LLM inference and serving

This project is a comprehensive educational curriculum and engineering handbook focused on the lifecycle of large language models. It serves as a structured knowledge base for machine learning practitioners, covering the fundamental mathematical and architectural principles of transformer-based sequence modeling, as well as the practical implementation of supervised instruction fine-tuning and preference-based model alignment. The repository distinguishes itself by providing a deep dive into advanced model composition and optimization techniques. It details methodologies for weight-space model merging and mixture-of-experts strategies, alongside practical guidance on low-precision parameter quantization and inference optimization to manage hardware requirements. Furthermore, it explores the development of autonomous agentic systems capable of tool-use orchestration and the construction of retrieval-augmented generation pipelines to ground model outputs in external data. The content spans the entire technical stack, from foundational deep learning concepts and neural network design to the complexities of deploying, evaluating, and securing models in production environments. It includes a curated collection of technical articles, blog posts, and interactive notebooks that track state-of-the-art research trends and experimental methodologies in generative artificial intelligence.

Transformers is a comprehensive library for machine learning that provides a unified interface for training, fine-tuning, and deploying transformer-based models. It supports a wide range of tasks, including text classification, language modeling, question answering, and sequence-to-sequence translation, while offering specialized architectures for both text and vision processing. The framework includes tools for managing the entire model lifecycle, from data preprocessing and tokenization to distributed training and inference. The library features extensive support for model optimization and performance, including techniques like quantization, speculative decoding, and paged memory management for key-value caches. It provides native integration for distributed training across multi-node clusters, as well as flexible APIs for serving models via compatible inference servers. Developers can also utilize built-in utilities for model patching, custom kernel execution, and automated documentation generation to streamline development workflows.

TensorFlow is a comprehensive machine learning framework designed for the construction, training, and deployment of complex mathematical models. It utilizes a graph-based execution model that represents operations as directed acyclic graphs, enabling automatic differentiation and efficient parallel processing. The system provides high-level interfaces for defining neural network architectures, alongside a robust engine for managing multidimensional array structures and tensor mathematics. The framework distinguishes itself through a scalable distributed runtime that orchestrates workloads across heterogeneous hardware accelerators and decentralized network nodes. It employs deferred-execution symbolic graphs to perform graph-level optimizations, fusion, and ahead-of-time kernel compilation for specific hardware architectures. To ensure consistent performance across production environments, it features a standardized serialization format for model graphs and specialized tools for model serving, quantization, and compression. Beyond core training capabilities, the platform includes a high-throughput data ingestion engine that supports asynchronous, multi-threaded pipelines to prevent bottlenecks. It also offers extensive support for hardware abstraction, allowing for pluggable device integration and containerized acceleration. The ecosystem is rounded out by utilities for data validation, federated learning, and specialized modeling tasks, providing a complete toolchain for moving models from research into high-availability production environments.

jetson-inference is a set of libraries and tools for executing optimized deep learning models on embedded GPU hardware. Its primary purpose is to enable real-time computer vision and AI inference at the edge with low latency and high throughput. The project distinguishes itself through high-performance streaming analytics and the ability to execute concurrent AI pipelines on auto-grade silicon. It provides specialized support for multi-sensor stream processing, utilizing zero-copy data transport to load camera frames directly into GPU memory. The codebase covers a broad surface of capabilities, including real-time video analytics, object detection and tracking, and image segmentation. It also integrates hardware-accelerated decoding and TensorRT-based inference to optimize model execution on embedded platforms. The project provides a TensorRT inference wrapper and an embedded vision SDK to facilitate the deployment of neural network primitives.

