Search Results for "gpu max performance" - Page 3
Sort By:
Showing 155 open source projects for "gpu max performance"
View related business solutions-
Build Agents and Models on One PlatformGemini Enterprise Agent Platform is Google Cloud's comprehensive platform for developers to build, scale, govern, and optimize agents and models. Choose from Google's most advanced models and third-party models like Anthropic's Claude Model Family.
-
$300 Free Credits to Build on Google CloudPut your $300 in credit toward real workloads, then keep building with free monthly usage for 20+ products. No commitment and no charge until you upgrade.
-
1
LightLLM
LightLLM is a Python-based LLM (Large Language Model) inference
LightLLM is a high-performance inference and serving framework designed specifically for large language models, focusing on lightweight architecture, scalability, and efficient deployment. The framework enables developers to run and serve modern language models with significantly improved speed and resource efficiency compared to many traditional inference systems. -
2
RLax
Library of JAX-based building blocks for reinforcement learning agents
...It supports both on-policy and off-policy learning, as well as value-based, policy-based, and model-based approaches. RLax is fully JIT-compilable with JAX, enabling high-performance execution across CPU, GPU, and TPU backends. The library implements tools for Bellman equations, return distributions, general value functions, and policy optimization in both continuous and discrete action spaces. It integrates seamlessly with DeepMind’s Haiku (for neural network definition) and Optax (for optimization), making it a key component in modular RL pipelines. -
3
DLRM
An implementation of a deep learning recommendation model (DLRM)
...The architecture combines dense (MLP) and sparse (embedding) branches, then interacts features via dot product or feature interactions before passing through further dense layers to predict click-through, ranking scores, or conversion probabilities. The implementation is optimized for performance at scale, supporting multi-GPU and multi-node execution, quantization, embedding partitioning, and pipelined I/O to feed huge embeddings efficiently. It includes data loaders for standard benchmarks (like Criteo), training scripts, evaluation tools, and capabilities like mixed precision, gradient compression, and memory fusion to maximize throughput. -
4
MoCo (Momentum Contrast)
Self-supervised visual learning using momentum contrast in PyTorch
...The core idea of MoCo is to maintain a dynamic dictionary with a momentum-updated encoder, allowing efficient contrastive learning across large batches. The repository includes implementations for both MoCo v1 and MoCo v2, the latter improving training stability and performance through architectural and augmentation enhancements. Training is optimized for distributed multi-GPU environments, using DistributedDataParallel for speed and simplicity. -
Earn up to 16% annual interest with Nexo.Put idle assets to work with competitive interest rates, borrow without selling, and trade with precision. All in one platform. Geographic restrictions, eligibility, and terms apply.
-
5
BentoML
Unified Model Serving Framework
...Parallelize compute-intense model inference workloads to scale separately from the serving logic. Adaptive batching dynamically groups inference requests for optimal performance. Orchestrate distributed inference graph with multiple models via Yatai on Kubernetes. Easily configure CUDA dependencies for running inference with GPU. Automatically generate docker images for production deployment. -
6
DINOv3
Reference PyTorch implementation and models for DINOv3
DINOv3 is the third-generation iteration of Meta’s self-supervised visual representation learning framework, building upon the ideas from DINO and DINOv2. It continues the paradigm of learning strong image representations without labels using teacher–student distillation, but introduces a simplified and more scalable training recipe that performs well across datasets and architectures. DINOv3 removes the need for complex augmentations or momentum encoders, streamlining the pipeline while... -
7
PyOpenCL
OpenCL integration for Python, plus shiny features
PyOpenCL is a Python wrapper for the OpenCL framework, providing seamless access to parallel computing on CPUs, GPUs, and other accelerators. It enables developers to harness the full power of heterogeneous computing directly from Python, combining Python’s ease of use with the performance benefits of OpenCL. PyOpenCL also includes convenient features for managing memory, compiling kernels, and interfacing with NumPy, making it a preferred choice in scientific computing, data analysis, and... -
8
MaxText
A simple, performant and scalable Jax LLM
MaxText is a high-performance, highly scalable open-source framework designed to train and fine-tune large language models using the JAX ecosystem. The project acts as both a reference implementation and a practical training library that demonstrates best practices for building and scaling transformer-based language models on modern accelerator hardware. It is optimized to run efficiently on Google Cloud TPUs and GPUs, enabling researchers and engineers to train models ranging from small... -
9
xFormers
Hackable and optimized Transformers building blocks
xformers is a modular, performance-oriented library of transformer building blocks, designed to allow researchers and engineers to compose, experiment, and optimize transformer architectures more flexibly than monolithic frameworks. It abstracts components like attention layers, feedforward modules, normalization, and positional encoding, so you can mix and match or swap optimized kernels easily. One of its key goals is efficient attention: it supports dense, sparse, low-rank, and... -
Train ML Models With SQL You Already KnowBuild and deploy ML models using familiar SQL. Automate data prep with built-in Gemini. Query 1 TB and store 10 GB free monthly.
