fp32

Lightweight C inference for Qwen3 GGUF. Multiturn prefix caching & batch processing.

Fast SGEMM emulation on Tensor Cores

An stress and benchmark utility for NVIDIA GPUs. Measures performance across various precisions (FP64, FP32, TF32, FP16, INT8) and monitors real-time vitals like power, temperature, and clock speeds.

Converts a floating-point number or hexadecimal representation of a floating-point numbers into various formats and displays them into binary/hexadecimal.

The floats package provides types for handling multi-precision floating-point numbers.

RTL implementation of a 32-bit IEEE-754 single-precision floating-point multiplier in SystemVerilog, featuring SVA-based property checking

A reproducible GPU benchmarking lab that compares FP16 vs FP32 training on MNIST using PyTorch, CuPy, and Nsight profiling tools. This project blends performance engineering with cinematic storytelling—featuring NVTX-tagged training loops, fused CuPy kernels, and a profiler-driven README that narrates the GPU’s inner workings frame by frame.

Strict-FP32 CUDA SGEMM with shape-dispatched SM80 cp.async kernels and reproducible A100 evidence

Neural Machine Translation Quantization: Optimizing Transformer Models for Efficient Deployment

Official code for the TMLR 2026 paper on BF16/FP16 divergence in LLM greedy decoding and selective FP32 lm_head recomputation.

🎬 Explore GPU training efficiency with FP32 vs FP16 in this modular lab, utilizing Tensor Core acceleration for deep learning insights.

Strict FP32 flash-style CUDA attention kernels for short-sequence NVIDIA workloads

TensorRT Inference Code by cpp

The goal of this reposotory is create a downcasting the Fashion MNIST model from FP32 to FP16 using pytorch

AI Performance Benchmark Tool. Unifies CPU, GPU (Metal), and NPU (Neural Engine) tests in GFLOPS and TOPS.

fp32/low-memory SFFT backend for phrosty — full 4088² Roman difference imaging in ~5 GB VRAM on an 8 GB consumer GPU, validated to 1e-5 of sky. Opt-in: SFFT_BACKEND=rfft