1
00:00:00,240 --> 00:00:06,680
Consider an original counting puzzle: three boxes contain four
beads each, then five extra beads are added.
看一个原创计数题：三个盒子各有四颗珠子，再加五颗珠子。

2
00:00:07,260 --> 00:00:13,240
The correct total is seventeen, but an unfinished answer still
contains gaps in its reasoning and final result.
正确总数为十七，但未完成回答的推理和结果中仍有缺口。

3
00:00:14,020 --> 00:00:19,840
Imagine some words becoming visible while other positions remain
masked, as a response gradually takes shape.
想象部分词逐渐可见，而其他位置仍被遮蔽，回答逐步成形。

4
00:00:20,620 --> 00:00:25,800
A later, more complete version can contain clues that the
earlier noisy view has not yet made explicit.
后来更完整的版本可能包含早期噪声视图尚未明确的线索。

5
00:00:26,560 --> 00:00:32,479
These token boxes are an illustrative schematic, not a
reproduced benchmark question or an actual model decoding trace.
这些词元框是示意图，不是基准题或实际模型解码轨迹。

6
00:00:33,180 --> 00:00:38,540
Can self-generated later information train a diffusion model to
improve its earlier prediction decisions?
论文探索自身生成的后期信息能否训练扩散模型，改进其早期决策。

7
00:00:39,167 --> 00:00:46,027
A diffusion language model generates by iterative denoising,
beginning with a response whose token positions are masked.
扩散语言模型从响应词元全部遮蔽的状态出发，通过迭代去噪生成。

8
00:00:46,927 --> 00:00:52,387
Each step predicts masked-position distributions, rather than
only the next left-to-right token.
每一步预测遮蔽位置的分布，而不是只确定从左到右的下一个词元。

9
00:00:53,187 --> 00:00:59,167
Confident predictions become visible; other positions stay
masked until later steps improve their reliability.
选中高置信预测变为可见，其余位置继续遮蔽，等待后续更可靠的信息。

10
00:01:00,007 --> 00:01:06,487
The current noisy state and clean final response are different
moments along one sampled generation trajectory.
当前噪声状态和最终干净响应因此属于同一采样生成轨迹的不同时刻。

11
00:01:07,327 --> 00:01:13,547
Arbitrary-order generation permits conditioning on later
information, rather than only a left-to-right prefix.
任意顺序结构允许利用后期信息，而不把指导限制在从左到右的前缀。

12
00:01:14,507 --> 00:01:20,507
During training, a completed self-generated response supplies
privileged teacher-specific context.
方法在训练中利用这一性质，把模型完成的响应转成教师专属上下文。

13
00:01:21,100 --> 00:01:28,220
The student first samples its own complete trajectory,
preserving earlier noisy states and the final generated
response.
学生先采样自己的完整轨迹，保存早期噪声状态与最终生成响应。

14
00:01:29,200 --> 00:01:35,200
For an earlier denoising step, its input remains that current
noisy state together with the original prompt.
对早期去噪步骤，学生输入仍是当前噪声状态与原始提示。

15
00:01:36,340 --> 00:01:42,160
The teacher view selectively reveals tokens from the completed
response at positions that are currently masked.
教师视图在当前遮蔽位置选择性揭示完成响应中的词元。

16
00:01:43,040 --> 00:01:47,900
Revealing a subset supplies clues while leaving prediction work
for the teacher distribution.
只揭示子集，因此特权上下文提供线索，同时仍留下教师分布的预测任务。

17
00:01:48,860 --> 00:01:54,060
Both roles start from the same diffusion model, but different
views inform their next-step predictions.
两角色源自同一扩散模型，但不同视图为下一步预测提供不同信息。

18
00:01:54,960 --> 00:02:01,060
Future information comes from its sampled response, not time
travel or external knowledge of unseen test answers.
后期信息来自已完成采样响应，并非时间旅行或未见测试答案的外部知识。

19
00:02:01,673 --> 00:02:06,713
The objective compares student and teacher distributions at the
current denoising step.
学习目标遵循去噪过程，比较当前生成步骤的学生与教师分布。

20
00:02:07,773 --> 00:02:14,813
Only selected confident positions among the currently masked
tokens actively determine the next state in that step.
当前遮蔽词元中，只有选中高置信位置主动决定该步下一状态。

21
00:02:15,853 --> 00:02:21,933
The divergence loss is applied to that subset, rather than
treating every position as a left-to-right continuation.
方法把散度损失集中在该子集，而非把每位置视为独立的从左到右续写。

22
00:02:23,013 --> 00:02:29,793
The teacher supplies dense vocabulary-level guidance, giving
more information than the final correctness reward alone.
教师在这些位置提供密集词表级指导，比单一最终正确性奖励提供更多信息。

23
00:02:30,813 --> 00:02:36,533
The default selects supervised masked positions by teacher
confidence, supported by the reported ablation.
默认用教师置信度选择监督遮蔽位置，这一选择由报告消融实验支持。

24
00:02:37,753 --> 00:02:43,333
Summing this guidance over the sampled trajectory aligns the
objective with the iterative denoising process.
训练沿采样轨迹汇总逐步指导，使目标与扩散响应实际展开方式一致。

25
00:02:43,953 --> 00:02:50,893
The completed response used for guidance is generated by the
student itself, keeping the trajectory on-policy.
用于指导的完整响应由学生自身生成，保持同策略采样。

26
00:02:51,752 --> 00:02:57,653
The default recipe samples until a correct final answer appears,
or reaches its budget of eight attempts.
默认方案持续采样直到出现正确最终答案，或达到八次尝试预算。

