Search Results for "throughput"
Sort By:
Showing 55 open source projects for "throughput"
View related business solutions-
$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.
-
Ship Agents FasterGemini Enterprise Agent Platform lets you rapidly build, scale, govern and optimize production-ready agents grounded in your organization's data. The platform enables developers to build custom or pre-built agents for virtually any use case. New customers get $300 in free credits.
-
1
vLLM
A high-throughput and memory-efficient inference and serving engine
vLLM is a fast and easy-to-use library for LLM inference and serving. High-throughput serving with various decoding algorithms, including parallel sampling, beam search, and more. -
2
DeepSpeed
Deep learning optimization library: makes distributed training easy
DeepSpeed is an easy-to-use deep learning optimization software suite that enables unprecedented scale and speed for Deep Learning Training and Inference. With DeepSpeed you can: 1. Train/Inference dense or sparse models with billions or trillions of parameters 2. Achieve excellent system throughput and efficiently scale to thousands of GPUs 3. Train/Inference on resource constrained GPU systems 4. Achieve unprecedented low latency and high throughput for inference 5. Achieve extreme compression for an unparalleled inference latency and model size reduction with low costs DeepSpeed offers a confluence of system innovations, that has made large scale DL training effective, and efficient, greatly improved ease of use, and redefined the DL training landscape in terms of scale that is possible. ... -
3
Parallax
Parallax is a distributed model serving framework
...A two-stage scheduling architecture determines how model layers are allocated to available hardware and how requests are routed across nodes during execution. This scheduling system optimizes latency, throughput, and hardware utilization even when nodes have different computational capabilities. The platform also supports model sharding and pipeline parallelism, allowing very large models to run across distributed resources. -
4
LitServe
Minimal Python framework for scalable AI inference servers fast
...Unlike traditional serving tools that enforce rigid abstractions, LitServe focuses on flexibility by letting users control request handling, batching strategies, and output processing directly in Python. LitServe is built on top of FastAPI and extends it with AI-specific optimizations such as efficient multi-worker execution, which can significantly improve throughput. It includes built-in capabilities for batching, streaming responses, and automatic scaling across CPUs and GPUs, enabling high-performance deployments. -
Cut Data Warehouse Costs by 54%BigQuery delivers 54% lower TCO with exabyte scale and flexible pricing. Free migration tools handle the SQL translation automatically.
-
5
Text Embeddings Inference
High-performance inference server for text embeddings models API layer
...It provides an API interface that allows developers to integrate embedding capabilities into applications without managing model internals directly. Text Embeddings Inference is optimized for throughput and low latency, enabling it to handle large volumes of requests reliably. It also emphasizes ease of deployment, often using containerization and configurable runtime options to adapt to different infrastructure setups. -
6
FlexLLMGen
Running large language models on a single GPU
...This design allows organizations to deploy powerful language models for high-volume tasks without the infrastructure costs typically associated with large-scale AI systems. The project is particularly useful for workloads that prioritize throughput over latency, including benchmarking experiments and large corpus analysis. -
7
MemOS
AI memory OS for LLM and Agent systems
...By abandoning some of the historical assumptions of Unix-style operating systems, MemOS attempts to unlock new performance and scalability tradeoffs for applications that need high throughput and low latency on memory-intensive workloads. -
8
Infinity
Low-latency REST API for serving text-embeddings
Infinity is a high-throughput, low-latency REST API for serving vector embeddings, supporting all sentence-transformer models and frameworks. Infinity is developed under MIT License. Infinity powers inference behind Gradient.ai and other Embedding API providers. -
9
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. -
Save Up to 91% on Cloud Compute With Spot VMsRun batch jobs at 60-91% off with Spot VMs. Long-running workloads get automatic discounts with sustained use.
-
10
SimpleLLM
950 line, minimal, extensible LLM inference engine built from scratch
...Designed to run efficiently on high-end GPUs like NVIDIA H100 with support for models such as OpenAI/gpt-oss-120b, Simple-LLM implements continuous batching and event-driven inference loops to maximize hardware utilization and throughput. Its straightforward code structure allows anyone experimenting with custom kernels, new batching strategies, or inference optimizations to trace execution from input to output with minimal cognitive overhead. -
11
Text Generation Inference
Large Language Model Text Generation Inference
Text Generation Inference is a high-performance inference server for text generation models, optimized for Hugging Face's Transformers. It is designed to serve large language models efficiently with optimizations for performance and scalability. -
12
FlashAttention
Fast and memory-efficient exact attention
...The project provides implementations of FlashAttention, FlashAttention-2, and newer iterations optimized for modern GPU architectures such as NVIDIA Hopper and AMD accelerators. By improving both forward and backward pass efficiency, it enables training and inference of large language models with longer sequence lengths and higher throughput. The library integrates with PyTorch and supports various attention configurations, including causal masking, multi-query attention, and rotary embeddings. -
13
LMCache
Supercharge Your LLM with the Fastest KV Cache Layer
...Instead of rebuilding KV states for repeated or shared text segments, LMCache persists and retrieves them from multiple tiers—GPU memory, CPU DRAM, and local disk—then injects them into subsequent requests to reduce TTFT and increase throughput. Its design supports reuse beyond strict prefix matching and enables sharing across serving instances, improving efficiency under real multi-tenant traffic. The broader project includes examples, tests, a server component, and public posts describing cross-engine sharing and inter-GPU KV transfers. These capabilities aim to lower latency, cut GPU cycles, and stabilize performance for production workloads with overlapping prompts or retrieval-augmented contexts. ... -
14
FramePack
Lets make video diffusion practical
