初赛提交指南

提交时间

初赛提交截止时间为 2025 年 11 月 30 日（周日）23:59:59。

你需要在截止时间之前使用以下提供的提交脚本向计分服务器发送你的结果，提交脚本会在开始执行时记录本地时间戳。计分服务器以该时间戳作为判断比赛结束的依据。

比赛集群登录节点与积分服务器的时间差异问题

由于机器间时间差异及网络延迟等因素，计分服务器将接受时间戳与本地时间相差在 10 分钟内的提交。

请不要尝试伪造该时间戳，组委会将可能要求临近结束时提交的队伍提供文件修改时间等佐证材料。

请参赛队伍注意截止时间，尽早提交，以免出现错过提交截止时间的情况。概不接受因机器时间差异等原因导致错过截止时间的补交请求。

准备提交文件

每道题目需要提交的文本文件如下：

优化报告提交要求

参赛队伍需在初赛结束时提交一份 优化报告，具体要求和提交方式见每个赛题的优化报告部分
截止时间是 2025 年 12 月 3 日 23:59:59，请在截止时间前提交优化报告

Benchmarks

HPL

日志文件xxx.out，包括正确性验证、flops输出等。内容类似于

================================================================
Job ID: 37851, Running on node: g0001, Start Time: Wed Oct 15 05:10:38 PM CST 2025
================================================================
...
--------------------------------------------------------------------------------
||Ax-b||_oo/(eps*(||A||_oo*||x||_oo+||b||_oo)*N)=   0.000553521720 ...... PASSED
||Ax-b||_oo  . . . . . . . . . . . . . . . . . = 0.0000000163628932
||A||_oo . . . . . . . . . . . . . . . . . . . = 47891.7234336114634061
||x||_oo . . . . . . . . . . . . . . . . . . . = 29.1905095333805278
||b||_oo . . . . . . . . . . . . . . . . . . . = 0.4999915975420884
================================================================================
================================================================
Job End Time: Wed Oct 15 05:10:56 PM CST 2025
================================================================

注：需在运行脚本run.sh中，手动加注命令，在log中标记运行起止时间（如赛题介绍示例脚本所示）。log若无时间标注，视为无效提交

HPCG

日志文件xxx.out，包括正确性验证、flops输出等。内容类似于

================================================================
Job ID: 37851, Running on node: g0001, Start Time: Wed Oct 15 05:10:38 PM CST 2025
================================================================
...
Final Summary=
Final Summary::HPCG result is VALID with a GFLOP/s rating of=3055.69
Final Summary::HPCG 2.4 rating for historical reasons is=3073.34
Final Summary::Results are valid but execution time (sec) is=314.3606
Final Summary::Official results execution time (sec) must be at least=1800
================================================================
Job End Time: Wed Oct 15 05:10:56 PM CST 2025
================================================================

注：实际运行时长不得少于300s，否则提交无效。

注：需在运行脚本run.sh中，手动加注命令，在log中标记运行起止时间（如赛题介绍示例脚本所示）。log若无时间标注，视为无效提交

Alphafold3

Alphafold3 共有12个测例，每个测例需要提交一份文件，测例的顺序与名称对应如下：

序号	测例
1	37aa_2jo9
2	107aa_1tce
3	301aa_3db6
4	436aa_1muu
5	583aa_1cf3
6	740aa_4a5s
7	959aa_7lzh
8	1024aa_5l6h
9	1284aa_4xwk
10	1290aa_1g9a
11	1303aa_1t3t
12	1491aa_5kis

每个测例需要准备两份文件，最后提交时，需要将结果汇总到一份文件中：

程序运行时的标准输出 stdout.*，内容类似于：

Alphafold3 sample standard output

Running AlphaFold 3. Please note that standard AlphaFold 3 model parameters are
only available under terms of use provided at
https://github.com/google-deepmind/alphafold3/blob/main/WEIGHTS_TERMS_OF_USE.md.
If you do not agree to these terms and are using AlphaFold 3 derived model
parameters, cancel execution of AlphaFold 3 inference with CTRL-C, and do not
use the model parameters.
Skipping running the data pipeline.
Found local device: cpu
Building model from scratch...
loading the model parameters...
Loading from /home/sccomp/WORK/alphafold3/weights/af3.bin
Processing fold input 959aa_7LZH
Skipping data pipeline...
Output directory: output/959aa_7lzh
Writing model input JSON to output/959aa_7lzh
Predicting 3D structure for 959aa_7LZH for seed(s) (1,)...
Featurising data for seeds (1,)...
Featurising 959aa_7LZH with rng_seed 1.
Featurising 959aa_7LZH with rng_seed 1 took 18.72 seconds.
Featurising data for seeds (1,) took  24.59 seconds.
Running model inference for seed 1...
Running model inference for seed 1 took  3839.33 seconds.
Extracting output structures (one per sample) for seed 1...
Extracting output structures (one per sample) for seed 1 took  1.90 seconds.
Running model inference and extracting output structures for seed 1 took  3841.23 seconds.
Running model inference and extracting output structures for seeds (1,) took  3841.23 seconds.
Writing outputs for 959aa_7LZH for seed(s) (1,)...
Done processing fold input 959aa_7LZH.

