Checkpoint-engine is a middleware tool designed for updating model weights in LLM inference engines efficiently. It provides implementations for both Broadcast and P2P weight update methods, orchestrating the transfer process and controlling the inference engine through ZeroMQ socket. The tool optimizes weight broadcast by arranging data transfer into stages and organizing transfers into a pipeline for performance. It supports flexible installation options and is tested with various models and device setups. Checkpoint-engine also allows reusing weights from existing instances and provides a patch for FP8 quantization in vLLM.

This repository contains the training and inference code needed to replicate the winning solution to the AI Mathematical Olympiad - Progress Prize 1. It consists of fine-tuning DeepSeekMath-Base 7B, high-quality training datasets, a self-consistency decoding algorithm, and carefully chosen validation sets. The training methodology involves Chain of Thought (CoT) and Tool Integrated Reasoning (TIR) training stages. Two datasets, NuminaMath-CoT and NuminaMath-TIR, were used to fine-tune the models. The models were trained using open-source libraries like TRL, PyTorch, vLLM, and DeepSpeed. Post-training quantization to 8-bit precision was done to improve performance on Kaggle's T4 GPUs. The project structure includes scripts for training, quantization, and inference, along with necessary installation instructions and hardware/software specifications.

SmolChat-Android is a mobile application that enables users to interact with local small language models (SLMs) on-device. Users can add/remove SLMs, modify system prompts and inference parameters, create downstream tasks, and generate responses. The app uses llama.cpp for model execution, ObjectBox for database storage, and Markwon for markdown rendering. It provides a simple, extensible codebase for on-device machine learning projects.

NineRec is a benchmark dataset suite for evaluating transferable recommendation models. It provides datasets for pre-training and transfer learning in recommender systems, focusing on multimodal and foundation model tasks. The dataset includes user-item interactions, item texts in multiple languages, item URLs, and raw images. Researchers can use NineRec to develop more effective and efficient methods for pre-training recommendation models beyond end-to-end training. The dataset is accompanied by code for dataset preparation, training, and testing in PyTorch environment.

NVIDIA NeMo Framework is a scalable and cloud-native generative AI framework built for researchers and PyTorch developers working on Large Language Models (LLMs), Multimodal Models (MMs), Automatic Speech Recognition (ASR), Text to Speech (TTS), and Computer Vision (CV) domains. It is designed to help you efficiently create, customize, and deploy new generative AI models by leveraging existing code and pre-trained model checkpoints.

Yalm (Yet Another Language Model) is an LLM inference implementation in C++/CUDA, emphasizing performance engineering, documentation, scientific optimizations, and readability. It is not for production use and has been tested on Mistral-v0.2 and Llama-3.2. Requires C++20-compatible compiler, CUDA toolkit, and LLM safetensor weights in huggingface format converted to .yalm file.

CompressAI-Vision is a tool that helps you develop, test, and evaluate compression models with standardized tests in the context of compression methods optimized for machine tasks algorithms such as Neural-Network (NN)-based detectors. It currently focuses on two types of pipeline: Video compression for remote inference (`compressai-remote-inference`), which corresponds to the MPEG "Video Coding for Machines" (VCM) activity. Split inference (`compressai-split-inference`), which includes an evaluation framework for compressing intermediate features produced in the context of split models. The software supports all the pipelines considered in the related MPEG activity: "Feature Compression for Machines" (FCM).

Atom is an accurate low-bit weight-activation quantization algorithm that combines mixed-precision, fine-grained group quantization, dynamic activation quantization, KV-cache quantization, and efficient CUDA kernels co-design. It introduces a low-bit quantization method, Atom, to maximize Large Language Models (LLMs) serving throughput with negligible accuracy loss. The codebase includes evaluation of perplexity and zero-shot accuracy, kernel benchmarking, and end-to-end evaluation. Atom significantly boosts serving throughput by using low-bit operators and reduces memory consumption via low-bit quantization.

