SOTA low-bit LLM quantization (INT8/FP8/MXFP8/INT4/MXFP4/NVFP4) & sparsity; leading model compression techniques on PyTorch, TensorFlow, and ONNX Runtime

[EMNLP 2024 & AAAI 2026] A powerful toolkit for compressing large models including LLMs, VLMs, and video generative models.

SNDR Core Engine (Genesis) — vLLM runtime patch-overlay for Qwen3.6 + Gemma4 on consumer NVIDIA (Ampere sm_86, 2× A5000/3090). Qwen3.6-35B-A3B FP8 ~240 tok/s, 27B-int4 hybrid GDN+Mamba, Gemma4 26B/31B AWQ, 256K ctx. 321 patches: TurboQuant k8v4 KV, MTP/DFlash spec-decode, FULL cudagraph, hybrid GDN. vLLM pin dev424 + Control Center GUI.

vLLM Qwen 3.6-27B (AWQ-INT4) + DFlash speculative decoding on AMD Strix Halo (gfx1151 iGPU, 128 GB UMA, ROCm 7.13). 24.8 t/s single-stream, vision, tool calling, 256K context, OpenAI-compatible, Docker. Matches DGX Spark FP8+DFlash+MTP at a third of the cost. No CUDA.

FastAPI wrapper around original Vibevoice 1.5B and 7B models, with support for AWQ4 quant

Native Windows build of vLLM 0.24.0 - no WSL, no Docker. Python 3.13 + CUDA 12.8 + PyTorch 2.11 cu128 for RTX 30/40/50-series, pre-built wheel, Windows patchset, 10 KV-cache compression dtypes, OpenAI API server fixes, Rust frontend, and Rust tool parser support.

An OpenAI Compatible API which integrates LLM, Embedding and Reranker. 一个集成 LLM、Embedding 和 Reranker 的 OpenAI 兼容 API

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

A light, transparent, and modular inference & quantization engine for studying LLMs.

Compress Any LLM Up to 6x in One Command. Unified CLI for GGUF, GPTQ, and AWQ quantization.

🧗‍♂️ harleyszhang 的个人博客

本来叫 nano 的，后来发现装不下 Qwen3.5，就改名叫 big 了

Research Test: REAP expert pruning + AWQ quantization of Qwen3-Coder-Next MoE model

Quantize LLM using AWQ

Quantization quality analyzer - benchmark GGUF/GPTQ/AWQ quantization accuracy.

Dockerized vLLM serving for Kimi-Linear-48B-A3B (AWQ-4bit), from 128K to 1M context.

[ICLR2026] When Reasoning Meets Compression: Understanding the Effects of LLMs Compression on Large Reasoning Models. Support interpretation of Qwen, Llama, etc.

Effortlessly quantize, benchmark, and publish Hugging Face models with cross-platform support for CPU/GPU. Reduce model size by 75% while maintaining performance.

Run Qwen 3.6-27B AWQ-INT4 models with DFlash speculative decoding on AMD Strix Halo hardware using vLLM for high-throughput inference.

White paper & reproducible benchmark suite for LLM inference optimization on AMD MI300X using ROCm 6.1