feat: collate_dedup 默认开(本地4.10->3.98s,AUC精确不变,减查表带宽)冲72

代码/code/infer.py

@@ -168,7 +168,7 @@ CONFIG = {
     "syncfree_mask": True,    # True=用searchsorted构造因果mask(无同步)；False=repeat_interleave(同步)
     "emb_fp16": True,         # True=Embedding表转FP16(查表带宽减半,实测AUC 0.75932≈无损)
     "use_embedding_bag": True,   # F.embedding_bag 融合查表+池化(单kernel,消dedup的unique同步,AUC≈无损)
-    "collate_dedup": False,      # collate(不计时)段内去重+计数→embedding_bag per_sample_weights,减查表带宽(数学等价)
+    "collate_dedup": True,       # collate(不计时)段内去重+计数→embedding_bag per_sample_weights,减查表带宽(本地4.10→3.98,AUC精确不变)
     "dedup_embedding": True,  # True=查表前对sign去重(只查唯一值再展开),本地7.80->6.49s,AUC逐位等价
     "sparse_pool": False,     # True=用(段×唯一)稀疏矩阵乘做池化,避免materialize整个[M,512](段内高重复时省)
     "compile": False,         # 是否 torch.compile（实测慢5×，勿开）