Model2Vec is a technique to turn any sentence transformer into a really small static model, reducing model size by 15x and making the models up to 500x faster, with a small drop in performance. It outperforms other static embedding models like GLoVe and BPEmb, is lightweight with only `numpy` as a major dependency, offers fast inference, dataset-free distillation, and is integrated into Sentence Transformers, txtai, and Chonkie. Model2Vec creates powerful models by passing a vocabulary through a sentence transformer model, reducing dimensionality using PCA, and weighting embeddings using zipf weighting. Users can distill their own models or use pre-trained models from the HuggingFace hub. Evaluation can be done using the provided evaluation package. Model2Vec is licensed under MIT.

The development of Natural Language Interfaces to Databases (NLIDBs) has been greatly advanced by the advent of large language models (LLMs), which provide an intuitive way to translate natural language (NL) questions into Structured Query Language (SQL) queries. DBCopilot is a framework that addresses challenges in real-world scenarios of natural language querying over massive databases by employing a compact and flexible copilot model for routing. It decouples schema-agnostic NL2SQL into schema routing and SQL generation, utilizing a lightweight differentiable search index for semantic mappings and relation-aware joint retrieval. DBCopilot introduces a reverse schema-to-question generation paradigm for automatic learning and adaptation over massive databases, providing a scalable and effective solution for schema-agnostic NL2SQL.

MSCCL++ is a GPU-driven communication stack for scalable AI applications. It provides a highly efficient and customizable communication stack for distributed GPU applications. MSCCL++ redefines inter-GPU communication interfaces, delivering a highly efficient and customizable communication stack for distributed GPU applications. Its design is specifically tailored to accommodate diverse performance optimization scenarios often encountered in state-of-the-art AI applications. MSCCL++ provides communication abstractions at the lowest level close to hardware and at the highest level close to application API. The lowest level of abstraction is ultra light weight which enables a user to implement logics of data movement for a collective operation such as AllReduce inside a GPU kernel extremely efficiently without worrying about memory ordering of different ops. The modularity of MSCCL++ enables a user to construct the building blocks of MSCCL++ in a high level abstraction in Python and feed them to a CUDA kernel in order to facilitate the user's productivity. MSCCL++ provides fine-grained synchronous and asynchronous 0-copy 1-sided abstracts for communication primitives such as `put()`, `get()`, `signal()`, `flush()`, and `wait()`. The 1-sided abstractions allows a user to asynchronously `put()` their data on the remote GPU as soon as it is ready without requiring the remote side to issue any receive instruction. This enables users to easily implement flexible communication logics, such as overlapping communication with computation, or implementing customized collective communication algorithms without worrying about potential deadlocks. Additionally, the 0-copy capability enables MSCCL++ to directly transfer data between user's buffers without using intermediate internal buffers which saves GPU bandwidth and memory capacity. MSCCL++ provides consistent abstractions regardless of the location of the remote GPU (either on the local node or on a remote node) or the underlying link (either NVLink/xGMI or InfiniBand). This simplifies the code for inter-GPU communication, which is often complex due to memory ordering of GPU/CPU read/writes and therefore, is error-prone.

Habitat-Lab is a modular high-level library for end-to-end development in embodied AI. It is designed to train agents to perform a wide variety of embodied AI tasks in indoor environments, as well as develop agents that can interact with humans in performing these tasks.

Stellar Chat is a multi-modal chat application that enables users to create custom agents and integrate with local language models and OpenAI models. It provides capabilities for generating images, visual recognition, text-to-speech, and speech-to-text functionalities. Users can engage in multimodal conversations, create custom agents, search messages and conversations, and integrate with various applications for enhanced productivity. The project is part of the '100 Commits' competition, challenging participants to make meaningful commits daily for 100 consecutive days.

AI Edge Torch is a Python library that supports converting PyTorch models into a .tflite format for on-device applications on Android, iOS, and IoT devices. It offers broad CPU coverage with initial GPU and NPU support, closely integrating with PyTorch and providing good coverage of Core ATen operators. The library includes a PyTorch converter for model conversion and a Generative API for authoring mobile-optimized PyTorch Transformer models, enabling easy deployment of Large Language Models (LLMs) on mobile devices.

hugegraph-ai aims to explore the integration of HugeGraph with artificial intelligence (AI) and provide comprehensive support for developers to leverage HugeGraph's AI capabilities in their projects. It includes modules for large language models, graph machine learning, and a Python client for HugeGraph. The project aims to address challenges like timeliness, hallucination, and cost-related issues by integrating graph systems with AI technologies.

