime interface for AI agents to execute shell commands in sandboxed environments—local, Docker, AWS, or Modal—with a unified API that abstracts away infrastructure details.
Loads optimized compute kernels from Hugging Face Hub into Python applications at runtime, enabling dynamic kernel loading without modifying PYTHONPATH.
However, it requires Python 3.10+ and a working compute environment; without those, it will not function.
Transforms categorical variables into numeric representations using scikit-learn-compatible encoders, supporting both unsupervised methods (one-hot, binary, ordinal) and supervised techniques (target encoding, CatBoost encoding).
Install it if you work with categorical data in machine learning projects.
Connects LangChain applications to Ollama, enabling use of locally-run language models through a unified LangChain interface.
Install it if you want to use Ollama models within LangChain applications.
onnx-ir provides an in-memory intermediate representation for ONNX models that supports the full ONNX specification, enabling graph construction, analysis, and transformation without requiring protobuf after initial conversion.
Provides standard schema, statistics, and problem statement representations for machine learning metadata that can be used with TensorFlow for data validation, exploration, and transformation.
SeleniumBase is a browser automation framework that combines Selenium WebDriver with pytest integration, CDP mode for bot-detection bypass, and tools for web testing, scraping, and crawling.
However, the 60 runtime dependencies are substantial—evaluate whether you need the full feature set or if a lighter alternative suits your use case.
Comfy Kitchen provides optimized GPU kernels for quantized tensor operations in diffusion inference, supporting multiple compute backends (eager, CUDA, Triton, HIP) with operations like FP8/INT8/INT4 quantization, RoPE, AdaLN, and fused attention.
However, alpha status means the API and QuantizedTensor behavior may change; verify compatibility with your PyTorch and CUDA/ROCm versions before production use.
Treescope is an interactive HTML pretty-printer and tensor visualizer for IPython notebooks that renders complex objects with expandable trees, faceted array visualizations, and copy-path buttons.
Install it if you spend time inspecting complex objects or tensors in IPython notebooks.
Records and streams data like images, tensors, point clouds, and text to the Rerun Viewer for live visualization or file-based replay.
WebDataset reads and streams large-scale training data from tar-based shards using sequential I/O, compatible with PyTorch's IterableDataset and usable with PyTorch, TensorFlow, and JAX.
However, the aging maintenance status (421 days since last release) and alpha development status mean you should verify that the library's behavior matches your exact…
Provides standardized FastAPI-based handlers and decorators for deploying ML models (vLLM, TensorRT-LLM) to Amazon SageMaker with unified `/ping` and `/invocations` endpoints.
Evaluates and analyzes speaker diarization systems by computing metrics, diagnostics, and error analysis for reproducible assessment of who-spoke-when predictions.
However, verify the license terms first, as the metadata does not declare one explicitly.
RedisVL is a Python client for building AI applications with Redis vector search, enabling semantic search, RAG pipelines, and vector-backed retrieval with metadata filtering and hybrid search capabilities.
Tilelang is a domain-specific language for writing high-performance GPU and CPU kernels (GEMM, attention, sparse operations) using Pythonic syntax, compiling to optimized code via TVM.
However, note the Beta status, recent project age (11 months), and dependency on torch and TVM—verify stability for your specific use case and target hardware before…
Integrates OpenAI's multi-modal language models with LlamaIndex, enabling applications to process and reason over images and text together.
However, the aging maintenance status means you should verify that supported OpenAI models and LlamaIndex versions match your requirements before production use.
Semantic Kernel is a Python SDK for building AI agents and multi-agent systems that orchestrate language models, plugins, and tools through a unified framework supporting OpenAI, Azure OpenAI, and other LLM providers.
Install it if you're building AI agents, need multi-provider LLM support, or want a structured framework for prompt orchestration.
Integrates MLflow experiment tracking and model management with Azure Machine Learning workspaces, allowing you to log metrics and artifacts to AzureML while using MLflow APIs.
Integrates OpenAI-powered program synthesis with LlamaIndex, enabling structured code generation through OpenAI's language models.
However, the 441-day gap since the last release is a significant concern—verify that the three runtime dependencies are current and compatible with your OpenAI API…
Generates follow-up questions from documents or context using OpenAI's language models, integrated with the LlamaIndex framework.
However, the aging maintenance status (441 days since last release) warrants checking compatibility with your current llama-index versions before committing to…
Provides BlingFire tokenization bindings for the LiveKit Agents framework, enabling fast text segmentation and linguistic analysis within voice agent applications.
Integrates Deepgram's voice AI services (speech-to-text and text-to-speech) into LiveKit Agents for real-time audio processing in agent applications.
Install it if you are building LiveKit Agents that need Deepgram's speech-to-text or text-to-speech services.
Provides experimental LangChain components and research prototypes for building LLM applications, intended for exploration rather than production use.
No, not for production.
MediaPipe provides pre-built machine learning models and cross-platform APIs for vision, text, and audio tasks that run on-device without sending input data to external servers.
DVC is a command-line tool for versioning data and models alongside code, running reproducible machine learning pipelines, and tracking experiments locally using Git.
TensorDict is a batched, nested dictionary container that behaves like a PyTorch tensor, allowing you to slice, reshape, move, and perform arithmetic on structured data while keeping all nested tensors synchronized.
Official Python SDK for IBM watsonx.ai that provides a unified interface to foundation models, AutoAI experiments, retrieval-augmented generation, model tuning, and deployment across the watsonx.ai platform.
Install it if you are building or deploying AI models on the watsonx.ai platform; skip it if you are not using that specific IBM service.
Provides NVIDIA CUBLAS native runtime libraries for CUDA 11, enabling GPU-accelerated linear algebra operations in Python applications on x86_64 and ARM64 Linux, Windows platforms.
Install only if you are maintaining legacy code explicitly pinned to CUDA 11 and cannot upgrade; otherwise, use current nvidia-cublas packages or let your framework…
Client library for the Runware inference API, enabling image generation, upscaling, background removal, and other AI-powered image operations via async Python.
Converts trained scikit-learn models to ONNX format for deployment and high-performance inference using ONNX Runtime or other compatible tools.
Detects end-of-turn in voice conversations for LiveKit Agents using a language model trained for this task, replacing simpler voice activity detection with more accurate interruption prevention.
Provides the NVIDIA CUDA nvcc compiler for building CUDA applications from Python, bundling the compiler with its runtime dependencies.