Packages
IceVision provides a unified framework for training and deploying object detection models, supporting multiple model architectures and training backends like PyTorch Lightning and Fastai.
Install only if you are maintaining legacy code already using IceVision and cannot migrate.
Provides embedding generation through Pinecone's Inference API, enabling you to create vector embeddings for documents and queries using supported embedding models.
However, the dormant maintenance status means you should verify that the current version and supported models meet your needs, as no updates are expected.
AQT provides quantization of tensor operations (matmul, einsum, conv) in JAX computations, enabling int8 and lower-bit model training with minimal configuration while maintaining training-serving consistency.
Semantic Router routes LLM requests to predefined decision paths using vector embeddings instead of waiting for slow LLM generations, enabling fast semantic-based intent classification.
OpenEnv provides a Gymnasium-style framework for creating and deploying isolated execution environments that agents can interact with via standard `reset()`, `step()`, and `state()` APIs over HTTP and WebSocket.
Not recommended for production systems requiring API stability.
Provides C++ implementations of multiple tokenizers (SentencePiece, Tiktoken, Huggingface, Llama2.c, Tekken) with Python bindings for use in PyTorch-based LLM applications and ExecuTorch runners.
GPTCache provides semantic caching for LLM API calls, storing and retrieving responses based on exact or semantic similarity matching to reduce API costs and latency.
However, be aware that the project is aging (last release August 2024, no commits since July 2025), two security vulnerabilities are known, and the API is explicitly…
Evaluates CLIP-like vision-language models on standard datasets using zero-shot classification, zero-shot retrieval, linear probing, and captioning tasks.
However, the aging maintenance status (385 days since release) and Pre-Alpha development status mean you should verify compatibility with your PyTorch and…
ExecuTorch exports and runs PyTorch models on mobile, embedded, and edge devices with support for hardware acceleration backends like XNNPACK, Core ML, and QNN.
Python SDK for connecting to and running computer vision models and workflows on a local or remote Inference server, enabling image and video stream processing with object detection, classification, segmentation, and custom model inference.
Install it if you need to programmatically interact with Inference from Python; if you only need the server itself, install inference-cli instead.
dm_control provides Python bindings to the MuJoCo physics engine and a suite of reinforcement learning environments for continuous control tasks, along with tools for composing and modifying physics-based simulations.
Install it if you need a physics simulation framework for continuous control research or prototyping.
PyBullet provides physics simulation, robotics modeling, and collision detection through Python bindings to the Bullet Physics SDK, supporting URDF/SDF file loading, kinematics, dynamics, and rendering.
Integrates Groq's fast inference API with LiveKit Agents, enabling real-time LLM responses for voice and conversational agents.
Install it if you are already using LiveKit Agents and want to use Groq as your LLM provider, or if you are evaluating Groq for real-time voice agent workloads.
Builds graph retrieval-augmented generation (GraphRAG) applications by integrating Neo4j graph databases with LLM-powered knowledge extraction and retrieval pipelines.
YDF trains, evaluates, and interprets decision forest models (Random Forests, Gradient Boosted Trees, and others) through a Python wrapper around the C++ Yggdrasil Decision Forests library.
Provides experimental reinforcement learning algorithms and tools that extend stable_baselines3 with newer implementations, including policy gradient variants, Q-learning methods, and environment wrappers.
Identifies the language of text in one of 97 languages using a pre-trained statistical model, with optimized speed for Python 3.6+.
ClearML Agent is a job scheduler and orchestration service that runs machine learning experiments on local or cloud resources, managing virtual environments, dependencies, and execution monitoring across Linux, macOS, and Windows.
Provides a thin wrapper to use OpenAI-compatible APIs (including local models) within the LlamaIndex framework, allowing you to swap in alternative LLM providers that follow the OpenAI protocol.
PromptLayer is a Python SDK for managing, testing, and monitoring LLM prompts and agent requests with built-in tracing, template retrieval, and provider proxy support.
