1
00:00:00,320 --> 00:00:06,640
Imagine searching delivery policies to learn whether batteries
can travel by air, among several related documents.
想象搜索配送政策，了解电池能否航空运输，候选中有几篇相关文档。

2
00:00:07,380 --> 00:00:14,400
The retriever finds passages about shipping, damaged products,
and air transport, but their relevance differs.
检索器找到配送、损坏商品和航空运输文段，但相关程度不同。

3
00:00:15,260 --> 00:00:21,320
A reranker must identify which supplied passage answers the
question, then improve the candidate ordering.
重排器需要找出能回答问题的输入文段，再改善候选顺序。

4
00:00:22,220 --> 00:00:26,660
These fictional documents describe teaching roles rather than
actual benchmark passages.
文档是虚构教学示例；展示标题描述角色，而非实际基准文段。

5
00:00:27,660 --> 00:00:33,400
Generating candidate identifiers adds computation beyond
estimating the relevance of supplied documents.
可以顺序生成候选标识，但这在估计文档相关性之外引入额外计算。

6
00:00:34,440 --> 00:00:39,200
CompRank directly scores candidates while reducing the tokens
needed for comparison.
CompRank 围绕直接给候选项评分组织计算，并减少比较所需词元。

7
00:00:39,740 --> 00:00:45,080
CompRank places independent document blocks before a query
decision block that scores the candidates.
CompRank 在查询决策块前放置独立文档块，随后给输入候选项评分。

8
00:00:45,980 --> 00:00:52,240
Tokens inside each document use causal self-attention, building
a representation from that document's own context.
每篇文档内部使用因果自注意力，根据自身上下文构建表示。

9
00:00:53,140 --> 00:00:58,400
Cross-document attention is prohibited, so document encoding
does not depend on neighboring candidates.
词元不能跨文档块关注，消除表示构建时对邻近候选项的依赖。

10
00:00:59,420 --> 00:01:05,060
Local positions avoid document-order offsets, reducing one
source of candidate-permutation sensitivity.
局部位置索引避免文档顺序偏移，减少一种候选排列敏感来源。

11
00:01:06,000 --> 00:01:11,780
The query block still sees all document blocks, maintaining a
global candidate view when relevance is estimated.
查询块仍可看到全部文档块，在估计相关性时保留全局候选视野。

12
00:01:12,560 --> 00:01:18,120
Independent document states can therefore be reused in principle
across queries and reordered candidate lists.
因此，独立文档状态原则上可以跨查询和重排候选列表复用。

13
00:01:18,700 --> 00:01:25,580
Independent encoding removes cross-document interactions, but
query attention still scales with document tokens.
独立文档编码移除文档间交互，但查询注意力仍随文档词元数增长。

14
00:01:26,420 --> 00:01:33,500
Compression exposes only selected document key-value states to
the query block, reducing the final interaction cost.
压缩只向查询块提供选定文档键值状态，减少最终交互成本。

15
00:01:34,680 --> 00:01:41,140
Document representations stay unchanged; selection controls
which states query-to-document attention sees.
文档表示保持不变；子集决定哪些词元参与查询到文档评分注意力。

16
00:01:42,100 --> 00:01:47,760
Segment-wise selection retains regularly spaced tokens, covering
the document rather than only its end.
分段选择每隔若干位置保留词元，覆盖全篇，而非只覆盖末尾。

17
00:01:48,720 --> 00:01:54,660
The illustrated position numbers show regular spacing, without
asserting which particular words an experiment retained.
示意位置编号展示规则间隔，不声称实验保留了哪些具体词。

18
00:01:55,520 --> 00:02:01,780
The default Step-ten strategy uses that coverage to balance a
smaller attention budget against ranking quality.
默认 Step-10 使用这种覆盖，在较小注意力预算与排序质量间权衡。

19
00:02:02,480 --> 00:02:08,320
Scoring aggregates query attention on retained document tokens,
treating that mass as relevance evidence.
评分汇总查询分给各文档保留词元的注意力权重，将其视为相关性证据。

20
00:02:09,200 --> 00:02:15,100
Attention is normalized across the retained document tokens and
then aggregated into a score for each candidate.
注意力跨保留文档词元归一化，再聚合成各候选项得分。

21
00:02:15,840 --> 00:02:22,100
The main Mistral configuration uses one decision token,
averaging scores across attention heads at a fixed layer.
主 Mistral 配置使用一个决策词元，在固定层跨注意力头平均得分。

22
00:02:23,120 --> 00:02:26,960
Inference directly sorts supplied documents by their
attention-derived scores.
推理按注意力得分排列文档，直接产生输入候选项的顺序。

23
00:02:27,440 --> 00:02:32,580
No identifier sequence is generated; scoring avoids sequential
decoding of a ranking string.
不生成自回归标识序列，因此评分避免顺序解码排序字符串。

24
00:02:33,240 --> 00:02:38,820
Its output is candidate order, not reconstructed document text
or a generated answer to the original query.
输出是候选顺序，而不是重建文档文本或生成原始查询的回答。

25
00:02:39,587 --> 00:02:46,047
Identifier generation uses the full vocabulary, including tokens
unrelated to valid input candidates.
生成目标通过完整词表学习文档标识输出，其中包含与有效候选项无关的词元。

