Prashant Kumar
Member Of Technical Staff at Cohere
London, England, United Kingdom
Summary
🤩
Rockstar🎓
Top SchoolPrashant Kumar is an experienced compiler and ML systems engineer with nine years building high-performance code generation pipelines for CPU and GPU backends. Currently a Member of Technical Staff at Cohere, he has deep open-source pedigree contributing to LLVM, Torch-MLIR, and IREE—projects that sit at the intersection of PyTorch and MLIR and power production ML runtimes. His work includes implementing operator lowerings, vectorization fixes, polynomial approximations for math ops, and quantized matrix-multiplication optimizations, demonstrating both low-level numerical care and compiler-scale design. Previously at AMD and Nod.ai he helped bootstrap MLIR backends (notably initial torch-mlir efforts) and improved developer tooling and inference APIs for SHARK/IREE. Based in London, he pairs academic training from IIT Delhi with practical DevOps and backend experience, making him effective across research-to-production paths. A less obvious strength is his attention to subtle numerical and masking semantics, which has improved correctness as well as performance in compiler transformations.
9 years of coding experience
4 years of employment as a software developer
Indian Institute of Technology Delhi (IIT Delhi)Bachelor of Engineering (BE) Electronics Engineering, Bachelor of Engineering (BE) Electronics Engineering at Rashtrasant Tukadoji Maharaj Nagpur University, NagpurHigh School Computer Science, High School Computer Science at Nazareth academy
English, Hindi
Github Skills (29)
pytorch10
c-language10
python10
operation10
scripting10
tensorrt10
machine-learning10
vectorization10
mlr10
script10
compiler-compiler10
tensorflow10
compiler-optimization10
compiler10
sh10
Programming languages (5)
Github contributions (5)
nod-ai/SHARK-Studio
Mar 2022 - Jan 2023
SHARK Studio -- Web UI for SHARK+IREE High Performance Machine Learning Distribution
Role in this project:
Back-end & DevOps Engineer
Back-end & DevOps EngineerContributions:149 reviews, 143 commits, 234 PRs in 10 months
Contributions summary:Prashant primarily contributed to setting up and improving the development environment. They added a script (`set_dep_pypaths.sh`) to configure the `PYTHONPATH` for dependencies, particularly those related to IREE and Torch-MLIR. Additionally, they introduced a basic inference API using the `shark_runner` with the `torch_mlir_utils` and `iree_utils` modules. This suggests a focus on making it easier to run machine learning models.
pytorchcudaamdheterogeneousdeep-learning
llvm/torch-mlir
Oct 2021 - Dec 2022
The Torch-MLIR project aims to provide first class support from the PyTorch ecosystem to the MLIR ecosystem.
Role in this project:Back-end Developer & ML Engineer
Back-end Developer & ML EngineerContributions:264 reviews, 54 commits, 110 PRs in 1 year 2 months
Contributions summary:Prashant primarily focused on implementing and integrating core functionality within the Torch-MLIR project, specifically aimed at bridging the gap between PyTorch and the MLIR ecosystem. Their contributions centered on adding lowering implementations for PyTorch operators like `aten.matmul`, `aten.Int.Tensor`, `aten.log`, `aten.log_softmax`, `aten.fill.Scalar`, `aten.nll_loss_forward`, `aten.nll_loss_backward`, `aten.gt.Tensor`, `aten.bernoulli`, `aten.hardswish`, and `aten.silu`. They also worked on refactoring existing code related to decomposition of complex ops, as well as supporting the return of element types.
pytorchmlirtorchcompilerecosystem
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.