ranking_scores.csv文件，用于正确性验证，内容类似于：

seed,sample,ranking_score
1,0,0.6900024153732506
1,1,0.6924695267712131
1,2,0.6819236425461754
1,3,0.6985983714101734
1,4,0.6822181319029088

你最后需要将这两个文件的内容整合到一起(文件名不重要)，内容类似于：

Alphafold3 sample submission file

seed,sample,ranking_score
1,0,0.6900024153732506
1,1,0.6924695267712131
1,2,0.6819236425461754
1,3,0.6985983714101734
1,4,0.6822181319029088
Running AlphaFold 3. Please note that standard AlphaFold 3 model parameters are
only available under terms of use provided at
https://github.com/google-deepmind/alphafold3/blob/main/WEIGHTS_TERMS_OF_USE.md.
If you do not agree to these terms and are using AlphaFold 3 derived model
parameters, cancel execution of AlphaFold 3 inference with CTRL-C, and do not
use the model parameters.
Skipping running the data pipeline.
Found local device: cpu
Building model from scratch...
loading the model parameters...
Loading from /home/sccomp/WORK/alphafold3/weights/af3.bin
Processing fold input 959aa_7LZH
Skipping data pipeline...
Output directory: output/959aa_7lzh
Writing model input JSON to output/959aa_7lzh
Predicting 3D structure for 959aa_7LZH for seed(s) (1,)...
Featurising data for seeds (1,)...
Featurising 959aa_7LZH with rng_seed 1.
Featurising 959aa_7LZH with rng_seed 1 took 18.72 seconds.
Featurising data for seeds (1,) took  24.59 seconds.
Running model inference for seed 1...
Running model inference for seed 1 took  3839.33 seconds.
Extracting output structures (one per sample) for seed 1...
Extracting output structures (one per sample) for seed 1 took  1.90 seconds.
Running model inference and extracting output structures for seed 1 took  3841.23 seconds.
Running model inference and extracting output structures for seeds (1,) took  3841.23 seconds.
Writing outputs for 959aa_7LZH for seed(s) (1,)...
Done processing fold input 959aa_7LZH.

Bagel

Bagel 只需要提交一份文件：

完整的日志文件 output.log，包含正确性验证结果、各测例用时等。内容类似于：

================================================================
Job ID: 17027, Running on node: fuse1, Start Time: Wed Oct 15 05:10:38 PM CST 2025
================================================================
Loading CUDA module...
...
19 : 68.575 ms
20 : 22.284 ms
================================================================
Job End Time: Wed Oct 15 05:10:56 PM CST 2025
================================================================

DIT

DIT 只需要提交一份文件：

完整的日志文件 dit_xxxxx.out，包含正确性验证结果、运行时长。内容类似于：

DiT sample submission file

 99%|█████████▉| 247/250 [02:30<00:01,  2.33it/s]
 99%|█████████▉| 248/250 [02:31<00:01,  1.71it/s]
100%|█████████▉| 249/250 [02:32<00:00,  1.64it/s]
100%|██████████| 250/250 [02:32<00:00,  1.61it/s]
100%|██████████| 250/250 [02:32<00:00,  1.64it/s]
============================================================
图像差异检查报告
============================================================
图像尺寸: (256, 256, 3)
RMSE: 3.19e-03
MSE: 1.02e-05
最大绝对差异: 1.00e+00
最小绝对差异: 0.00e+00
平均绝对差异: 1.02e-05
阈值: 1.00e-05
检查结果: 通过
============================================================
差异可视化图像已保存: diff_visualization.png