GPU-accelerated machine learning algorithms for classification, regression, clustering, and dimensionality reduction, built on CUDA and integrated with scikit-learn-compatible APIs.
Computes Short Term Objective Intelligibility (STOI) as a PyTorch loss function for training speech enhancement models, wrapping the classical and extended STOI metrics in differentiable form.
However, the dormant maintenance status (no releases for 683 days) and the documented approximation—not exact replication—of the original metric mean you should…
Integrates Amazon AWS AI services (Bedrock, Polly, Transcribe, Nova) into LiveKit Agents for speech-to-speech, text-to-speech, speech-to-text, and LLM capabilities.
Install it if you are building LiveKit voice agents and need AWS AI services (Bedrock, Transcribe, Polly, Nova).
Fits beta calibration models to classifier outputs, transforming predicted probabilities into better-calibrated confidence estimates.
However, verify compatibility with your numpy and scikit-learn versions before relying on it in production, given the last release was April 2021.
Vocos is a neural vocoder that synthesizes audio waveforms from acoustic features (mel-spectrograms or EnCodec tokens) using a GAN-based model that generates spectral coefficients for fast reconstruction via inverse Fourier transform.
Provides NVIDIA GPU support for JAX by enabling CUDA 13 compilation and execution on compatible GPUs through the jax-cuda13-pjrt runtime.
PyTorch Ignite provides a high-level training and evaluation engine that simplifies neural network workflows with an event-driven architecture and built-in metrics, reducing boilerplate while maintaining full control over training logic.
NIXL provides Python bindings to accelerate point-to-point communications in AI inference frameworks, abstracting over CPU/GPU memory and various storage backends through a modular plugin architecture.
Integrates Anthropic's Claude language models into LlamaIndex applications, enabling text completion, chat, streaming, and structured prediction workflows.
Install it if you are building LlamaIndex applications and want to use Claude models; skip it if you are using Anthropic's SDK directly without LlamaIndex.
Lime explains individual predictions from any black-box machine learning classifier by fitting local linear models around specific instances, supporting text, tabular, and image data.
Install it if you need model-agnostic explanations for existing classifiers and can verify it works with your specific dependencies; avoid it if you require active…
Recognizes text and detects objects in captcha images using offline neural network models, supporting digits, letters, Chinese characters, and slider captchas.
Applies graph-level optimizations to ONNX models, including a library of prepackaged passes for common transformations like operator fusion and constant elimination.
InterpretML provides interpretable machine learning models and explainers for blackbox systems, letting you understand both global model behavior and individual predictions.
Adds programmable guardrails to LLM-based conversational applications to control output behavior, prevent unsafe responses, and enforce structured dialog flows.
LangExtract uses LLMs to extract and ground structured information from unstructured text, mapping each extraction to its exact source location and generating interactive visualizations of results.
Install it if you need to extract and ground structured data from documents with traceability and schema enforcement.
Computes Fréchet Inception Distance (FID) and Kernel Inception Distance (KID) scores for evaluating generative models, with corrected image resizing and quantization to ensure consistent results across implementations.
However, consider that maintenance is aging (last release December 2022); verify compatibility with your current PyTorch version before relying on it for new…
Integrates Google's Gemini and Vertex AI embedding models with llama-index for generating text embeddings in your applications.
Wraps OpenTelemetry with Arize-aware defaults to instrument and send LLM application traces to Arize or custom OTLP endpoints.
Install it if you're already using Arize or a compatible OTLP collector and want to avoid writing OTEL configuration boilerplate; skip it if you need raw…
Calls Azure's Computer Vision service to analyze images and extract visual features like captions, text (OCR), and object detection in a single API call.
OpenViking is a context database for AI agents that organizes memories, resources, and skills as a virtual filesystem under the `viking://` protocol, with tiered content loading (abstract, overview, details) and observable retrieval trajectories.
However, the AGPL-3.0 license requires copyleft compliance in derivative works, and the 58 runtime dependencies create medium install friction.