27
00:02:58,613 --> 00:03:06,733
Task-specific checks select useful future information; Sudoku
uses a score threshold rather than binary answer matching.
任务专属检查筛选有用后期信息；Sudoku 使用分数门槛，而非二元答案匹配。

28
00:03:07,693 --> 00:03:14,293
Training follows the whole sampled trajectory, rather than one
artificial forward pass over a static response.
训练沿完整采样去噪轨迹展开，而非只对静态参考响应进行一次人工前向计算。

29
00:03:15,413 --> 00:03:21,193
The default fixes the initial teacher policy, providing a stable
reference as student parameters change.
默认还把教师策略固定在初始检查点，使学生参数变化时拥有稳定参考。

30
00:03:22,133 --> 00:03:27,073
Teacher information comes from enriched context, while model
updates apply to the learner.
教师额外信息来自增强条件视图，而模型更新应用于学习者。

31
00:03:27,547 --> 00:03:34,827
The experiments use LLaDA Instruct, an eight-billion-parameter
diffusion model, on four math and planning tasks.
实验使用LLaDA-8B-Instruct扩散模型，评估四项数学推理与规划任务。

32
00:03:35,486 --> 00:03:43,087
Tasks include GSM eight K, MATH five hundred, Countdown, and
four-by-four Sudoku, with retained benchmark settings.
任务为GSM8K、MATH500、Countdown及四乘四Sudoku，沿用基准设置。

33
00:03:43,867 --> 00:03:50,007
Training uses low-rank adaptation and reverse KL, retaining a
quarter of eligible tokens for teacher context.
训练使用低秩适应，默认反向 KL，并为教师上下文保留四分之一适用词元。

34
00:03:50,747 --> 00:03:55,547
Training length is two hundred fifty-six tokens; evaluation
lengths depend on the task category.
默认训练响应长度为二百五十六词元，评估长度则取决于任务类别。

35
00:03:56,367 --> 00:04:01,627
Evaluation occurs every twenty-five steps before five hundred
one, reporting the best evaluated checkpoint.
五百零一步前每二十五步检查性能，报告结果选择评估过的最佳检查点。

36
00:04:02,327 --> 00:04:08,747
The comparison specifies the model, distillation recipe, task,
generation length, and checkpoint-selection rule.
这些细节明确比较对象：特定模型、蒸馏方案、任务、生成长度和检查点选择策略。

37
00:04:09,193 --> 00:04:17,953
Consider GSM eight K at length two hundred fifty-six, using the
eight-billion-parameter LLaDA Instruct diffusion model.
看GSM8K生成长度256列，使用报告的LLaDA-8B-Instruct扩散模型。

38
00:04:18,673 --> 00:04:24,113
The base scores seventy-six percent; diffusion GRPO reaches
seventy-nine point eight percent.
基础模型得分为百分之七十六，报告的 diffusion GRPO 基线为百分之七十九点八。

39
00:04:24,753 --> 00:04:31,073
Diffusion on-policy self-distillation reaches eighty-one
percent, exceeding that baseline in this specific setting.
扩散同策略自蒸馏达到百分之八十一，在此设置下优于该基线。

40
00:04:32,273 --> 00:04:37,693
Other Table one tasks and generation lengths differ, so the
comparison remains tied to this configuration.
表一还有其他任务与生成长度，因此比较必须限定在展示配置。

41
00:04:38,653 --> 00:04:44,093
At MATH five hundred length five hundred twelve, diffusion GRPO
reports thirty-nine point two.
MATH500 长度五百一十二时，GRPO 报告三十九点二。

42
00:04:44,753 --> 00:04:49,093
The proposed method reports thirty-seven point eight for that
same task and generation length.
所提方法在同一任务与生成长度下报告三十七点八。

43
00:04:49,893 --> 00:04:55,453
The reported improvement therefore does not imply that the
proposed method wins every individual cell in the table.
因此，报告提升并不意味着所提方法赢得表中每一格。

44
00:04:56,333 --> 00:05:01,833
This chart preserves the precise model, baseline, generation
length, and scope of the reasoning comparison.
图表展示一项精确推理比较，保留基础模型、基线、生成长度与评估范围。

45
00:05:02,520 --> 00:05:08,160
Optimization-step counts also describe how quickly dense teacher
guidance can improve the learner.
论文还报告优化步数，展示有用教师分布如何更快指导学习者。

46
00:05:09,220 --> 00:05:16,620
GSM eight K lists seven thousand seven hundred baseline updates
versus four hundred twenty-five for d-OPSD.
GSM8K 表二列出基线七千七百步，d-OPSD 为四百二十五步。

47
00:05:17,500 --> 00:05:23,840
Baseline counts come from its original report: these are
optimization steps, rather than matched wall-clock timings.
基线步数来自原报告，因此这些数字描述优化步数，而非匹配的运行时间测量。

48
00:05:24,720 --> 00:05:30,940
Dense supervision is linked to efficiency, while each training
step still spends computation on sampled trajectories.
方法把密集监督联系到这种效率，但每步采样仍消耗轨迹生成计算。

49
00:05:31,940 --> 00:05:37,860
Return to the bead example: a checked later response guides
earlier masked prediction states.
回到未完成珠子回答：经过检查的后期响应，为早期遮蔽预测状态提供指导线索。

50
00:05:38,940 --> 00:05:45,560
Ordinary generation evaluates the trained model, whose
parameters improve through training-specific privileged context.
随后用普通生成过程评估学习模型；训练专属特权上下文用于改善参数。

51
00:05:46,640 --> 00:05:53,380
The method pairs self-generated future context with step-aligned
supervision, tailored to iterative diffusion reasoning.
实际思想是把自身生成的后期上下文与逐步对齐监督配对，适配迭代扩散推理。