ViperEkura 2667b8116d refactor: unify paged and contiguous attention kernels via KVSource policy ...

- merge AttentionParams and PagedAttentionParams into one struct
- add attn_kv_source.cuh with ContigKV/PagedKV addressing policies
- template prefill/decode kernels (MMA + scalar) on the KV policy, deleting the four duplicated attn_paged_*.cuh variants
- template dispatcher launchers on KV; single combine kernel
- verify: all correctness tests pass and SASS matches baseline (no perf regression)

2026-08-05 14:06:13 +08:00

..

architecture.md

docs: align CUDA kernel and RoPE docs with code

2026-08-03 20:54:40 +08:00

cuda_kernels.md

refactor: unify paged and contiguous attention kernels via KVSource policy

2026-08-05 14:06:13 +08:00

dataflow.md

docs: audit non-CUDA documentation

2026-08-02 07:39:24 +08:00

internals.md

docs: align CUDA kernel and RoPE docs with code

2026-08-03 20:54:40 +08:00