#cuda #nccl #nvidia

#attention #grouped-query-attention #linear-attention #position-embedding #RMSNorm #RoPE #transformer