brrr cuda dtypes float8 inference llama mx offloading optimizer pytorch quantization sparsity training transformer

bluefield cloud datacenter dpu filesystem framework linux offloading smartnic storage virtio-fs virtualization