Tools tagged with "Speculative decoding"

SGLang is a self-hosted inference framework for teams serving AI models to applications. It supports production workloads on a single GPU or across distributed clusters, including AI agents and reinforcement learning rollouts.

vLLM serves language models on hardware you control. It's for developers and teams building model APIs that need to handle concurrent requests while using memory and compute efficiently.

vLLM Metal lets developers serve language models on Apple Silicon Macs through vLLM. It's for people who want to use their Mac's hardware for local AI inference while keeping vLLM's API server and request scheduling.

Lucebox is a local AI workstation for developers and teams who want to run models on their own hardware. Its inference engine is also available separately for supported AMD and NVIDIA machines.

MTPLX is a native Mac app, command-line tool and local LLM server for people running chat or coding models on Apple Silicon. It uses a model's own multi-token prediction heads to accelerate generation without a separate draft model.

FastLLM is an open-source inference engine for people running large language models on their own machines or self-hosted servers. Its main draw is hardware flexibility: large mixture-of-experts (MoE) models can spread their weights across GPU memory, system RAM and disk when they don't fit entirely on a graphics card.

Sidekick is a native macOS app for people who want AI answers grounded in their documents and research materials. Its built-in llama.cpp engine runs GGUF models such as Qwen3.5 without a separate model runner.

TabbyAPI is a self-hosted LLM API server for people who want to run ExLlamaV3 models on their own hardware. It’s built for personal projects and small groups; its maintainers don’t recommend it for production servers.

TensorSharp runs local AI models for .NET developers and people who want chat, media generation and agent tools on their own hardware. You can embed it in a C# application or use its CLI, browser interface or companion app, TensorAgent.

Llama Desktop runs language models on your computer for private chat and use in other apps. It's for people who want a local ChatGPT alternative and developers connecting coding agents, editors or scripts to their own models.

vllm-mlx is a local LLM inference server for developers who want to connect AI apps and coding assistants to models running on an Apple Silicon Mac. One process exposes both OpenAI-compatible and Anthropic-compatible APIs. It's open source under Apache 2.0 and runs on macOS with at least 8 GB of unified memory.

MiniCPM is a family of compact language models for developers building local assistants and coding agents. It targets on-device workloads and hardware with limited computing resources.

NeMo Automodel is a Python library for researchers and developers who want to train or adapt AI models on their own GPU workstations or clusters. Its PyTorch-based training works with Hugging Face models without converting checkpoints.

Deltafin runs full Kimi K3 on a single device for researchers and developers exploring large models on their own hardware. It preserves the original expert weights and complete expert bank, but responses are slow and long conversations take substantially longer than short completions.

Pooled runs a single open model across laptops, desktops and phones in browser tabs. It lets people combine their devices' memory to use a model that wouldn't fit on one device, for shared chat or coding tasks.

OlliteRT lets you use an Android phone as a private AI server for apps on your local network. It's for people who want to run models locally for chat clients or home automation without a cloud inference service.

Sonar serves language and multimodal models for developers and teams running AI applications on their own hardware. It builds on vLLM and works with Hugging Face-compatible models.

TensorRT-LLM is an inference library for developers serving LLMs on their own NVIDIA GPUs. It supports application-level inference and server deployments, with performance optimizations for workloads spread across one or several machines.

llama.cpp is a local LLM inference engine for developers building AI applications or running models on hardware they control. It supports language and vision-language models, with command-line tools and a server API for using the engine directly or connecting other software.

Soup is a local LLM fine-tuning tool for developers and teams adapting models to their own datasets. It brings training, evaluation and export into one workflow that runs on your own hardware without requiring a cloud service.

llama-cpp-python is for developers adding local LLM inference to Python applications or connecting existing API clients to models on their own hardware. It brings llama.cpp into a Python library and a self-hosted server.

FastDeploy is a self-hosted inference toolkit for teams deploying language and vision-language models on Linux servers. Built on PaddlePaddle, it supports offline inference and online serving.

SpecForge is an open-source framework for teams training draft models to speed up LLM inference in SGLang. It suits developers who want to train on their own hardware and use the resulting drafts directly with SGLang serving.

OpenVINO Model Server is a self-hosted inference server for developers and teams serving AI models on their own machines or servers. It uses OpenVINO to run workloads on x86 CPUs, including Xeon, with acceleration for Intel integrated and discrete GPUs and NPUs. It's open source under Apache 2.0.

ExLlamaV2 and V3 are local LLM inference libraries for developers building AI applications or hosting models on consumer GPUs. V2 is archived for now and isn't currently maintained; development continues in V3.

Atomic Chat is a local LLM app for people who want to chat with models on their own device or use them with coding assistants and AI agents. It began as a fork of Jan by Menlo Research.

MLX VLM is a Python package for developers and researchers who want to run multimodal models on their own Mac. It supports local inference and fine-tuning through MLX.

Slotstream runs Qwen3.8-Flash-Next on Apple Silicon Macs with too little RAM to hold the whole model. It's aimed at Macs with 16 to 64 GB of memory, for people who want local chat, answers about pictures or a backend for coding assistants.

DwarfStar is a local LLM inference engine for people who want to run large DeepSeek, GLM and Qwen models on their own computers or servers. It focuses on selected models and hardware setups, and requires the project's own GGUF files rather than accepting arbitrary GGUF downloads.

NeMo RL is NVIDIA's Python library for researchers and engineering teams adapting language and multimodal models on their own hardware. It supports experiments on a GPU workstation and distributed training across multiple nodes.

FlashInfer is a GPU kernel library for developers building LLM inference systems on their own hardware or servers. It supplies the attention, matrix multiplication and mixture-of-experts operations used by model serving software.

Intel IPEX-LLM is a library for developers running or fine-tuning language models on Intel hardware. The project is archived and no longer maintained, and Intel has identified known security issues in the code.