Recent Posts#

Empowering Developers to Build a Robust PyTorch Ecosystem on AMD ROCm™ with Better Insights and Monitoring
At AMD, the PyTorch ecosystem team is committed to delivering an exceptional out-of-the-box experience for developers. Over the past year, the team has made significant progress in expanding PyTorch ecosystem support, improving CI test coverage across a wider range of GPU architectures, enhancing training and inference capabilities, streamlining the developer experience, introducing new functionality and performance optimizations, and strengthening quality monitoring. This blog showcases our ongoing efforts to build a robust PyTorch ecosystem on AMD ROCm™ Software, including the production-readiness of PyTorch across N-1, N, and N+1 releases aligned with ROCm versions. We also introduce the AI SoftWare Heads-Up Dashboard (AISWHUD), a powerful new tool that provides deep insights into the health and performance of the PyTorch ecosystem on ROCm, empowering developers with greater visibility and control.

ROCm 7.9 Technology Preview: ROCm Core SDK and TheRock Build System
Introduce ROCm Core SDK, and learn to install and build ROCm components easily using TheRock.

Kimi-K2-Instruct: Enhanced Out-of-the-Box Performance on AMD Instinct MI355 Series GPUs
Learn how AMD Instinct MI355 Series GPUs deliver competitive Kimi-K2 inference with faster TTFT, lower latency, and strong throughput.

Gumiho: A New Paradigm for Speculative Decoding — Earlier Tokens in a Draft Sequence Matter More
Gumiho boosts LLM inference with early-token accuracy, blending serial + parallel decoding for speed, accuracy, and ROCm-optimized deployment.

GEMM Tuning within hipBLASLt– Part 2
Learn how to use hipblaslt-bench for offline GEMM tuning in hipBLASLt—benchmark, save, and apply custom-tuned kernels at runtime.

Announcing MONAI 1.0.0 for AMD ROCm: Breakthrough AI Acceleration for Medical Imaging Models on AMD Instinct™ GPUs
Learn how to use Medical Open Network for Artificial Intelligence (MONAI) 1.0 on ROCm, with examples and demonstrations.

Medical Imaging on MI300X: Optimized SwinUNETR for Tumor Detection
Learn how to setup, run and optimize SwinUNETR on AMD MI300X GPUs for fast medical imaging 3D segmentation of tumors using fast, large ROIs.

Optimizing FP4 Mixed-Precision Inference with Petit on AMD Instinct MI250 and MI300 GPUs: A Developer’s Perspective
Learn how FP4 mixed-precision on AMD GPUs boosts inference speed and integrates seamlessly with SGLang.

Optimizing Drug Discovery Tools on AMD MI300s Part 2: 3D Molecular Generation with SemlaFlow
Learn how to set up, run, and optimize SemlaFlow, a molecular generation tool, on AMD MI300X GPUs for faster drug discovery workflows

Elevating 3D Scene Rendering with GSplat
ROCm Port of GSplat - GPU accelerated rasterization of Gaussian splatting

From Ingestion to Inference: RAG Pipelines on AMD GPUs
Build a RAG enhanced GenAI application that improves the quality of model responses by incorporating data that is missing in the model training data.

Enabling FlashInfer on ROCm for Accelerated LLM Serving
FlashInfer is an open-source library for accelerating LLM serving that is now supported by ROCm.