-
10
Qwen2.5-Omni
Capable of understanding text, audio, vision, video
Qwen2.5-Omni is an end-to-end multimodal flagship model in the Qwen series by Alibaba Cloud, designed to process multiple modalities (text, images, audio, video) and generate responses both as text and natural speech in streaming real-time. It supports “Thinker-Talker” architecture, and introduces innovations for aligning modalities over time (for example synchronizing video/audio), robust speech generation, and low-VRAM/quantized versions to make usage more accessible. It holds... -
11
JAX Toolbox
Public CI, Docker images for popular JAX libraries
JAX Toolbox is a development toolkit designed to streamline and optimize the use of JAX for machine learning and high-performance computing on NVIDIA GPUs. It provides prebuilt Docker images, continuous integration pipelines, and optimized example implementations that help developers quickly set up and run JAX workloads without complex configuration. The project supports popular JAX-based frameworks and models, including architectures used for large-scale pretraining such as GPT and LLaMA... -
12
LitServe
Minimal Python framework for scalable AI inference servers fast
LitServe is a minimal Python framework designed for building custom AI inference servers with full control over how models are executed and served. It allows developers to define their own inference logic, making it suitable for complex systems such as multi-model pipelines, agents, and retrieval-augmented generation workflows. Unlike traditional serving tools that enforce rigid abstractions, LitServe focuses on flexibility by letting users control request handling, batching strategies, and... -
13
dots.ocr
Multilingual Document Layout Parsing in a Single Vision-Language Model
dots.ocr is a cutting-edge multilingual document parsing system built on a unified vision-language model that combines layout detection, text recognition, and structural understanding into a single architecture. Unlike traditional OCR pipelines that rely on multiple specialized components, dots.ocr integrates these processes end-to-end, reducing error propagation and improving consistency across tasks. The model is designed to recognize virtually any human script, making it highly effective... -
14
Mosec
A high-performance ML model serving framework, offers dynamic batching
Mosec is a high-performance and flexible model-serving framework for building ML model-enabled backend and microservices. It bridges the gap between any machine learning models you just trained and the efficient online service API. -
15
Selkies-GStreamer
Open-Source Low-Latency Accelerated Linux WebRTC HTML5 Remote Desktop
selkies-gstreamer is a GStreamer-based media streaming component used in the Selkies project, a cloud-native platform designed for interactive desktop and application streaming. This module acts as a high-performance media pipeline that captures video, encodes it with low latency, and streams it via WebRTC to client browsers. It is optimized for GPU-accelerated encoding and integrates with Kubernetes-based deployments to enable scalable, real-time remote desktop sessions. This component plays a critical role in delivering smooth, responsive experiences for cloud-based workstations, gaming, or visualization tools. -
16
SageAttention
NeurIPS2025 Spotlight] Quantized Attention
SageAttention is an open-source optimization library designed to accelerate the attention mechanism used in transformer-based neural networks. Since attention operations are often the most computationally expensive component of modern AI models, SageAttention introduces quantization techniques that significantly reduce computational overhead while preserving model accuracy. The system achieves this by using low-precision numerical formats such as INT4, FP8, or INT8 to represent key matrices... -
17
RLHF-Reward-Modeling
Recipes to train reward model for RLHF
RLHF-Reward-Modeling is an open-source research framework focused on training reward models used in reinforcement learning from human feedback for large language models. In RLHF pipelines, reward models are responsible for evaluating generated responses and assigning scores that guide the model toward outputs that better match human preferences. The repository provides training recipes and implementations for building reward and preference models using modern machine learning frameworks. It... -
18
Stable Diffusion Version 2
High-Resolution Image Synthesis with Latent Diffusion Models
