Last released Aug 8, 2026
Memory-efficient PyTorch optimizer with block-wise adaptive log-space quantization, fused CUDA kernels, and APOLLO/CAME paths
Supported by
