Packages
Provides NVIDIA GPU acceleration for JAX numerical computing via the PJRT plugin, enabling compiled array operations and automatic differentiation on CUDA 12 hardware.
FlashRank re-ranks search results using lightweight cross-encoder and LLM-based models, optimized for speed and minimal resource overhead in retrieval pipelines.
Install it if you're building retrieval-augmented generation systems and want to improve result quality without adding significant latency or infrastructure overhead.
Comet ML is an experiment tracking and logging platform that automatically captures scripts, output, code, and metrics from machine learning workflows and sends them to a cloud dashboard for centralized monitoring and comparison.
ai-edge-litert runs machine learning models on mobile and embedded devices with low latency and small binary footprint, supporting Android, iOS, and other operating systems.
Flyteidl provides the protobuf-based IDL (Interface Definition Language) and generated Python client code for Flyte's backend APIs, enabling programmatic interaction with Flyte's workflow orchestration platform.
Qwen3-ASR provides speech recognition and language identification for 52 languages and dialects, plus forced-alignment for timestamping speech in 11 languages, with both streaming and offline inference modes.
ONNX GraphSurgeon lets you programmatically create and modify ONNX neural network models by working with an intermediate representation of graphs, nodes, and tensors.
Install it if you need to create or modify ONNX graphs in code.
tf-nightly provides nightly builds of TensorFlow, an open-source framework for numerical computation and machine learning that runs on CPUs, GPUs, TPUs, and edge devices.
Compresses LLM prompts and agent outputs before they reach the model, reducing token usage by 60–95% on structured data while preserving accuracy, via library, proxy, CLI wrapping, or MCP server.
Python bindings for llama.cpp that enable running large language models locally with CPU or GPU acceleration, offering both low-level C API access and high-level text completion APIs compatible with OpenAI, LangChain, and LlamaIndex.
However, expect high install friction due to C compilation requirements—use pre-built wheels where available to reduce setup complexity.
LangMem provides memory management tools for LangGraph agents, enabling them to store, search, and recall information from conversations to maintain context and adapt behavior across sessions.
However, the aging maintenance status (291 days since last release) warrants checking that its dependencies remain compatible with your langchain and langgraph…
KeyBERT extracts keywords and keyphrases from documents by computing BERT embeddings and finding n-grams most similar to the document as a whole using cosine similarity.
Builds hierarchical multi-agent systems where a supervisor agent coordinates and delegates tasks to specialized worker agents using LangGraph.
However, review the library's own recommendation to use tool-calling directly for most new projects—install this primarily to upgrade existing code or when the…
A Python client library for interacting with Hindsight, a semantic memory system that uses personality-driven thinking to process and store information.
Integrates IBM watsonx.ai models with LangChain, providing chat, embedding, text generation, and reranking capabilities through a unified interface.
Implements rotary positional embeddings for transformer attention layers in PyTorch, enabling efficient encoding of absolute and relative position information along any tensor axis.
Provides archived experimental features for Haystack LLM framework that have either graduated into the main haystack-ai package or been discontinued; the final release (0.19.0.post1) is no longer maintained.
Identifies the language of text input across 97 languages using a pre-trained statistical model, available as a command-line tool, Python library, or WSGI web service.
However, do not use it if you require active maintenance, support for modern Python versions beyond basic compatibility, or confidence that the model reflects current…
Tests whether a graph is planar, computes planar embeddings, draws planar graphs as ASCII art, and isolates forbidden subgraphs using algorithms from the Edge Addition Planarity Suite.
fastai is a deep learning library that provides high-level APIs for training neural networks on images, text, tabular data, and recommendations, built on top of PyTorch.
Provides hardware-efficient implementations of linear attention, sparse attention, state space models, and hybrid sequence model architectures, verified across NVIDIA, AMD, and Intel platforms.
MXNet is a deep learning framework that enables you to build and train neural networks with support for multiple programming interfaces and hardware accelerators.