Llama is a computational framework and runtime environment designed for executing transformer-based neural networks locally. It functions as a generative AI inference engine, enabling the processing of input sequences through pre-trained model weights to produce text completions and structured data outputs directly on your own hardware. The system distinguishes itself through specialized memory and computation management techniques, including memory-mapped weight loading and quantization-aware inference, which allow for efficient execution on standard consumer hardware. It utilizes a stateless request execution model and a tensor-based computation graph to handle token-based sequence processing, ensuring that each inference task operates independently without reliance on persistent server state. This project provides the necessary tools for local large language model deployment, including a command-line interface for retrieving authorized model checkpoints and configuration files. It supports offline research and the integration of text generation capabilities into custom software applications, allowing users to manage model parameters such as sequence length and batch size to meet specific performance requirements.

LightLLM is a high-performance serving framework for deploying and executing large language models. It functions as a multi-GPU inference engine and server capable of handling dense architectures, mixture-of-experts designs, and multimodal models that process both text and images. The system is distinguished by its specialized support for Mixture-of-Experts models using expert parallelism and fused kernels. It implements structured text generation through deterministic state machines and pushdown automata to enforce precise output formats. To optimize throughput, the framework employs speculative decoding, paged key-value cache management, and a separated prefill and decode pipeline. The platform covers a broad range of operational capabilities, including tensor and data parallelism for scaling across hardware, multi-tier cache offloading for long context windows, and tool use integration for executing external functions. It also provides a standard interface for chat completions and dedicated tools for measuring request throughput and latency under real-world workloads. The project is implemented in Python and includes base classes for integrating custom model architectures.

Llama.cpp is an inference engine designed for the local execution of text-based and multimodal language models on consumer hardware. It provides a core environment for running models that process both text and image inputs, utilizing hardware-accelerated backends to optimize performance across diverse CPU and GPU architectures. The project distinguishes itself by offering a lightweight HTTP server that adheres to standard API specifications, enabling chat completion, embeddings, and reranking services. It includes a suite of tools for model quantization and conversion, which reduces memory usage and improves performance, alongside a command-line interface for managing chat templates and inference parameters. The ecosystem further supports structured data generation through grammar-based output constraints and provides diagnostic utilities for visualizing computational graphs. Comprehensive documentation is available, including a reference matrix that details the compatibility of computational operations across supported hardware backends.

vLLM is a high-throughput inference engine designed for the efficient serving and execution of large language models. It functions as a production-ready distributed model server, providing standard API protocols for online serving while also supporting offline batch processing. The system is built to maximize token generation speed and memory efficiency, enabling both large-scale cloud deployments and local execution on personal hardware. The project distinguishes itself through advanced memory management and request scheduling techniques, most notably its use of non-contiguous key-value cache blocks to eliminate fragmentation and its ability to dynamically insert new sequences into batches as they arrive. It provides a hardware-agnostic abstraction layer that maps complex mathematical operations to diverse accelerators, including specialized GPUs and consumer-grade silicon like Apple hardware. This is further supported by custom kernel fusion and a flexible quantization framework that allows for the compression of neural networks to fit resource-constrained environments. Beyond its core runtime, the framework offers extensive support for custom

This repository provides a collection of practical demonstrations and implementation guides for machine learning tasks using TensorFlow.js. It serves as a resource for developers to explore model architectures, training workflows, and data manipulation techniques across domains such as computer vision, natural language processing, and reinforcement learning. The project covers the full lifecycle of machine learning development, including tensor-based mathematical operations, model construction via high-level layer APIs or low-level tensor logic, and model serialization for various storage mediums. It includes utilities for converting models into browser-compatible formats and provides infrastructure for executing these models across diverse backends, including WebGL, WebAssembly, and CPU-accelerated environments. Documentation and examples are organized by task type, allowing users to browse implementations for regression, object detection, and generative models. The repository also includes deployment guides for hosting server-side applications on cloud platforms, alongside tools for managing tensor memory and asynchronous training processes.

