Packages
Rewrites attributes and constants in ONNX model operators via CLI or Python API, supporting both file-based and in-memory graph modifications.
Converts ONNX model input tensor dimensions and channel ordering (NCHW ↔ NHWC, RGB ↔ BGR) by rewriting input operations and inserting compensating transposes.
Integrates Modal serverless compute sandboxes with LangChain's Deep Agents framework, allowing agents to execute commands in isolated Modal environments.
However, it is a narrow integration—only install if you specifically need Modal sandbox execution within agents; it adds no value if you're not using both Modal and…
Converts JSON files to ONNX model format, accepting either a JSON file path or a dictionary and returning an ONNX ModelProto object.
However, dormant maintenance (last update 1077 days ago) means you should test thoroughly before production use and be prepared to fork or find alternatives if issues…
Compresses ONNX model files by deduplicating constant tensor values throughout the graph, with options to extract weights to external .npy files or downcast numeric types for further size reduction.
However, the project is abandoned and has not been maintained since September 2022, so test thoroughly with your specific ONNX models and toolchain versions before…
An MCP server that connects AI tools to Amazon Bedrock Knowledge Bases, enabling natural-language queries, result filtering by data source, and optional reranking of retrieved passages.
Install only if you have AWS credentials, appropriate IAM permissions, and at least one tagged knowledge base; otherwise, setup will fail at runtime.
Prompt flow is a development suite for building, testing, and deploying LLM-based applications, integrating prompts, Python code, and external tools into executable flows with built-in tracing and evaluation.
Install it if you are building or managing LLM-based applications and want structured tooling for testing, evaluation, and deployment.
Analyzes and displays the structure of ONNX models, including those too large for graphical tools like Netron, reporting operation counts and model size.
However, dormant maintenance (last update over 1055 days ago) means you should verify compatibility with your ONNX opset version before relying on it in production…
Adapters adds parameter-efficient fine-tuning methods to HuggingFace Transformers models, letting you train and compose multiple adapter modules on top of pre-trained models with minimal parameter overhead.
Install it if you're doing transfer learning on transformer models and want to reduce training costs and memory usage.
Modifies the input and output shapes of ONNX model files, allowing you to adapt pre-trained models to different batch sizes, dimensions, or symbolic shape parameters via CLI or Python API.
Modifies ONNX model files to initialize or replace their batch size dimension with a specified value, useful for making models accept variable or fixed batch sizes.
However, maintenance is dormant (last update 808 days ago), so expect no active bug fixes or feature development.
Infers and fills missing tensor shape information in ONNX model files, available as both a CLI tool and a Python library function.
However, note that the project is dormant (last update 808 days ago), so expect no active maintenance or updates for recent ONNX features.
Adds new operations (nodes) to ONNX model graphs at specified connection points, allowing insertion of custom operators between existing model layers.
Applies specialized optimization techniques to ONNX neural network models, particularly those with Einsum and OneHot operations that standard optimizers cannot effectively reduce.
However, the dormant maintenance status means no active support for new onnxsim versions or bug fixes—treat it as a specialized tool for a specific problem, not a…
Changes the opset version of an ONNX model graph, either via command-line or Python API, without altering the model's structure or weights.
Encodes and decodes Base64-formatted constant values for ONNX model files, supporting multiple numeric and string data types via CLI or Python API.
However, if you need ongoing support, compatibility updates, or are starting a new ONNX workflow, consider whether an actively maintained alternative exists or…
sod4onnx removes specified output operations from ONNX model files, either via command-line or Python API, producing a modified ONNX model with those outputs deleted.
Removes embedded doc_strings from ONNX model files, reducing file size and metadata bloat in serialized neural network graphs.
Implements conditionally routed efficient attention mechanisms for PyTorch, routing a subset of tokens through lighter and heavier branches to reduce computational cost in transformer architectures.
Swaps connections between output and input variables in ONNX model graphs, allowing you to redirect one node's output to feed into a different node's input.
