Learn how to semantically split vision datasets using foundation model embeddings on AMD GPUs to reduce leakage and improve evaluation.

Quantize Kimi-K2.5 to W4A8 and W8A8 using AMD Quark and serve on MI325X with FlyDSL and AITER for further inference acceleration.

We show that the MI355X delivers better performance than the B200 for ComfyUI after enabling PyTorch attention for gfx950.

Use ATOM as an out-of-tree vLLM plugin to keep vLLM compatibility while enabling AMD-optimized attention, model execution, and multi-model support including Kimi-K2.5.

Run Street Gaussians on AMD Instinct MI300: migrate to latest gsplat, install on ROCm, and render dynamic street scenes.

Explore a new MoE architecture designed for native computation-communication overlap, enabling efficient distributed execution.

Explore how TraceLens automates profiler trace analysis to pinpoint bottlenecks and optimize AI workloads.

Learn how to use the Primus projection tool to estimate memory and performance for large-scale LLM training on AMD Instinct™ accelerator platforms.

Hands-on, reproducible guide to train and run Eruku on LUMI supercomputer, powered by AMD Instinct MI250X GPUs.

A practical guide to installing and using FlyDSL nightly wheels on ROCm for fast, Python-native GPU kernel development

This blog explores a new training-free loosely speculative decoding method, that can accept mismatches that are semantically valid and speedup original SPD method.

Profiling tools