LMCache is a distributed key-value cache manager and tiering system designed to accelerate large language model inference. It functions as a tiered storage layer that offloads tensors from GPU memory to CPU RAM, local disks, or remote object stores, enabling the reuse of cached prefixes across different inference sessions and serving engines. The system differentiates itself through a disaggregated prefill-decode model, which separates prompt processing from token generation by transferring caches between distributed compute nodes. It utilizes peer-to-peer orchestration to share and retrieve cached states across a cluster of servers, supported by a centralized coordinator for node membership and heartbeat monitoring. Broad capabilities include multi-tier storage management with support for S3, Redis, and POSIX filesystems, as well as performance optimizations such as asynchronous offloading, zero-copy shared memory transfers, and data quantization. The project also provides comprehensive observability through Prometheus and OpenTelemetry exports, alongside Kubernetes-based orchestration for deploying cache servers as DaemonSets.

LlamaFactory is a unified framework for fine-tuning and adapting large language models. It provides a comprehensive platform that standardizes training workflows across diverse machine learning architectures, allowing users to execute both full-tuning and parameter-efficient methods through a single interface. The project distinguishes itself by offering a low-code visual dashboard that enables users to configure experiments and monitor performance metrics in real time without writing extensive custom scripts. It also features a configuration-driven orchestration system that decouples experiment logic from the underlying execution engine, alongside an OpenAPI-compliant server that exposes trained models as standard network endpoints for integration with external software. Beyond its core training capabilities, the platform supports real-time experiment tracking by streaming performance data to external monitoring services. This allows for the evaluation of model progress and the optimization of parameters throughout the development lifecycle. The software is designed to be installed and configured as a standalone environment for managing the end-to-end lifecycle of language model adaptation.

Dynamo is a distributed inference orchestration platform designed for large language models. It functions as a system to coordinate prefill and decode phases across GPU nodes, utilizing a multi-backend runtime adapter to connect engines like vLLM and TensorRT-LLM through a unified block-oriented memory interface. An OpenAI-compatible API server provides the frontend for integration with existing tools and clients. The project is distinguished by its disaggregated serving architecture, which separates prompt processing and token generation onto independent GPU pools to optimize throughput and memory. It employs a key-value cache-aware request router that directs queries to workers holding relevant cache entries to reduce recomputation. High-speed data transfer mechanisms move cache blocks and weights directly between GPU VRAMs over RDMA or NVLink to minimize latency. The platform includes comprehensive capabilities for distributed fault tolerance, allowing in-flight requests to migrate and resume from failure points via token-state continuation. It features SLA-based autoscaling and performance profiling to right-size GPU pools and a Kubernetes-native operator for topology-aware scheduling. Additional support covers multimodal inference for images, video, and audio, alongside dynamic swapping of LoRA adapters. Installation is available via wheels, container images, charts, and crates, with support for major Linux distributions and NVIDIA GPU architectures from Ampere through Blackwell.

GPT4All is a cross-platform runtime environment designed to execute large language models directly on local consumer hardware. By leveraging an optimized C++ inference backend, it enables private, offline AI interactions without requiring an internet connection or external cloud services. The project provides a comprehensive ecosystem for managing the entire model lifecycle, including discovery, downloading, and configuration of local weights. What distinguishes the platform is its integrated retrieval-augmented generation engine, which allows users to index local documents into semantic vector spaces. This capability enables context-aware chat sessions where the model can reference private files, notes, and spreadsheets to provide grounded, relevant responses. The system also features a local HTTP server that exposes an OpenAI-compatible API, allowing developers to integrate these private, self-hosted models into existing applications and workflows. Beyond its core inference and retrieval capabilities, the project includes a graphical desktop interface for end-user interaction and a Python software development kit for programmatic access. These tools support advanced configuration of model parameters, performance monitoring, and the management of local embedding pipelines for custom semantic search tasks. The software is distributed as a unified application package, with documentation available to guide users through installation and local environment setup.