Axone is a public dPoS layer 1 designed for connecting, sharing, and monetizing resources in the AI stack. It is an open network for collaborative AI workflow management compatible with any data, model, or infrastructure, allowing sharing of data, algorithms, storage, compute, APIs, both on-chain and off-chain. The 'axoned' node of the AXONE network is built on Cosmos SDK & Tendermint consensus, enabling companies & individuals to define on-chain rules, share off-chain resources, and create new applications. Validators secure the network by maintaining uptime and staking $AXONE for rewards. The blockchain supports various platforms and follows Semantic Versioning 2.0.0. A docker image is available for quick start, with documentation on querying networks, creating wallets, starting nodes, and joining networks. Development involves Go and Cosmos SDK, with smart contracts deployed on the AXONE blockchain. The project provides a Makefile for building, installing, linting, and testing. Community involvement is encouraged through Discord, open issues, and pull requests.

Edgen is a local GenAI API server that serves as a drop-in replacement for OpenAI's API. It provides multi-endpoint support for chat completions and speech-to-text, is model agnostic, offers optimized inference, and features model caching. Built in Rust, Edgen is natively compiled for Windows, MacOS, and Linux, eliminating the need for Docker. It allows users to utilize GenAI locally on their devices for free and with data privacy. With features like session caching, GPU support, and support for various endpoints, Edgen offers a scalable, reliable, and cost-effective solution for running GenAI applications locally.

EasyDist is an automated parallelization system and infrastructure designed for multiple ecosystems. It offers usability by making parallelizing training or inference code effortless with just a single line of change. It ensures ecological compatibility by serving as a centralized source of truth for SPMD rules at the operator-level for various machine learning frameworks. EasyDist decouples auto-parallel algorithms from specific frameworks and IRs, allowing for the development and benchmarking of different auto-parallel algorithms in a flexible manner. The architecture includes MetaOp, MetaIR, and the ShardCombine Algorithm for SPMD sharding rules without manual annotations.

Since the emergence of chatGPT in 2022, the acceleration of Large Language Model has become increasingly important. Here is a list of papers on LLMs inference and serving.

LLM-Viewer is a tool for visualizing Language and Learning Models (LLMs) and analyzing performance on different hardware platforms. It enables network-wise analysis, considering factors such as peak memory consumption and total inference time cost. With LLM-Viewer, users can gain valuable insights into LLM inference and performance optimization. The tool can be used in a web browser or as a command line interface (CLI) for easy configuration and visualization. The ongoing project aims to enhance features like showing tensor shapes, expanding hardware platform compatibility, and supporting more LLMs with manual model graph configuration.

ServerlessLLM is a fast, affordable, and easy-to-use library designed for multi-LLM serving, optimized for environments with limited GPU resources. It supports loading various leading LLM inference libraries, achieving fast load times, and reducing model switching overhead. The library facilitates easy deployment via Ray Cluster and Kubernetes, integrates with the OpenAI Query API, and is actively maintained by contributors.

Sarathi-Serve is the official OSDI'24 artifact submission for paper #444, focusing on 'Taming Throughput-Latency Tradeoff in LLM Inference'. It is a research prototype built on top of CUDA 12.1, designed to optimize throughput-latency tradeoff in Large Language Models (LLM) inference. The tool provides a Python environment for users to install and reproduce results from the associated experiments. Users can refer to specific folders for individual figures and are encouraged to cite the paper if they use the tool in their work.

