SANA: Efficient High-Resolution Image Synthesis with Linear Diffusion Transformer

A SOTA quantization algorithm for high-accuracy low-bit LLM inference, seamlessly optimized for CPU/XPU/CUDA, with multi-datatype support and full compatibility with vLLM, SGLang, and Transformers.

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

记录量化LLM中的总结。

A production-ready Docker setup for ComfyUI that unlocks the full potential of NVIDIA Blackwell GPUs (RTX 50 series) through 4-bit quantization with NVFP4.

[ACL 2026 Main] Code for the paper "ARCQuant: Boosting NVFP4 Quantization with Augmented Residual Channels for LLMs"

DFlash vLLM for DGX Spark — Plug & Play Block-Diffusion Speculative Decoding

LLM fine-tuning with LoRA + NVFP4/MXFP8 on NVIDIA DGX Spark (Blackwell GB10)

Easy recipes to speed up latency of Flux, QwenImage, and LTX-2 with NVFP4 and MXFP8 on Blackwell.

DGX Spark research and tests - containers, benchmarks, and investigation notes for running models on GB10 (SM 12.1)

Blackwell-optimized llama.cpp Docker image – works on all NVIDIA GPUs, but tuned for RTX 50 series. Built from scratch with CUDA 12.8, sm_120, NVFP4-ready. 250+ tok/s on 4B F16. Includes llama-chat script.

(Experimental) A high-throughput and memory-efficient inference and serving engine for LLMs optimized for GB10 homelabs

NVFP4 AWQ Full quantization of SuperGemma4-26B-Abliterated-Multimodal for Blackwell GPUs — pre-built vLLM container + patches included

Guide for serving fine-tuned Qwen3.5-27B (dense, NVFP4) on DGX Spark via native vLLM. Includes critical config fixes for modelopt export_hf_checkpoint() that prevent silent FP32 dequantization.

WanGP v10.61 RTX 50XX Pinokio Upgrade - Python 3.11, PyTorch 2.10, CUDA 13.0, NVFP4 kernels. Copy files to wan.git folder, Reset + Install + Update

🔧 Fine-tune large language models efficiently on NVIDIA DGX Spark with LoRA adapters and optimized quantization for high performance.

NVFP4 inference on Blackwell GeForce (RTX 5090/5080/5070 Ti/RTX PRO 6000) — SM120 patches for vLLM + FlashInfer + CUTLASS. 175 tok/s on Qwen3.6-35B MoE.

EAGLE E4B speculative decoding drafter for Gemma 4 31B DECKARD HERETIC Uncensored NVFP4 — optimized for NVIDIA DGX Spark

My solutions for the GPUMODE NVFP4 competition, written in CuTe DSL