Greg Diamos

I did not set out to write a paper about small models. I had a data processing problem.

Large Language Models compress massive amounts of training data into their parameters. This compression is lossy but highly effective-billions of parameters can encode the essential patterns from terabytes of text. However, what's less obvious is that thi

This blog covers the performance of the MI300X GPU in the context of a bfloat16 GEMM benchmark.

This blog covers the performance of the MI300X GPU in the context of a memcpy peer benchmark.

This blog covers the performance of the MI300X GPU in the context of a memcpy benchmark.

Today we are excited to announce the release ScalarLM v0.5 (pronounced KRAY-lem)

ROCm (Radeon Open Compute) provides an open-source software foundation for GPU computing on AMD hardware. In this guide, we'll walk through the process of building ROCm-enabled containers for ScalarLM, enabling you to leverage AMD GPUs for large language

Introduction Large Language Models (LLMs) are transforming software development by excelling in tasks such as generating new code. However, their performance with complex legacy codebases has limitations. To bridge this gap, we adopt the ScalarLM unified

Background

In this blog I will cover detailed articles on high performance computing and machine learning research topics and open source software development. Stay tuned!