Skip to content
View PranavShashidhara's full-sized avatar
2 followers · 2 following
  • MulticoreWare

Achievements

Achievement: Pull SharkAchievement: Quickdraw

Achievements

Achievement: Pull SharkAchievement: Quickdraw

Block or report PranavShashidhara

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
PranavShashidhara/README.md

Pranav Shashidhara

GPU performance and ML systems. MS in Data Science, University of Maryland (2026).

I write and benchmark GPU code, and I study how inference workloads use the memory hierarchy. Before graduate school I spent three years at Oracle building data pipelines and ML-ready feature processing at scale.

Focus

  • CUDA kernels and roofline analysis (memory-bound vs. compute-bound behavior)
  • LLM inference: KV-cache management, batching, quantization
  • Open-source work on ROCm support in vLLM (in progress)

LinkedIn

Pinned Loading

  1. cuda-memory-hierarchy-benchmarks cuda-memory-hierarchy-benchmarks Public

    Roofline-driven CUDA microbenchmarking framework for studying memory-bound vs compute-bound transitions on edge GPUs (Jetson Orin Nano, SM87), comparing hand-written GEMM kernels, Tensor Core imple…

    Cuda

  2. triton_kernel_benchmark triton_kernel_benchmark Public

    Hand-tuned CUDA GEMM and fused attention kernels ported to Triton, with a systematic study of how the autotuner's configs compare to manual pipeline scheduling (reaching 80.1% of cuBLAS at N=4096).

    Jupyter Notebook

  3. kvserve kvserve Public

    Paged KV-cache LLM inference engine with dynamic batching and roofline-guided performance analysis (A100, FP16)

    Jupyter Notebook

  4. llm-orchestration-stack llm-orchestration-stack Public

    End-to-end platform for fine-tuning and deploying Llama-3.1 8B-Instruct on SQL generation using QLoRA, 4-bit quantization, and multi-GPU inference. Focuses on memory-efficient, high-throughput LLM …

    Jupyter Notebook

  5. HPC-and-CUDA-kernels HPC-and-CUDA-kernels Public

    This project is to learn HPC and build projects on it.

    Python

  6. modular modular Public

    Forked from modular/modular

    The Modular Platform (includes MAX & Mojo)

    Mojo