OwnerSunshine530 cdc2dd490b feat: Triton varlen因果flash attention(块对角,单kernel,消逐块调用+mask构造开销) ...

每program处理(用户段query块,head),只遍历段内<=该块的key(因果),在线softmax,
fp16读写fp32累加。CONFIG.attn=triton(默认仍chunked);_triton_block_meta每batch算一次
block→段映射8层复用;_resolve_attn在无triton/CPU时回退chunked。等价测试+bench --attn triton。
数学等价(FlashAttention同类,规则允许),不改组网。

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

2026-06-17 00:14:53 +08:00

..

tests

feat: Triton varlen因果flash attention(块对角,单kernel,消逐块调用+mask构造开销)

2026-06-17 00:14:53 +08:00

bench.py

feat: Triton varlen因果flash attention(块对角,单kernel,消逐块调用+mask构造开销)

2026-06-17 00:14:53 +08:00

build_env.sh

fix: build_env.sh 简化为纯净版本（避免 CUDA 预热导致异常）

2026-06-12 21:55:09 +08:00

EXPERIMENTS.md

docs: 收尾 — 最终67.998/记录RepEncoder预计算尝试与结论

2026-06-16 13:18:48 +08:00

infer.py

feat: Triton varlen因果flash attention(块对角,单kernel,消逐块调用+mask构造开销)

2026-06-17 00:14:53 +08:00

requirements.txt

revert: requirements.txt 还原为原始完整依赖列表

2026-06-12 21:24:22 +08:00

RISKS.md

docs: 潜在风险说明(RepEncoder预计算合规灰区/max_feasign一致性)与合规保底

2026-06-15 20:44:57 +08:00