ROCm still fights for territory AMD already won
Someone built a tool to make AMD GPUs run local LLMs without the usual driver nightmares. The benchmarks show Vulkan beating HIP by 20 percent.
Read the note 1 min read
From the notebook
gpu.
4 notes on this topic.
Someone built a formal verifier for LLM-written CUDA code
A new tool checks if ChatGPT's GPU kernels are actually correct before you run them in production.
Read the note 1 min read
Training LLMs on an 8GB GPU is now a weekend project
Someone just released a minimal codebase for supervised fine-tuning, direct preference optimisation, and group relative policy optimisation on consumer hardware.
Read the note 1 min read
Norway built an LLM training cluster on 2 PB of Huawei flash
The Norwegian University of Science and Technology is running LLM workloads on Chinese storage hardware. The performance numbers are interesting.
Read the note 1 min read