flashinfer.cake_fused_qk_rope_fp8_append

cake_fused_qk_rope_fp8_append is the FP8 variant of flashinfer.cake_fused_qk_rope_append: in one launch it applies the per-head RMSNorm of the query and key projections and the NeoX-style rotary embedding, quantizes the query to FP8 E4M3 with a per-row scale, and appends the quantized key and value rows into a paged FP8 KV cache. It serves prefill and decode (head_dim 128; 8 query heads with 1 KV head, or 64 with 8) on SM90a, SM100a and SM103a GPUs, writes only into caller-owned buffers, and is CUDA Graph capturable. launch_plan() exposes the host-side stage selection.

cake_fused_qk_rmsnorm_rope_quantize_fp8_append_paged_kv_cache(...)	Fused Q/K RMSNorm + NeoX RoPE + FP8 quantization + paged K/V append in one launch.
launch_plan(num_rows, num_requests, num_kv_heads)	Grid of the fused kernel's default build (mirrors the Cake kernel's launch_plan).