Open-source alternatives to Tensorflow Tpu

Search-R1 is a distributed training system and reinforcement learning framework designed to create search-augmented language models. It provides an architecture for scaling model workloads across head and worker nodes while optimizing how models interleave internal reasoning with external tool calls. The system focuses on refining model behavior through custom reward signals and reinforcement learning to improve tool-use formatting and information retrieval. It implements an interleaved reasoning-search loop that allows models to alternate between internal thought generation and external data

AISystem is a comprehensive AI full-stack infrastructure project covering the entire pipeline from AI chip architecture to high-level training frameworks. It encompasses the development of AI compiler frameworks, inference engines, and distributed training orchestrators designed to coordinate workloads across a heterogeneous compute stack of CPUs, GPUs, and NPUs. The project focuses on the deep integration of software and hardware, employing software-hardware co-design to align tensor layouts with physical memory structures. It provides specialized capabilities for accelerating Transformer mo

ZenML is an orchestration platform designed for building, deploying, and monitoring reproducible machine learning pipelines and agentic workflows. It provides a unified framework that manages the entire lifecycle of machine learning assets, from data processing and model training to the deployment of persistent inference services. By decoupling pipeline logic from underlying compute and storage, the platform enables teams to transition workflows seamlessly from local development environments to production-grade cloud infrastructure. The platform distinguishes itself through a service-oriented

This project is a JAX-based transformer framework and large language model trainer designed for building and training distributed models on TPU hardware accelerators. It provides a system for pretraining and fine-tuning autoregressive models by splitting weights and computations across a mesh of devices to reduce memory overhead and increase processing speed. The framework includes a TPU compute orchestrator for provisioning resources and automating dependency installation across remote distributed nodes. It also features a model weight converter capable of transforming and resharding checkpo

xtuner is a comprehensive training engine for large language models, offering a toolkit for pre-training, supervised fine-tuning, and the optimization of vision-language multimodal models. It serves as a distributed training accelerator and a specialized framework for scaling Mixture-of-Experts models and aligning model behavior through reinforcement learning from human feedback. The project distinguishes itself through advanced memory and compute optimizations, such as sequence parallelism for ultra-long context windows and interleaved pipeline parallelism to reduce GPU idle time. It provide

AReaL is a system for agent orchestration, distributed model training, and parameter-efficient tuning. It provides a framework for developing multi-turn reasoning agents and training large models using reinforcement learning from human feedback. The project implements a toolkit for improving the visual reasoning and geometry problem solving capabilities of vision-language models. It utilizes a memory-efficient tuning system to optimize mathematical and reasoning models across different inference backends. The infrastructure supports large-scale training through tensor, pipeline, and expert p

Pythia is a multimodal research framework and distributed training system designed for building, training, and evaluating large models that combine visual and linguistic data. It provides a modular environment for developing vision-language models, focusing on the integration of image and text inputs into shared feature representations. The framework utilizes a modular architecture that decouples model building blocks into interchangeable components, allowing for flexible configuration of vision and language modules. It includes a benchmark suite for executing reference models against standar

Lit-llama is a PyTorch-based implementation framework for the LLaMA language model, providing a system for pre-training, fine-tuning, and high-performance inference. It includes a pre-training pipeline for creating foundational language models from scratch and tools for running pretrained weights to generate natural text and predict sequences. The project provides specialized toolkits for parameter-efficient fine-tuning using low-rank adaptation and lightweight adapters. It also includes a quantization library that reduces model memory footprints through four-bit and eight-bit precision to en

Torchtune is a PyTorch-native library for fine-tuning, aligning, and quantizing large language models. It provides a config-driven system for instantiating components, orchestrating distributed training, and managing parameter-efficient fine-tuning with quantization support, all through YAML-based configurations and command-line overrides. The library distinguishes itself through its comprehensive post-training workflow orchestration, combining supervised fine-tuning, preference optimization (DPO, PPO, GRPO), knowledge distillation, and quantization-aware training in a single configurable pip

Open CLIP is an open source framework for training and deploying Contrastive Language-Image Pre-training models. It serves as a vision-language training framework and multimodal embedding engine that maps images and text into a shared vector space for similarity searches and zero-shot classification. The project provides a toolkit for distributed training of contrastive models and includes an image-to-text generative model for producing natural language descriptions. It supports custom text encoder integration and utilizes teacher-student model distillation to transfer knowledge from large pr

Cube Studio is a cloud-native MLOps platform and Kubernetes-based AI orchestrator designed for the entire machine learning lifecycle. It provides a distributed training framework for large-scale model fine-tuning, a GPU resource manager for hardware virtualization, and an ML pipeline orchestrator that uses visual directed acyclic graphs to manage end-to-end workflows. The platform distinguishes itself through its specialized LLM inference server, which supports retrieval-augmented generation and the construction of private knowledge bases. It features a dedicated system for supervised fine-tu

This project is a research framework and toolkit designed for training large-scale vision transformers and multimodal language models. It provides a comprehensive suite for vision-language pretraining, enabling the development of models that map images and text into shared latent spaces. The framework is distinguished by its capabilities in high-fidelity image generation and multimodal research, utilizing normalizing flows and variational autoencoders to produce images from text prompts or class labels. It supports the development of both generative and contrastive models, allowing for a wide

