00:00
2026-07-13
rocm.blogs.amd.com
artificial-intelligence
Serving NVFP4 Models on AMD Instinctβ’ MI355 Accelerators
AMD announced an emulation-based serving pipeline integrated into vLLM that enables AMD Instinct MI355 accelerators to load and serve standard NVFP4 quantized checkpoints directly, despite the CDNA4 aβ¦