msharmavikram Goto Github PK

carml

CarML (Cognitive ARtifacts for Machine Learning) is an open source distributed platform to easily deploy and benchmark machine learning (ML) and deep learning (DL) frameworks and models across hardware infrastructures, within a common interface. CarML allows ML/DL developers to publish and evaluate their models, users to experiment with published models, and system architects to profile end-to-end workflows to inform system designs.

dgl

Python package built to ease deep learning on graph, on top of existing DL frameworks.

dlrm

An implementation of a deep learning recommendation model (DLRM)

dragon-1

A host-based framework that transparently extends the GPU addressable global memory space beyond the host memory using NVM-backed data pointers

ece511_assignments

ensemblemlbp

EnsembleMLBP is a ECE511 course project created by Vikram, Umur and Andrew. Here we explore ensemble+ML based branch predictor. Initially we analyse tournamentBP and provide it loopholes. In second half we discuss RNN-LSTM and reinforcement learning schemes for BP. More details of the project will be added soon. The project is in early stage of development.

faiss

This is a forked version of FAISS that supports PowerPC ISA. (Tested in Power 8 with 4 NVIDIA GPU). HNSW is not supported as the code migration to Power ISA vectors yet to be done.

galois

Galois: C++ library for multi-core and multi-node parallelization

gem5_experiments

Trying out different experiments in gem5

llm.c

LLM training in simple, raw C/CUDA

memxct-gpu-nonmpi-1

Non-MPI version for single GPU

mlsync

Sync your ML data with your favorite productivity tools!

msharmavikram.github.io

nn_dataflow

Modified version of the "Explore the energy-efficient dataflow scheduling for neural networks. "

nvvl

This is a forked repository from Nvidia/nvvl. This fork has lower resolution (240p, 360p) enabled for data decoding.

pytorch-cifar

This is a forked version from kuangliu/pytorch-cifar with integration of torchgpipe. It is currently non-functional with more than one GPU.

rai

The RAI client allows one to interact with a cluster of machine to submit and evaluate code. RAI is a scalable job submission system designed for diverse workloads. RAI’s design addresses challenges of scalability, configurability, security, and cost in delivering a flexible programming environments.

reference

Reference implementations of MLPerf benchmarks

scanner

Efficient video analysis at scale

simplegpuhashtable

A simple GPU hash table implemented in CUDA using lock free techniques

timesfm

TimesFM (Time Series Foundation Model) is a pretrained time-series foundation model developed by Google Research for time-series forecasting.

tutorials

This repository contains some good tutorials to read when you get time! A must know skills for an Engineer.

two_branch_network

website