ChatGLM-6B is a generative AI inference engine designed for local execution of transformer-based language models. It provides a comprehensive runtime environment that allows users to load and run pre-trained neural network weights directly on their own hardware, ensuring data privacy and independence from external cloud services. The project distinguishes itself through a hardware-agnostic execution backend that supports deployment across diverse environments, including standard processors, Apple Silicon, and multi-GPU configurations. It incorporates advanced optimization techniques such as weight quantization and parameter-efficient fine-tuning via low-rank adaptation, which significantly reduce memory requirements and computational overhead. These features enable the deployment of large models on consumer-grade hardware while maintaining high throughput and performance. Beyond core inference, the toolkit includes a suite of utilities for programmatic integration, allowing developers to embed model capabilities into custom software workflows via standard interfaces. It also provides multiple interactive interfaces, including web-based graphical environments for text and vision tasks and a command-line interface for rapid prototyping and evaluation. The software is distributed as a Python-based package, requiring standard environment configuration to manage dependencies and hardware resource allocation.

Chitu is a distributed serving platform and orchestrator for large language model inference. It functions as a compute manager designed to deploy and scale model workloads across diverse hardware architectures, including GPUs, CPUs, and heterogeneous hardware clusters. The platform enables model deployment across a wide range of targets, including NVIDIA GPUs, regional chipsets, and legacy hardware. It manages the execution of models across these varying environments to increase available computing capacity and optimize resource utilization. The system includes capabilities for distributed inference orchestration and heterogeneous hardware scaling, allowing models to run on configurations ranging from single devices to large production clusters. It also incorporates concurrent traffic management and request queueing to maintain stability during high-demand workloads.

Nanochat is a lightweight execution environment designed for training and running language models on standard consumer hardware. It functions as both a neural network training framework and an inference engine, enabling users to perform backpropagation-based training and model execution directly on general-purpose processors without the need for dedicated graphics hardware. The project distinguishes itself through a suite of optimization tools that prioritize efficiency on local machines. By utilizing memory-mapped weight loading and CPU-optimized vector math, it maximizes throughput for interactive sessions. Furthermore, the framework includes a quantization toolkit that allows users to adjust the numerical precision of weights and activations, effectively balancing memory consumption against computational speed. The platform supports a range of capabilities for transformer architecture experimentation, including the configuration of training parameters and the management of local data pipelines. It employs a stateless generation loop to process tokens through self-contained execution cycles, facilitating the development and fine-tuning of custom models in a private, local environment.

This project is a containerized local AI infrastructure stack designed to deploy large language models and vector databases on private hardware. It functions as an orchestration platform that combines AI runners, knowledge graphs, and a visual workflow builder for creating agentic chatflows and automating tasks via tool integration. The platform distinguishes itself through a low-code approach to agent orchestration, utilizing a visual interface to design complex sequences and connect agents to external tools and search engines. It includes a dedicated local observability stack to track prompts, traces, and application performance, as well as hardware-specific optimization profiles to maximize inference speed on graphics processors and central processing units. The system covers a broad range of operational capabilities, including retrieval-augmented generation via vector database storage, centralized traffic routing with reverse proxy encryption, and shared-volume filesystem mounting for local data synchronization. It also manages network exposure to toggle between private and public web traffic configurations. The infrastructure is deployed as a pre-configured set of Docker-based services.

ColossalAI is a distributed deep learning framework designed for training and deploying massive artificial intelligence models across clusters of hardware accelerators. It functions as a parallel computing engine that partitions model workloads and data across multiple processors to maximize memory efficiency and throughput. The platform distinguishes itself through a comprehensive suite of parallelization strategies, including multi-dimensional tensor parallelism and pipeline-based model parallelism, which segment neural network layers and stages across devices. To support large-scale generative models in production, it provides a distributed inference runtime that utilizes dynamic request batching and optimized communication primitives to manage high volumes of concurrent traffic and minimize latency. The framework incorporates a large model optimization suite that enables the execution of complex models on limited hardware. This includes heterogeneous memory offloading, which moves parameters between GPU memory and system storage, and kernel-level computation optimizations that replace standard operations to reduce memory overhead. These capabilities facilitate both the training of massive models and the deployment of generative applications in production environments.