OwnerSunshine530 a5ee660523 perf: chunk_users 退回 4(评测最优67.998;3更慢8持平→chunk维度榨干) ...

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

2026-06-16 23:58:56 +08:00

..

tests

feat: sparse_pool 选项 — (段×唯一)稀疏矩阵乘做池化,避免materialize[M,emb]

2026-06-15 15:15:13 +08:00

bench.py

feat: movedev_rep — 在move_batch_to_device(不计时/主进程/有模型有数据)算rep,model跳过embedding

2026-06-16 19:37:34 +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

perf: chunk_users 退回 4(评测最优67.998;3更慢8持平→chunk维度榨干)

2026-06-16 23:58:56 +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