Packages
MONAI is a PyTorch-based framework for building deep learning models on medical imaging data, providing pre-processing, network architectures, loss functions, and evaluation metrics tailored to healthcare applications.
Automates feature engineering and preprocessing for machine learning pipelines, integrating with AutoGluon's broader ML automation framework to handle tabular, time series, and multimodal data preparation.
QuDiDA performs pixel-level image domain adaptation using scikit-learn transformers, allowing you to adjust image style or appearance to match a reference image.
LLM is a CLI tool and Python library for running prompts against Large Language Models from OpenAI, Anthropic, Google, Meta, and dozens of other providers, both via remote APIs and locally-installed models.
Integrates a Microsoft Bot Framework bot into an aiohttp web application, enabling asynchronous HTTP request handling for bot interactions.
Install only if you are maintaining an existing bot deployment or do not require ongoing support.
A Python client library for programmatically interacting with the DataRobot platform API, enabling model building, deployment, and management workflows.
e3nn provides PyTorch-based operations for building E(3)-equivariant neural networks, including tensor products, spherical harmonics, and linear layers that respect rotation, translation, and mirror symmetries.
Embedchain is a framework for building retrieval-augmented generation (RAG) applications that personalize LLM responses by ingesting, chunking, embedding, and retrieving unstructured data from various sources.
However, be aware that the last release was 507 days ago—expect slower bug fixes and feature updates.
PyTorch implementation of EfficientDet object detection models with configurable BiFPN, multiple backbones from timm, and support for training on COCO and other datasets.
BERTopic performs topic modeling on text documents using transformer embeddings and c-TF-IDF clustering to identify and label coherent topics with interpretable keywords.
Detects voiced versus unvoiced segments in audio by wrapping Google's WebRTC Voice Activity Detector with pre-built binary wheels for Windows, macOS, and Linux.
Install it if you need reliable voice activity detection in Python.
sherpa-onnx-core provides pre-built ONNX runtime binaries for speech recognition, synthesis, speaker identification, audio tagging, and other audio processing tasks across multiple platforms and architectures.
CopilotKit is a Python SDK for building AI copilots and agents that integrate with LangGraph, LangChain, and FastAPI to add conversational AI capabilities to applications.
Install it if you are building a FastAPI or LangChain application and want a higher-level agent framework; skip it if you prefer lower-level control over your agent…
RF-DETR is a transformer-based object detection, instance segmentation, and keypoint detection library built on a DINOv2 vision transformer backbone, designed for real-time inference.
Lineax solves linear systems and least-squares problems in JAX, handling both explicit matrices and implicit linear operators without materializing them.
Provides LangChain-compatible chat models and embeddings that route requests through UiPath's LLM infrastructure, supporting multiple providers including OpenAI, Google, Anthropic, AWS Bedrock, Azure AI, and Fireworks.
However, verify the license treatment before production use, and confirm that the normalized API mode meets your feature and stability requirements if you plan to use…
TinySegmenter is a compact Japanese tokenizer that breaks Japanese text into morphological tokens without requiring external dictionaries or machine learning models.
Provides PyTorch implementations of the Conformer architecture—a convolutional module and full transformer blocks that add local inductive bias to transformer models, commonly used in speech recognition.
However, if you require ongoing maintenance, compatibility updates, or active support, look for an actively maintained alternative or be prepared to maintain a fork…
Integrates Azure OpenAI's embedding models with LlamaIndex, enabling vector representations of text for semantic search and retrieval-augmented generation workflows.
Prompty is a Python runtime for executing LLM prompts defined in a language-agnostic `.prompty` file format, with built-in support for Azure OpenAI, OpenAI, and serverless model invokers, plus tracing and CLI tooling.
However, verify the status of the two known vulnerabilities (GHSA-wxhm-2mq7-7697, PYSEC-2026-3538) before deploying to production, and confirm whether optional…
Converts non-negative integers to their word representations in Indian languages and English, with optional digit-by-digit or variation modes.
Orbax Export serializes JAX models to TensorFlow SavedModel format, enabling JAX-trained models to be deployed and used in TensorFlow environments.
Kosong is an LLM abstraction layer that unifies message structures and asynchronous tool orchestration across multiple chat providers, letting you build AI agents without vendor lock-in.
However, the package is very new (first release 2025-10-15) and has no recorded license metadata—verify the license terms before using in proprietary work, and expect…
nixl is a meta package that automatically selects and installs the correct CUDA 12 or CUDA 13 backend for PyTorch at runtime based on your system's CUDA version.
FunASR is a speech recognition toolkit that transcribes audio offline or via streaming, with integrated voice activity detection, speaker identification, punctuation restoration, and emotion/audio-event tagging across multiple languages and deployment targets.
Install it if you need speaker diarization, emotion detection, streaming support, or self-hosted deployment.
Applies state-of-the-art model optimization techniques—quantization, pruning, Neural Architecture Search, distillation, speculative decoding, and sparsity—to accelerate PyTorch, Hugging Face, and ONNX models for deployment.
NumKong provides mixed-precision linear algebra and distance kernels with automatic accumulator widening, GIL-free batched operations, and low-precision dtype support (BFloat16, Float8, Float6, packed bits) across x86, ARM, RISC-V, and other architectures.
mlforecast trains machine learning models on time series data and generates forecasts, with built-in support for feature engineering, multiple models, and distributed training across remote clusters.
Instrument LLM applications with event tracking and OpenTelemetry-based tracing, capturing AI interactions, user signals, and tool calls for observability and analysis.
baml-py is the Python runtime for executing BAML programs, a language designed for building agents with strong typing and error handling.
However, verify the license terms first (currently unclear) and confirm that BAML's compilation tooling fits your workflow—baml-py is the runtime only.
A fast, Rust-backed tokenizer library that encodes and decodes text using BPE, WordPiece, and Unigram algorithms, compatible with HuggingFace tokenizer models.
USearch provides approximate nearest-neighbor vector search using HNSW indexing, supporting custom metrics, multiple data types, and on-disk index serving without loading into RAM.
Integrates Ollama LLM provider support into Microsoft Agent Framework, enabling agents to run local or remote Ollama models as their reasoning backbone.
Provides optimized GPU implementations of scaled dot-product attention (FlashAttention and FlashAttention-2) that reduce memory usage and improve speed compared to standard attention, with support for NVIDIA CUDA and AMD ROCm.
PyCaret automates machine learning workflows by wrapping scikit-learn, LightGBM, and related libraries, letting you build classification, regression, clustering, time-series, and anomaly-detection models with minimal code.
agnoctl is a command-line interface for creating and managing AgentOS projects, with built-in support for nine deployment platforms including Docker, AWS, Azure, Fly, GCP, Helm, Modal, Railway, and Render.
Provides a Python interface to Google's WebRTC Voice Activity Detector, classifying audio frames as voiced or unvoiced for speech recognition and telephony applications.
However, high install friction (compiled extension), dormancy since 2017-01-07, and uncertainty about modern Python compatibility mean you should verify it builds on…
Provides RAG (Retrieval Augmented Generation) context providers for Azure AI Search, supporting both semantic hybrid search and agentic multi-hop reasoning modes within the Microsoft Agent Framework.
Converts chat messages to token IDs for LLM inference and training, with deterministic parsing of model responses and multi-turn conversation extension that preserves token identity across turns.
Qwix is a JAX quantization library that applies Quantization-Aware Training (QAT) and Post-Training Quantization (PTQ) to neural network models, supporting deployment on XLA devices (CPU/GPU/TPU) and LiteRT targets.