Stable Diffusion (the stablediffusion repo by Stability-AI) is an open-source implementation and reference codebase for high-resolution latent diffusion image models that power many text-to-image systems. The repository provides code for training and running Stable Diffusion-style models, instructions for installing dependencies (with notes about performance libraries like xformers), and guidance on hardware/driver requirements for efficient GPU inference and training. It’s organized as a practical, developer-focused toolkit: model code, scripts for inference, and examples for using memory-efficient attention and related optimizations are included so researchers and engineers can run or adapt the model for their own projects. ... -
19
DeepSpeed MII
MII makes low-latency and high-throughput inference possible
MII makes low-latency and high-throughput inference possible, powered by DeepSpeed. The Deep Learning (DL) open-source community has seen tremendous growth in the last few months. Incredibly powerful text generation models such as the Bloom 176B, or image generation model such as Stable Diffusion are now available to anyone with access to a handful or even a single GPU through platforms such as Hugging Face. While open-sourcing has democratized access to AI capabilities, their application is... -
20
Covalent workflow
Pythonic tool for running machine-learning/high performance workflows
Covalent is a Pythonic workflow tool for computational scientists, AI/ML software engineers, and anyone who needs to run experiments on limited or expensive computing resources including quantum computers, HPC clusters, GPU arrays, and cloud services. Covalent enables a researcher to run computation tasks on an advanced hardware platform – such as a quantum computer or serverless HPC cluster – using a single line of code. Covalent overcomes computational and operational challenges inherent... -
21
autoresearch for AMD
AI agents running research on single-GPU nanochat training
autoresearch for AMD is a framework for autonomous scientific experimentation in machine learning, enabling AI agents to iteratively improve models through a continuous loop of hypothesis generation, experimentation, and evaluation. The system is built around a minimal structure that includes a data preparation module, a training script that can be modified, and a program specification that guides the agent’s decision-making process. During each iteration, the agent edits the training code,... -
22
The SpeechBrain Toolkit
A PyTorch-based Speech Toolkit
...Spectral masking, spectral mapping, and time-domain enhancement are different methods already available within SpeechBrain. Separation methods such as Conv-TasNet, DualPath RNN, and SepFormer are implemented as well. SpeechBrain provides efficient and GPU-friendly speech augmentation pipelines and acoustic features extraction. -
23
ERNIE
The official repository for ERNIE 4.5 and ERNIEKit
ERNIE is an open-source large-model toolkit and model family from the PaddlePaddle ecosystem that focuses on training, fine-tuning, compression, and practical application of ERNIE large language models. The repository positions ERNIEKit as an industrial-grade development toolkit, emphasizing end-to-end workflows that span high-performance pre-training, supervised fine-tuning, and alignment. It supports both full-parameter training and parameter-efficient approaches so teams can choose between maximum quality and lower-cost adaptation depending on their constraints. The project also emphasizes optimization techniques for large-scale training, including mixed-precision and hybrid-parallel strategies that are commonly needed for multi-node GPU clusters. ... -
24
Halo
Framework built by White Circle for training large language
Performance features include FlashAttention, DeepEP, Liger kernels, Grouped GEMM, BF16 optimizers, LoRA, and QLoRA. RL rollouts can run through vLLM or SGLang while overlapping with training. Halo also provides Docker images, YAML recipes, checkpoint tools, dataset utilities, benchmarks, and agent-oriented technical documentation. -
25
JAX
Composable transformations of Python+NumPy programs
With its updated version of Autograd, JAX can automatically differentiate native Python and NumPy functions. It can differentiate through loops, branches, recursion, and closures, and it can take derivatives of derivatives of derivatives. It supports reverse-mode differentiation (a.k.a. backpropagation) via grad as well as forward-mode differentiation, and the two can be composed arbitrarily to any order. What’s new is that JAX uses XLA to compile and run your NumPy programs on GPUs and...
