nvfp4 开源项目推荐（329 个）

NVlabs /Sana

SANA：使用线性扩散Transformer的高效高分辨率图像合成

NVlabs /LongLive

Long Video Gen / World Model / World Action Model (Embodied) Infrastructure

intel /auto-round

一种面向高精度低比特 LLM 推理的 SOTA 量化算法，无缝优化 CPU/XPU/CUDA，支持多种数据类型，并完全兼容 vLLM、SGLang 和 Transformers。

NVIDIA /cudnn-frontend

cuDNN 前端是 NVIDIA 现代开源的 cuDNN 库入口，以及一个不断增长的高性能开源内核集合。

Avarok-Cybersecurity /atlas

纯锈推断引擎

AEON-7 /Qwen3.6-27B-AEON-Ultimate-Uncensored-DFlash

Qwen 3.6 - 27 B完全未经审查、功能增强的版本。NVFP 4 + z-Lab DFlash推测性解码（n=12）在统一的ghcr.io/aeon-7/aeon-vllm-ultimate:latest容器上进行，针对DGX Spark上的长上下文草稿接受进行了调整。6种HF变体（BF 16/NVFP 4/STP/MTP-XS）、docker-compose和QuickStart。

hasso5703 /alzeo

dgx-spark-qwen38 is an open-source installer that turns an NVIDIA DGX Spark, or any GB10 machine (ASUS, Dell, HP, Lenovo, MSI, GIGABYTE, Acer), into a private AI server in one command: measured model lanes, OpenAI and Anthropic APIs, and a local cockpit.

sergqwer /strata-nvfp4

Strata fork: Qwen3.8-Flash-Next 125B MoE in NVFP4 on one RTX 20-50 card (12 GB+) and 64 GB of RAM or more. Setup installs our GPTQ quants (huihui abliterated, OrcaRouter) from Hugging Face. CPU experts on AVX-512 or AVX2 (+AVX-VNNI), all experts in RAM from 92 GB, a low-RAM mode below, vision, W4A8 prompts. Windows / CUDA 13.

jpezzulli /sglang-rtxpro6000

Optimized SGLang runtime for Qwen3.8-27B FP8 with DFlash2 and Qwen3.8 Flash-Next NVFP4 with FR-Spec on one NVIDIA RTX PRO 6000 Blackwell 96 GB GPU (SM120): 524K context, HiCache and NIXL.

dnv2003 /v100-skinny

Hand-written NVFP4 W4A16 CUDA kernels for Volta

0xSero /glm-5.2-sm120

GLM-5.2-NVFP4-REAP-469B，服务于SM120（4×RTX PRO 6000 Blackwell）——一命令vLLM启动方案，250K上下文，DeepSeek Sparse Attention + MTP推测解码

BenChaliah /NVFP4-on-4090-vLLM

AdaLLM is an NVFP4-first inference runtime for Ada Lovelace (RTX 4090) with FP8 KV cache and custom decode kernels. This repo targets NVFP4 weights and keeps the entire decode path in FP8

ValerioDolci /capy-flash

Qwen3.8-Flash-Next (MoE) and Qwen3.8-27B on one or two 16 GB RTX 50-series GPUs: experts offloaded to CPU, two-GPU tensor parallel. 64 GB of RAM is enough for Flash-Next. Formerly ninfer-tp2.

AEON-7 /comfyui-aeon-spark

适用于NVIDIA DGX Spark（GB 10/Blackwell/sm_121a）的前沿舒适UI。CUDA 13 + SageAttention v3（sm_121a）+NVFP 4 + 14个自定义节点包+ Flux 2 Dev / LTX 2.3 22 B/ ACE-Step v1.5 XL Turbo预捆绑有已创建的文本编码器路径。

headpiece747 /ninfer-5090-windows

Native Windows port of NInfer engine for RTX 5090. Features Qwen3.8-27B with QUASAR and NInfer models, MTP/DFlash2 with vision and 262,144 context

Anemll /dspark-vllm-gx10

Two-node DGX Spark/ASUS GX10 DeepSeek V4 Flash DSpark NVFP4 port for vLLM 0.25, with live dashboard and reproducible deployment.

taishan1994 /LLM-Quantization

记录量化LLM中的总结。

kekzl /imp

该项目在 GitHub 上没有填写描述

0xBakeer /TandemLLM

An environment where open models work as agents: a first-level harness with routing, skills, memory and sandboxed tools, on an engine that drafts and verifies its own tokens.

gpillon /ignis

OpenAI-compatible inference server for Qwen3.8 on one RTX 5090 / RTX PRO 6000: the 27B at ~170 tok/s (peaks ~370), Flash-Next (125B MoE) at ~100 tok/s, built for local coding agents. Rust + CUDA, Windows and Linux.