Aphrodite is the official backend engine for PygmalionAI, serving as the inference endpoint for the website. It allows serving Hugging Face-compatible models with fast speeds. Features include continuous batching, efficient K/V management, optimized CUDA kernels, quantization support, distributed inference, and 8-bit KV Cache. The engine requires Linux OS and Python 3.8 to 3.12, with CUDA >= 11 for build requirements. It supports various GPUs, CPUs, TPUs, and Inferentia. Users can limit GPU memory utilization and access full commands via CLI.

uTensor is an extremely light-weight machine learning inference framework built on Tensorflow and optimized for Arm targets. It consists of a runtime library and an offline tool that handles most of the model translation work. The core runtime is only ~2KB. The workflow involves constructing and training a model in Tensorflow, then using uTensor to produce C++ code for inferencing. The runtime ensures system safety, guarantees RAM usage, and focuses on clear, concise, and debuggable code. The high-level API simplifies tensor handling and operator execution for embedded systems.

llm-resource is a comprehensive collection of high-quality resources for Large Language Models (LLM). It covers various aspects of LLM including algorithms, training, fine-tuning, alignment, inference, data engineering, compression, evaluation, prompt engineering, AI frameworks, AI basics, AI infrastructure, AI compilers, LLM application development, LLM operations, AI systems, and practical implementations. The repository aims to gather and share valuable resources related to LLM for the community to benefit from.

LitServe is a high-throughput serving engine designed for deploying AI models at scale. It generates an API endpoint for models, handles batching, streaming, and autoscaling across CPU/GPUs. LitServe is built for enterprise scale with a focus on minimal, hackable code-base without bloat. It supports various model types like LLMs, vision, time-series, and works with frameworks like PyTorch, JAX, Tensorflow, and more. The tool allows users to focus on model performance rather than serving boilerplate, providing full control and flexibility.

This repository documents how to optimize common algorithms based on CUDA. It includes subdirectories with code implementations for specific optimizations. The optimizations cover topics such as compiling PyTorch from source, NVIDIA's reduce optimization, OneFlow's elementwise template, fast atomic add for half data types, upsample nearest2d optimization in OneFlow, optimized indexing in PyTorch, OneFlow's softmax kernel, linear attention optimization, and more. The repository also includes learning resources related to deep learning frameworks, compilers, and optimization techniques.

AIAC is a library and command line tool to generate Infrastructure as Code (IaC) templates, configurations, utilities, queries, and more via LLM providers such as OpenAI, Amazon Bedrock, and Ollama. Users can define multiple 'backends' targeting different LLM providers and environments using a simple configuration file. The tool allows users to ask a model to generate templates for different scenarios and composes an appropriate request to the selected provider, storing the resulting code to a file and/or printing it to standard output.

ENOVA is an open-source service for Large Language Model (LLM) deployment, monitoring, injection, and auto-scaling. It addresses challenges in deploying stable serverless LLM services on GPU clusters with auto-scaling by deconstructing the LLM service execution process and providing configuration recommendations and performance detection. Users can build and deploy LLM with few command lines, recommend optimal computing resources, experience LLM performance, observe operating status, achieve load balancing, and more. ENOVA ensures stable operation, cost-effectiveness, efficiency, and strong scalability of LLM services.

Jina is a tool that allows users to build multimodal AI services and pipelines using cloud-native technologies. It provides a Pythonic experience for serving ML models and transitioning from local deployment to advanced orchestration frameworks like Docker-Compose, Kubernetes, or Jina AI Cloud. Users can build and serve models for any data type and deep learning framework, design high-performance services with easy scaling, serve LLM models while streaming their output, integrate with Docker containers via Executor Hub, and host on CPU/GPU using Jina AI Cloud. Jina also offers advanced orchestration and scaling capabilities, a smooth transition to the cloud, and easy scalability and concurrency features for applications. Users can deploy to their own cloud or system with Kubernetes and Docker Compose integration, and even deploy to JCloud for autoscaling and monitoring.

Vidur is a high-fidelity and extensible LLM inference simulator designed for capacity planning, deployment configuration optimization, testing new research ideas, and studying system performance of models under different workloads and configurations. It supports various models and devices, offers chrome trace exports, and can be set up using mamba, venv, or conda. Users can run the simulator with various parameters and monitor metrics using wandb. Contributions are welcome, subject to a Contributor License Agreement and adherence to the Microsoft Open Source Code of Conduct.

AI System School is a curated list of research in machine learning systems, focusing on ML/DL infra, LLM infra, domain-specific infra, ML/LLM conferences, and general resources. It provides resources such as data processing, training systems, video systems, autoML systems, and more. The repository aims to help users navigate the landscape of AI systems and machine learning infrastructure, offering insights into conferences, surveys, books, videos, courses, and blogs related to the field.