Command-line tool for scaffolding, configuring, and deploying LangChain Deep Agents to the LangGraph Platform, with support for MCP server registration and sandbox backend configuration.
FlashOptim provides drop-in replacements for PyTorch optimizers (SGD, Adam, Lion variants) that reduce training memory by quantizing optimizer states, master weights, and gradients while maintaining model convergence.
However, it is early-stage (Alpha), Linux/CUDA-only, and the first optimizer step incurs Triton JIT overhead.
Integrates Google's Gemini LLM with LlamaIndex, though this integration is no longer supported as of version 0.6.2.
Provides pretrained neural network models for audio tagging and sound event detection on audio files using PANNs (Pretrained Audio Neural Networks).
However, do not expect bug fixes or updates—verify that the pretrained models and PyTorch compatibility meet your production requirements before committing to it for…
SAE Lens trains and analyzes sparse autoencoders for mechanistic interpretability research, with built-in support for PyTorch models and deep integration with TransformerLens and Hugging Face Transformers.
Integrates Google Vertex AI embedding models into LlamaIndex applications, providing access to text and multimodal embedding APIs with flexible credential handling.
Integrates Inworld's text-to-speech and speech-to-text APIs into LiveKit voice agents, enabling real-time voice synthesis and transcription for conversational AI applications.
Detects and segments QR codes in images using a YOLOv8-based model, returning bounding boxes, confidence scores, and precise polygon coordinates even in difficult positions.
Integrates LiteLLM with LlamaIndex to provide unified access to multiple LLM providers (OpenAI, Cohere, and others) through a single interface.
Install only if you actually need multi-provider flexibility; single-provider projects may not benefit.
Integrates LiteLLM with PandasAI to enable natural-language queries against dataframes using any LLM provider that LiteLLM supports.
However, the 0.0.1 version and absence of updates since release suggest early-stage maturity; verify compatibility with your target Python version and litellm setup…
Accelerates Transformer model training and inference on NVIDIA GPUs using low-precision formats including FP8, MXFP8, and NVFP4, with optimized kernels and mixed-precision support.
Reads and writes ARFF (Attribute-Relation File Format) files, a standard format for machine learning datasets, converting them to and from Python dictionaries and lists.
However, its abandoned status since 2020 makes it risky for new projects—install only if you need ARFF support for legacy datasets or existing workflows, and consider…
Hindsight-api-slim provides a persistent memory system for AI agents that stores facts, tracks entities and relationships, and enables temporal and semantic reasoning using PostgreSQL with pgvector.
Provides a MLServer runtime that loads and serves MLflow models, enabling MLflow-serialized models to run as inference endpoints with support for dictionary-based tensor inputs.
However, the aging maintenance status (434 days since release) means you should verify that the package's feature set matches your MLflow and MLServer versions before…
A plugin that extends LLM to access OpenAI models via the Responses API, enabling use of models like o1-pro that are unavailable through LLM's default Chat Completions implementation.
However, the aging maintenance status and preview designation mean you should expect possible API changes and verify that your target models are actually available…
Haystack is an end-to-end NLP framework for building LLM applications with pipelines, agents, and retrieval-augmented generation (RAG), connecting components like vector databases, language models, and document stores.
tpu-inference is a hardware plugin for vLLM that enables high-performance serving of large language models on Google TPUs, supporting both PyTorch and JAX model definitions through a unified backend.
However, it is strictly tied to TPU infrastructure—it cannot run on CPU or GPU systems.
TorchGeo provides PyTorch datasets, samplers, transforms, and pre-trained models for geospatial and remote sensing data, handling multispectral satellite imagery with automatic coordinate reference system alignment.
Connects LlamaIndex agents to MCP (Model Context Protocol) servers to expose and call remote tools, and converts LlamaIndex workflows into MCP servers.
Provides a lightweight Flash-Attention-3 forward-only kernel compiled to a Python wheel, optimized for inference workloads on CUDA hardware without backward pass or optional features.