MYSTERY

MYSTERY 只需要提交一份文件：

完整的日志文件 run.log，包含正确性验证结果、运行时长。内容类似于：

Mystery sample submission file

X Ray Image task:
================================================================
[2025-11-17 21:30:38.308] Running...
================================================================
0
1
...
511
[[111, 351, np.float64(0.010245295879398755)], [195, 273, np.float64(0.008082491342037493)], [201, 187, np.float64(0.009585248501416939)], [222, 72, np.float64(0.007761181292119802)], [228, 447, np.float64(0.008935738327156805)], [237, 30, np.float64(0.009258154027198118)], [282, 186, np.float64(0.009897658125727224)], [345, 294, np.float64(0.008458769186054356)], [373, 472, np.float64(0.009994601066936188)], [429, 297, np.float64(0.010314554610626176)], [458, 57, np.float64(0.007845900909236063)], [482, 176, np.float64(0.0074326216306861885)]]
================================================================
[2025-11-17 21:44:24.265] Checking...
================================================================
source rate of 0 is 0.007
source rate of 1 is 0.007
source rate of 2 is 0.007
source rate of 3 is 0.008
source rate of 4 is 0.008
source rate of 5 is 0.008
source rate of 6 is 0.009
source rate of 7 is 0.009
source rate of 8 is 0.009
source rate of 9 is 0.01
source rate of 10 is 0.01
source rate of 11 is 0.01
det rate of 0 is 0.010245295879398755
det rate of 1 is 0.008082491342037493
det rate of 2 is 0.009585248501416939
det rate of 3 is 0.007761181292119802
det rate of 4 is 0.008935738327156805
det rate of 5 is 0.009258154027198118
det rate of 6 is 0.009897658125727224
det rate of 7 is 0.008458769186054356
det rate of 8 is 0.009994601066936188
det rate of 9 is 0.010314554610626176
det rate of 10 is 0.007845900909236063
det rate of 11 is 0.0074326216306861885
source #0: distance = 0.0, delta rate = 0.00024529587939875475
source #1: distance = 0.0, delta rate = 8.249134203749259e-05
source #2: distance = 1.0, delta rate = 0.0005852485014169392
source #3: distance = 1.0, delta rate = 0.00023881870788019805
source #4: distance = 1.4142135623730951, delta rate = 0.0009357383271568053
source #5: distance = 3.0, delta rate = 0.0002581540271981188
source #6: distance = 0.0, delta rate = 0.00010234187427277629
source #7: distance = 1.0, delta rate = 0.0014587691860543555
source #8: distance = 1.0, delta rate = 0.0009946010669361887
source #9: distance = 0.0, delta rate = 0.00031455461062617623
source #10: distance = 4.123105625617661, delta rate = 0.0008459009092360624
source #11: distance = 1.0, delta rate = 0.0004326216306861884
max distance: 4.123105625617661
max delta rate: 0.0014587691860543555
================================================================
[2025-11-17 21:44:24.454] Done.
================================================================

准备提交脚本

将提交脚本 submit.sh 下载到集群上，然后添加执行权限：

wget https://lab.cs.tsinghua.edu.cn/scc/submit.sh
chmod +x submit.sh

不要修改提交脚本

在通常情况下，你不需要也不应该修改 submit.sh 文件中的内容。

如果你认为确有必要进行某种修改，请先在参赛群中与工作人员交流并得到许可。

使用脚本提交

提交脚本的用法为 ./submit.sh [--dry] <problem> <file>，其中：

--dry 为可选项，表示只提交到服务器进行合法性检验，不计入成绩计算
problem 为题目名称，在 HPL，HPCG，AF3-i，BAGEL，DIT，MYSTERY 中选择（不区分大小写，AF3- 后的 i 表示第i个测例，具体顺序可查看本页面 Alphafold3 部分， \(1 \leq i \leq 12\)）
file 为提交文件的路径

例如，以下指令将提交 HPL 题目的 HPL.dat 文件：

./submit.sh HPL HPL.dat

以下指令将测试 HPL 题目的 HPL.dat 文件是否合法，不计入成绩计算：

./submit.sh --dry HPL HPL.dat

例如，以下指令将提交 Alphafold3 第 7 个测例（959aa_7lzh）的结果：

./submit.sh AF3-7 alphafold3-7.txt

在第一次提交时，脚本会提示输入提交用户名和密码，提交用户名、密码将通过邮件发送到各队队长的邮箱中。在第一次输入后，这些信息会被保存在 ~/.thuscc_credentials 文件中，之后的提交无需再次输入密码。若密码输入错误，可以在删除该文件后重新进行提交。

如果一切顺利，你将看到如下输出：

Server Message: Submission Accepted

其他可能出现的结果包括：

File Missing: 积分服务器没有收到你提交的文件
Bad Request (... are required): 提交脚本发送的其他参数错误
Time Error (difference between server time and submit time is more than 10 minutes)：提交时间与计分服务器时间相差超过 10 分钟
Submission Already Closed：提交时间已过截止时间
Unknown Team：提交脚本没有找到正确的队伍名，请确认用户名是否正确
Wrong Password：提交密码错误
Unknown Problem (choose from ...)：提交时指定的题目名称错误
Multiple Results：你的提交文件中可能包含了多次运行或多个测试规模的结果，请只保留一个结果重新提交
Invalid Result (...)：你的提交没有通过基本的合法性校验，括号中包含了具体的错误原因。如果你看到了令人困惑的错误原因，请首先确认是否交错了题目
Validation Passed (dry run)：使用了 --dry 选项，只进行了合法性检验，不计入成绩计算

其他特殊情况

如果你得到了 Failed Storing Submission File 或 Unknown Error (please contact admin or staff) 等指示您联系管理员的信息，请立即在参赛群中联系工作人员。

实时排行榜

你可以在 THUSCC25 实时排行榜页面查看根据所有队伍当前提交计算的实时成绩排名。