[Optimization]【Hackathon 10th Spring No.49】Port ngram_match and hybrid_mtp_ngram kernels to CUDA#6960

Open

cloudforge1 wants to merge 30 commits intoPaddlePaddle:developfrom

CloudForge-Solutions:task/049-spec-decode-gpu-kernel

Commits on Mar 6, 2026

Merge remote-tracking branch 'upstream/develop' into develop
cloudforge1
committed
Merge remote-tracking branch 'upstream/develop' into develop
cloudforge1
committed

Commits on Mar 9, 2026

Merge remote-tracking branch 'upstream/develop' into develop
cloudforge1
committed
Merge remote-tracking branch 'upstream/develop' into develop
cloudforge1
committed

Commits on Mar 10, 2026

Commits on Mar 11, 2026

Merge remote-tracking branch 'upstream/develop' into develop
cloudforge1
committed

Commits on Mar 17, 2026

Merge remote-tracking branch 'upstream/develop' into develop
cloudforge1
committed

Commits on Mar 19, 2026

Merge remote-tracking branch 'upstream/develop' into develop
cloudforge1
committed

Commits on Mar 21, 2026

Commits on Mar 22, 2026

fix: keep input_ids_len on CPU in __init__, move to GPU in _run_impl
cloudforge1
committed

Commits on Mar 25, 2026

Extract shared ngram search into __device__ helper (ngram_match_common.cuh)
cloudforge1
committed

Commits on Mar 30, 2026