Assemble Attention.forward

Combine KV writes, varlen prefill and paged decode into a complete attention forward. A free ML coding interview task on NudgeCode: write it in Python and check it with tests in your browser. Hard · Attention, KV Cache, GQA · from Paged KV Cache.

Combine KV writes, varlen prefill and paged decode into a complete attention forward.