【Hackathon 10th Spring No.53】[Feature][Kernel] Optimize AppendAttention for discrete head-wise block_idx PR2 [cf]#7718

Open

bob-cloudforge wants to merge 13 commits intoPaddlePaddle:developfrom

CloudForge-Solutions:task/h10-053-pr2-discrete-block-idx-v4

Commits on May 4, 2026

feat(append_attn): head-wise SWA recycle + discrete-block-idx ABI
bob-cloudforge
committed
fix(t53/pr2): close CUDA-graph capture, swap_states reorder, doc gap
bob-cloudforge
committed

Commits on May 6, 2026

fix(t53/pr2): PD_CHECK INT32 dtype on Python-supplied tensors before .data<int>()
bob-cloudforge
committed
fix(t53/pr2): assert GQA divisibility before kv_num_heads_global // tp_size
bob-cloudforge
committed
fix(kvcache): use float division in available_gpu_resource for partial-recycle accuracy
bob-cloudforge
committed
fix(scheduler): document SWA recycle block-boundary guard
bob-cloudforge
committed
fix(t53): unstall V1 scheduler in head-wise mode
bob-cloudforge
committed
fix(t53): available_gpu_resource use is-not-None guard, drop dead legacy path
bob-cloudforge
committed