Paris

Mamy Ratsimbazafy

Elite
@mratsim

Ethereum Blockchain R&D ✦ ZK Cryptography Engineering ✦ Data Scientist ✦ High Performance, Parallel, Scientific and GPU computing

Arraymancer. A fast, ergonomic and portable tensor library in Nim with a deep learning focus for CPU, GPU and embedded devices via OpenMP, Cuda and OpenCL backends

1.4k

weave. A state-of-the-art multithreading runtime: message-passing based, fast, scalable, ultra-low overhead

585

constantine. Constantine: modular, high-performance, zero-dependency cryptography stack for verifiable computation, proof systems and blockchain protocols.

496

Amazon-Forest-Computer-Vision. Amazon Forest Computer Vision: Satellite Image tagging code using PyTorch / Keras with lots of PyTorch tricks

371

laser. The HPC toolbox: fused matrix multiplication, convolution, data-parallel strided tensor primitives, OpenMP facilities, SIMD, JIT Assembler, CPU detection, state-of-the-art vectorized BLAS for floats and integers

295

trace-of-radiance. An educational raytracer

105

Synthesis. Synthesis is a compiletime, procedure-based, low-overhead, no-allocation, state-machine generator optimized for communicating processes and threads

99

Arch-Data-Science. Archlinux PKGBUILDs for Data Science, Machine Learning, Deep Learning, NLP and Computer Vision

97

home-credit-default-risk. Default risk prediction for Home Credit competition - Fast, scalable and maintainable SQL-based feature engineering pipeline

81

McKinsey-SmartCities-Traffic-Prediction. Adventure into using multi attention recurrent neural networks for time-series (city traffic) for the 2017-11-18 McKinsey IronMan (24h non-stop) prediction challenge

62

loop-fusion. Loop efficiently over a variadic number of containers

41

rustygo. A Go bot written in Rust with Monte Carlo Tree Search & a brain dump on go bot optimizations

28

photon-jit. A low-overhead JIT assembler for Nim

20

number-theory. Give integers super-powers!

20

agent-smith. Reinforcement Learning Agents & Algorithms exploration

19

weave-io. A latency and fairness optimized threadpool. Tuned for async IO and decent for compute tasks.

19

Apartment-Interest-Prediction. Predict people interest in renting specific NYC apartments. The challenge combines structured data, geolocalization, time data, free text and images.

18

glyph. WIP Snes emulator

17

chirp8. A Chip-8 emulator written in Nim

16

weave-io-research. HTML

15

monocle. A lightweight interactive data visualization library

14

humpback-whale-identification. Kaggle Humpback whale identification: 2xGPU Data augmentation + FP16 mixed precision training

14

nim-rmad. Autograd (backpropagation, reverse-mode auto differentiation) in Nim

12

meilleur-data-scientist-france-2018. My solution for the competition "Le meilleur data scientist de France 2018" (Best Data Scientist of France 2018)

12

talks. A repo of my talks

11

finite-fields. Experimental finite field primitives (and maybe more)

11

blocksmith. Blocksmith: Safe & fast interfacing network blocks and attestations with Casper FFG finality & LMD Ghost fork choice

10

haskell-numbertheory. Exploration of primes, factorization and number theory through haskell

10

golem-prime. A Go bot written in Nim

9

compute-graph-optim. Experiments in compute graph optimisations and ML and HPC compilers frontend

9

jitterland. Experiments in lexing, parsing, interpreting, VM and JITs

9

quantizers. LLM quantization scripts

9

nim-project-euler. Solving Project Euler math problems through Nim language (https://projecteuler.net)

8

nim-clblast. A Nim wrapper for CLBlast, a tuned OpenCL BLAS library

7

nim-isl. A wrapper for isl - the Integer Set Library for manipulating sets and relations of integer points bounded by linear constraints.

7

hydra. Parametric Integer Linear Programming Solver and Polyhedral compilation backend

7

ReactFincal. iOS/Android Financial Calculator written in React Native + Typescript

5

MachineLearning_Kaggle. Kaggle Data Science & Machine Learning competitions - using Python or Julia

3

batch-reencode. Batch reencode videos to AV1 using ffmpeg

3

nim-julia-challenge. Nim solution to https://nextjournal.com/sdanisch/the-julia-challenge

2

ff-arm64. Experiments in accelerating finite fields on ARM64

2
41
Apply