FramePack explores compact representations for sequences of image frames, targeting tasks where many near-duplicate frames carry redundant information. The idea is to “pack” frames by detecting shared structure and storing differences efficiently, which can accelerate training or inference on video-like data. By reducing I/O and memory bandwidth, datasets become lighter to load while models still see the essential temporal variation. The repository demonstrates both packing and unpacking... -
15
Lemonade
Lemonade helps users run local LLMs with the highest performance
...The repository highlights easy onboarding with downloads, docs, and a Discord for support, suggesting an active user community. Messaging centers on squeezing maximum throughput/latency from modern accelerators without users having to hand-tune kernels or flags. Releases further reinforce the “server” framing, pointing developers toward a service that can be integrated into apps and tools. -
16
TensorRT LLM
TensorRT LLM provides users with an easy-to-use Python API
...It provides a Python-based API built on top of PyTorch that allows developers to define, customize, and deploy LLMs efficiently across a variety of hardware configurations, from single GPUs to large multi-node clusters. The library focuses on maximizing throughput and minimizing latency through advanced techniques such as quantization, custom attention kernels, and optimized memory management strategies. It includes support for cutting-edge inference methods like speculative decoding and inflight batching, enabling real-time and large-scale AI applications. TensorRT-LLM integrates seamlessly with NVIDIA’s broader inference ecosystem, including Triton Inference Server and distributed deployment frameworks, making it suitable for production environments. -
17
slime LLM
slime is an LLM post-training framework for RL Scaling
slime is an open-source large language model (LLM) post-training framework developed to support reinforcement learning (RL)-based scaling and high-performance training workflows for advanced LLMs, blending training and rollout modules into an extensible system. It offers a flexible architecture that connects high-throughput training (e.g., via Megatron-LM) with a customizable data generation pipeline, enabling researchers and engineers to iterate on new RL training paradigms effectively. The framework is designed to support a wide range of training modes, allowing both synchronous and asynchronous RL workflows and programmable rollout interfaces that simplify experimentation with custom environments and reward signals. ... -
18
GLM-OCR
Accurate × Fast × Comprehensive
GLM-OCR is an open-source multimodal optical character recognition (OCR) model built on a GLM-V encoder–decoder foundation that brings robust, accurate document understanding to complex real-world layouts and modalities. Designed to handle text recognition, table parsing, formula extraction, and general information retrieval from documents containing mixed content, GLM-OCR excels across major benchmarks while remaining highly efficient with a relatively compact parameter size (~0.9B),... -
19
Loki Mode
Multi-agent autonomous startup system for Claude Code
...By supporting multiple AI providers (like Claude Code, OpenAI Codex CLI, and Google Gemini CLI), loki-mode dynamically selects and spawns only the needed agents for a given project, optimizing computational resources and task throughput. Its Reason-Act-Reflect-Verify (RARV) cycle with self-verification loops emphasizes quality and resilience, automating end-to-end development lifecycles. -
20
Pocket TTS
A TTS that fits in your CPU (and pocket)
...Because it is CPU-oriented, it fits well in server environments where GPU access is limited, in desktop apps, or in edge deployments where simplicity matters more than maximum throughput. It also emphasizes developer ergonomics, providing a straightforward API surface that can be integrated into pipelines, assistants, accessibility tools, or batch generation scripts. -
21
FastDeploy
High-performance Inference and Deployment Toolkit for LLMs and VLMs
...The platform enables developers to deploy trained models quickly using optimized inference pipelines that support GPUs, specialized AI accelerators, and other hardware architectures. FastDeploy includes advanced acceleration technologies such as speculative decoding, multi-token prediction, and efficient KV cache management to improve throughput and latency during inference. It also offers compatibility with OpenAI-style APIs and vLLM-like interfaces, allowing developers to integrate deployed models easily into existing applications and services. -
22
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... -
23
DFlash
Block Diffusion for Ultra-Fast Speculative Decoding
DFlash is an open-source framework for ultra-fast speculative decoding using a lightweight block diffusion model to draft text in parallel with a target large language model, dramatically improving inference speed without sacrificing generation quality. It acts as a “drafter” that proposes likely continuations which the main model then verifies, enabling significant throughput gains compared to traditional autoregressive decoding methods that generate token by token. This approach has been shown to deliver lossless acceleration on models like Qwen3-8B by combining block diffusion techniques with efficient batching, making it ideal for applications where latency and cost matter. The project includes support for multiple draft models, example integration code, and scripts to benchmark performance, and it is structured to work with popular model serving stacks like SGLang and the Hugging Face Transformers ecosystem. -
24
fairseq2
FAIR Sequence Modeling Toolkit 2
fairseq2 is a modern, modular sequence modeling framework developed by Meta AI Research as a complete redesign of the original fairseq library. Built from the ground up for scalability, composability, and research flexibility, fairseq2 supports a broad range of language, speech, and multimodal content generation tasks, including instruction fine-tuning, reinforcement learning from human feedback (RLHF), and large-scale multilingual modeling. Unlike the original fairseq—which evolved into a... -
25
Modular Platform
The Modular Platform (includes MAX & Mojo)
...It is closely associated with the Mojo programming language and related tooling that aims to combine Python usability with systems-level performance. Modular’s ecosystem is designed to simplify deployment of AI workloads across heterogeneous hardware while maximizing throughput. The repository reflects an effort to modernize the AI development pipeline from compilation to runtime execution. Overall, Modular represents an ambitious attempt to unify performance engineering and developer ergonomics for large-scale AI systems.