26
00:02:47,067 --> 00:02:52,847
CompRank uses attention-derived scores to form a pointer
distribution over the input candidates.
CompRank 用同一组注意力文档得分，在输入候选项上构造指针分布。

27
00:02:53,667 --> 00:03:00,027
The CopyNet-style loss supervises the relevant candidate,
training the same scores inference later sorts.
CopyNet 式损失直接监督相关候选项，使学习评分与推理排序一致。

28
00:03:01,207 --> 00:03:07,247
Candidate selection replaces identifier generation, matching
supervision to the reranker decision.
这预测候选选择，而非自回归标识，让训练目标匹配重排所需决策。

29
00:03:08,247 --> 00:03:14,827
Experiments use Mistral seven B with thirty-document training
lists and positive examples derived from MS MARCO.
实验使用 Mistral-7B、三十文档训练列表，以及来自 MS MARCO 的正例。

30
00:03:15,667 --> 00:03:23,087
BEIR evaluation compares rerankers on the same one hundred BM
twenty-five retrieved candidates across seven datasets.
七个 BEIR 数据集上，重排器共用每查询的百个 BM25 候选。

31
00:03:23,627 --> 00:03:32,827
Table one compares compression strategies within CompRank,
measuring average NDCG at ten over seven BEIR datasets.
表一在 CompRank 内比较压缩策略，指标为七 BEIR 数据集平均 NDCG@10。

32
00:03:33,627 --> 00:03:40,307
The full-token version retains all document tokens and scores
thirty-nine point seven on the reported average metric.
完整词元版本保留全部文档词元，报告平均指标为三十九点七。

33
00:03:41,207 --> 00:03:47,287
Step-ten retains ten point two percent of tokens and scores
thirty-nine point two, half a point lower.
Step-10 保留百分之十点二的词元，得分三十九点二，低半点。

34
00:03:47,987 --> 00:03:53,867
Token retention measures attention budget; NDCG measures ranking
quality, so they have different meanings.
词元保留描述注意力输入预算；NDCG 描述排序效果，两者含义不同。

35
00:03:54,547 --> 00:04:00,787
This averages seven dataset scores, rather than measuring task
completion or promising every-cell improvement.
平均值综合七数据集得分，而非任务完成率，也不承诺每格提升。

36
00:04:01,487 --> 00:04:07,447
The matched test supports structured coverage, while stronger
compression still trades away ranking quality.
匹配比较支持结构化文档覆盖的高效性，但更强压缩仍会牺牲排序性能。

37
00:04:08,047 --> 00:04:14,267
A TREC-COVID test trains on thirty-document lists, then
evaluates up to five hundred without further fine-tuning.
另一项 TREC-COVID 测试在三十文档列表训练，随后无需再微调评估最多五百候选。

38
00:04:15,007 --> 00:04:21,027
At five hundred candidates, generation-based Direct-List takes
one hundred fourteen point five seconds per query.
五百候选时，生成式 Direct-List 每查询耗时一百一十四点五秒。

39
00:04:21,947 --> 00:04:27,987
Full-token CompRank takes sixteen point two two seconds;
Step-ten takes twelve for the same candidate count.
相同候选数下，完整词元 CompRank 为十六点二二秒，Step-10 为十二秒。

40
00:04:29,187 --> 00:04:34,967
These end-to-end timings use one A one hundred GPU with eighty
gigabytes of memory and batch size one.
端到端测时使用一张八十 GB 显存的 A100 GPU，批大小为一。

41
00:04:35,927 --> 00:04:41,307
Measured evaluation recomputes document blocks; optimized
caching remains a separate design opportunity.
实测流程仍重新计算文档块；优化 KV 缓存是此测量之外的设计机会。

42
00:04:42,167 --> 00:04:48,107
The gain combines decoding-free scoring and compressed
attention, rather than unimplemented serving optimizations.
实测收益因此结合免解码评分与压缩查询注意力，不归因于尚未实现的服务优化。

43
00:04:48,647 --> 00:04:55,007
Return to policy search: choose useful supplied evidence, rather
than write a document identifier sequence.
回到政策搜索：选择有用输入证据，而非在输出阶段编写文档标识。

44
00:04:55,947 --> 00:05:02,307
CompRank independently encodes documents, compresses
query-visible states, and ranks attention scores.
CompRank 分离文档表示，让查询看到结构化子集，并排列注意力候选得分。

45
00:05:03,267 --> 00:05:08,847
Candidate-level supervision trains those scores, aligning the
learning objective with inference-time sorting.
候选级监督训练这些相同得分，把学习目标连接到推理时使用的直接操作。

46
00:05:09,867 --> 00:05:15,607
Computation targets sparse comparison signals, instead of dense
interactions between every document pair.
设计把计算用于比较所需的稀疏信号，而非每对文档之间的密集交互。

47
00:05:16,587 --> 00:05:22,327
Quality and latency evidence is configuration-specific; caching
potential differs from measured serving.
质量与延迟证据对应指定基准配置，并区分状态复用潜力与实测端到端服务。

48
00:05:23,287 --> 00:05:29,607
Align representation, computation, and supervision with the
relevance decision the retrieval system needs.
实际启示是让表示、计算和监督与检索系统真正需要的相关性决策一致。