Pinned Loading

1. BitSqueeze BitSqueeze Public

   BitSqueeze is a tiny C library for compressing float32 tensors with GGML-style integer quantization (Q8_0, Q4_0, Q2_K, NF4, NVFP4), compact floating formats (FP4, MXFP4, NF4_DQ, FP8, MXFP8, FP16, B…

   C

2. numa-gpu-xfer-bench numa-gpu-xfer-bench Public

   This NUMA-aware benchmark provides latency and bandwidth measurements for CPU-GPU data transfers using libnuma for interleaved allocation, cudaMemcpyAsync for async operations, and concurrent multi…

   C++

3. numa-allocation numa-allocation Public

   This package provides a NUMA-aware CPU tensor allocation function for PyTorch. It allows you to allocate tensors on specific NUMA nodes or interleave them across nodes, with optional page-lock pinn…

   Python

4. liger-kernel-for-cpu-offloading-study liger-kernel-for-cpu-offloading-study Public

   This repository contains code, experiments, and a report for a study on the Liger Kernel and its application in memory-efficient training of Large Language Models (LLMs). The report analyzes the Li…

   Python 4