No, not for new projects.
sgl-deep-gemm provides optimized GEMM (matrix multiplication) kernels for GPU acceleration, packaged as a PyPI-installable wheel with ABI compatibility across Python versions via apache-tvm-ffi.
Provides PyTorch-based neural network models for image semantic segmentation, with 12 encoder-decoder architectures and 800+ pretrained encoders, usable in two lines of code.
Install it if you need to build or fine-tune semantic segmentation models; the breadth of pretrained encoders and multiple architectures make it a practical choice…
Python SDK for the Supervisely computer vision platform, providing programmatic access to data labeling, model management, project workflows, and app deployment capabilities.
However, clarify the license implications beforehand (treatment is currently unclear), and verify that the 53 dependencies align with your environment constraints.
Provides a PySpark-compatible implementation of the Hierarchical Navigable Small World (HNSW) algorithm for approximate nearest neighbor search with support for multiple distance metrics.
However, the aging maintenance status (no release since 2022-12-30) and unclear license warrant checking the repository before production use.
Faster-COCO-Eval provides a high-performance replacement for COCO metric evaluation that accelerates object detection, instance segmentation, and keypoint detection result computation through C++ optimizations while maintaining a compatible API.
Exports Hugging Face transformer models to ONNX format and runs them via ONNX Runtime for optimized inference.
Integrates LangChain with Milvus vector database to enable vector storage, similarity search, and retrieval for AI applications.
Install it if you are building LangChain applications that need vector storage and retrieval—it is the direct integration point between LangChain and Milvus.
Treelite serializes and exchanges decision tree forest models in a universal format, enabling C++ applications and other tools to store and transmit trained tree models efficiently.
BERTScore computes automatic evaluation metrics for text generation by comparing candidate and reference sentences using contextual embeddings from BERT, producing precision, recall, and F1 scores.
However, the package is dormant (last release 2023-02-20) and computationally heavy—it requires PyTorch and GPU access for practical use, and first-run model…
Feast is an open-source feature store that manages offline and online feature storage, retrieval, and materialization for machine learning training and real-time inference pipelines.
However, it brings substantial dependencies (30 runtime packages) and requires Python 3.10+.
Chronos provides pretrained transformer-based models for zero-shot time series forecasting, supporting univariate, multivariate, and covariate-informed tasks with quantile-based probabilistic predictions.
Install it if you need time series forecasting without the overhead of training your own model.
Qwen-Agent is a framework for building LLM applications with tool usage, planning, memory, and instruction-following capabilities, including pre-built assistants for browsing, code execution, and custom workflows.
However, verify the actual license terms (metadata shows unclear treatment despite Apache 2.0 headers), confirm Python version support for your environment, and…
Generates detailed summaries of PyTorch neural network models, showing layer names, input/output shapes, parameter counts, and computational operations—similar to Keras's model.summary() API.
Install it if you develop or analyze PyTorch models and want structured visibility into layer structure and computational cost.
FLAML automates machine learning workflows by selecting models and optimizing hyperparameters under resource constraints, supporting both standard ML tasks and custom function tuning.
Cache-DiT accelerates inference for Diffusion Transformers (DiTs) via hybrid caching, parallelism, quantization, and CPU offloading, integrating with the Diffusers library and supporting NVIDIA, AMD, and Ascend hardware.
Install it if you run Diffusers pipelines on GPU and want to reduce latency or memory usage.
sglang-kernel provides optimized CUDA compute primitives for LLM inference engines, enabling efficient inference for large language models and vision-language models through custom kernel operations.
InstructorEmbedding generates task-specific text embeddings by accepting natural-language instructions alongside text, without requiring model fine-tuning.
However, the lack of maintenance since May 2023 and unspecified Python version support mean you should test compatibility in your environment and be prepared to…
Sherpa-onnx runs speech recognition, text-to-speech, speaker identification, voice activity detection, keyword spotting, and other audio processing tasks locally using ONNX models across multiple platforms and architectures.
Install it if you need on-device speech or audio processing without external dependencies.