EasyR1 is a distributed model training system and reinforcement learning framework for large language and vision-language models. It functions as a multimodal trainer and an implementation of a Proximal Policy Optimization pipeline designed to refine the reasoning and perception capabilities of models that process both text and images. The system specializes in distributing reinforcement learning workloads across multiple compute nodes to manage high memory requirements. It optimizes hardware utilization through padding-free training and fine-tuning to fit large models onto available graphics

llm-foundry is a training framework for large language models, providing a system for foundation model pre-training and supervised fine-tuning. It includes a distributed trainer for scaling workloads across multiple nodes and GPUs, a dataset streaming pipeline for loading data from cloud storage, and a parameter-efficient fine-tuning implementation. The framework distinguishes itself through its use of parameter sharding and high-throughput data streaming to maintain stability during large-scale training. It incorporates low-rank adaptation to reduce computational costs and uses eight-bit flo

mmagic is a multimodal training pipeline and framework for generative AI, focusing on visual synthesis and restoration. It provides the infrastructure to build and train models for tasks such as text-to-image and text-to-video generation, 3D-aware content synthesis, and high-fidelity image translation using diffusion models and generative adversarial networks. The project distinguishes itself through specialized capabilities for generative model personalization, including techniques for fine-tuning subjects and styles. It also supports advanced visual manipulations such as latent space interp

Torchtune is a PyTorch-native library for fine-tuning, aligning, and quantizing large language models. It provides a configurable training pipeline orchestrated through YAML recipes, with CLI overrides and component swapping, distributed training via FSDP2, memory optimizations, and parameter-efficient fine-tuning methods like LoRA, DoRA, and QLoRA. The library distinguishes itself through its YAML-driven configuration system that defines all training parameters and instantiates components from config files, with full CLI override capability for any field or component at launch time. It suppo

Corenet is a deep learning training framework and computer vision model library designed for developing neural networks across vision, text, and audio modalities. It functions as a distributed training orchestrator for scaling workloads across multiple compute nodes and provides a multimodal data pipeline for processing image, text, and video data. The project includes a model conversion toolkit for transforming weights and architectures between different machine learning frameworks. It also provides tools for optimizing model performance on Apple Silicon and reducing response latency in gene

This is a machine learning framework for treating diverse natural language processing tasks as a unified text-to-text problem. It provides a toolkit for pre-training and fine-tuning large-scale transformer models, utilizing a system where both inputs and outputs are formatted as raw text sequences. The framework is distinguished by its distributed training system, which uses mesh-based strategies to scale model weights and training batches across multiple TPU cores. It supports multi-task learning by combining diverse datasets into a single training stream using configurable mixture rates, al

MindSpore is a deep learning framework designed for building and training neural networks across cloud, edge, and mobile environments. It functions as a distributed training system and a hardware accelerated AI toolkit capable of executing workloads on CPUs, GPUs, and specialized AI processors. The project includes an automatic differentiation engine that computes gradients through source transformation and static compilation. It enables distributed model training by splitting workloads across hardware using data and model parallelism. The framework covers cross-platform AI deployment and mo

This project is an implementation of the Transformer-XL language model, a neural network architecture designed for long-context language modeling. It provides frameworks for training and deploying models that capture long-term dependencies and relationships in text sequences that extend beyond a fixed context window. The implementation supports both PyTorch and TensorFlow, allowing for distributed training across multiple GPUs and host nodes. It employs a recurrent mechanism to maintain coherence in extended sequences, utilizing segment-level recurrence and state-based memory reuse. The code

kops is a Kubernetes cluster provisioner and lifecycle manager designed to automate the creation, maintenance, and destruction of production-grade clusters on cloud infrastructure. It functions as a declarative infrastructure manager, synchronizing the live state of a cluster with versioned manifests stored in remote object storage to ensure idempotent operations. The project distinguishes itself by offering comprehensive automation for the entire cluster lifecycle, including high-availability control plane deployment, incremental rolling updates, and automated version upgrades. It also serve

karpenter-provider-aws is a Kubernetes node autoscaler and infrastructure provider for AWS. It serves as a node lifecycle manager and cluster cost optimizer that automatically provisions and removes compute instances based on the resource requirements of pending pods. The project distinguishes itself through advanced AWS spot instance orchestration and price-capacity optimized selection to reduce cloud spend. It minimizes costs by consolidating underutilized nodes and prioritizing spot or reserved instances over on-demand capacity, while proactively migrating workloads before cloud provider i

This project is a comprehensive educational resource and curriculum focused on the design and implementation of the full machine learning software and hardware stack. It serves as a technical reference for architecting machine learning systems, spanning from low-level programming interfaces to large-scale deployment infrastructure. The project provides instructional guidance on several specialized domains, including the development of AI compilers through intermediate representations and graph optimizations. It covers the architectural patterns required for distributed training across GPU clu

RLinf is a distributed reinforcement learning orchestrator and embodied AI training framework. It provides the infrastructure to train vision-language-action models and robotic policies using a combination of reinforcement learning and supervised fine-tuning. The system is designed for scaling workloads across GPU clusters, managing the placement of actors, rollout workers, and environment components. It features a specialized robotics data collection pipeline for gathering teleoperated demonstrations and simulation trajectories into standardized replay buffers, alongside a hardware interface

This project is a natural language processing framework focused on a generalized autoregressive pretrainer designed for unsupervised language representation. It implements a language model that combines permutation-based training with a Transformer-XL backbone to function as a long-context text processor. The system is distinguished by its ability to handle text sequences that exceed standard length limits through the use of segment-level recurrence and relative positional encoding. It scales high-performance pretraining across multiple GPUs and TPU clusters using distributed training impleme

Minigo is a TensorFlow-based reinforcement learning engine designed to master the game of Go. It functions as a comprehensive system for training neural networks to predict board policies and game outcomes, utilizing a model trainer to generate self-play data and optimize weights. The project is distinguished by its ability to perform large-scale game simulations using Kubernetes to distribute worker nodes across CPU, GPU, and TPU hardware. It employs a Monte Carlo Tree Search implementation to identify optimal moves and supports specialized hardware acceleration, including inference on Edge

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 genera