LinkedIn
GitHub
X / Twitter
Google ScholarAdvancing GPU Programming with the CUDA Tile IR Backend for Triton
A talk on the CUDA Tile IR backend for Triton.
SOFTWARE ENGINEERING · NVIDIA
Jhalak Patel
AI compilers. GPU programming.
Efficient deep learning inference.
I'm a software engineer at NVIDIA. My technical interests span how we express computation, compile it for GPUs, and run neural networks efficiently.
I've spoken about the CUDA Tile IR backend for Triton and quantization-aware inference with TensorRT. I'm interested in the practical details that connect an idea to an implementation: performance, numerical accuracy, and programming models.
Talks & presentations
Public recordings and presentation materials.
Toward INT8 Inference: Deploying Quantization-Aware Trained Networks using TensorRT
Quantization-aware training and an end-to-end workflow for deploying quantized networks with TensorRT.
Granted U.S. patent
Worth a read
Articles and ideas that caught my attention.
Together Compiler: Continual Compilation for AI inference
Qwenguard: the CPU inside an LLM checks its own tool calls
Technical resources
Background on the technologies discussed in my talks.