2024-AICS-EXP, a full archive of the AICS intelligent computing system lab course

An archive of one course, not a framework

The repository describes itself as the complete archive of the 2024 new edition lab course files for the AICS intelligent computing system course. The Chinese description names the author, the course and the year, and the point of the project is stated plainly: it is a snapshot of what a lab course actually ran.

The structure is a set of top level directories, one per kind of artefact. There is a directory for the lab manuals, one for the reference answers, one for the assignment statements, one for the development and practice course files, one for the new edition of the theory lecture slides, and one for the electronic version of the first edition textbook.

So this is not a code library. There is no package, no build system and no API. What you get is the paperwork of a course plus the code that goes with it, which is a different and less common kind of open source artefact.

That shape has an obvious audience. The first is students of the course, for whom the repository is a map of what is expected. The second is instructors elsewhere who want to see how a lab sequence in this subject is structured, with reference solutions that have been graded and a score table showing what the ceiling on each experiment is. The third is the author themselves, who keeps the archive current as the course runs.

What changed between the old course and the 2024 edition

The README explains that the 2024 edition made large scale adjustments to the knowledge system and to the lab assignments, and lists what those changes were.

TensorFlow is no longer used and PyTorch is used throughout. Many of the old assignment code bases were heavily modified. New large model experiments were added, and files for a related course on developing and practising intelligent computing systems were added as well.

That last point is why there is a separate directory for the development and practice course files, and why the score table has a second block: three experiments from that companion course appear at the end, covering YOLOv5 object detection inference, BERT on the SQuAD task, and Swin-Transformer image classification, each with a maximum of 100.

The switch from TensorFlow to PyTorch is the change most likely to affect you if you are looking for reference code, because it invalidates any pre-2024 snippet you might otherwise copy from elsewhere. The repository's own answer for anyone with older code is the updated experiment statements and answers, not the archived ones.

Eighteen graded experiments, and a score table that says what the ceiling is

The score table is the most concrete thing in the repository, and it maps every experiment to its maximum points.

The early experiments cover the fundamentals: a handwritten digit classification lab at 100 points, the same lab implemented on the DLP platform at 100, a VGG19 image classification lab in Python at 100, the DLP platform version of image classification at 100, and a non-realtime image style transfer lab at 100. The second block is a PyTorch VGG19 image classification lab, a realtime style transfer inference lab, a realtime style transfer training lab, and a custom PyTorch CPU operator lab, all at 100.

Then the three-way choice experiments, where the numbers differ because the options are not equally weighted. The YOLOv5 option is 80 points, the Tacotron2 option is 90, and the BERT option is 100, which means the choice of topic directly changes the ceiling. The ModelSim simulation lab is 120 points, the largest single item. The intelligent programming language operator development and performance optimisation labs are 100 each. The large model block covers Stable Diffusion at 100, LLAMA3.2 at 100, and CodeLLAMA at 100.

For a student, this is a study plan. For an instructor, it is a rubric.

How the archive is updated, and the two honest caveats

The project status section describes the update policy and contains two admissions that are worth reading before you plan around this repository.

The first is that the project is updated in step with the course schedule, and that as of 2025-06-11 the content had been brought up to date with the large model experiments 8.1, 8.2 and 8.3. The second is that the content is adapted to the spring 2025 semester, and that if later course experiments change, students are welcome to submit pull requests.

There is a third piece of feedback in the same section: a student reported that the Triton optimisation part of experiment 8.1 passes the evaluation even though the code is wrong, and the author invites pull requests to improve it. That is a useful signal about what kind of repository this is. It is a working archive that its own users are correcting, not a finished artefact.

The update log runs from 2024-04-20 to 2026-06-02, and it records the kinds of change that accumulate in a course: slides updated, lab manuals and answers replaced, the Tacotron2 and YOLOv5 options added to the three-way choice experiment, the ModelSim simulation code fixed because the original had too many bugs and could only compare one of four result values, and the parallel_pe part of experiment six updated. The most recent entry, 2026-06-02, updated the code for experiment two, exp_2_2, to add the ratio of average time between DLP and CPU, in order to get full marks under the new experiment requirements.

Practical warnings in the log, from datasets over 100MB to a blue screen

Two entries in the update log read like support tickets rather than changelog lines, and they are the most immediately useful content in the repository.

The first is about data. On 2024-04-20 the datasets accompanying the assignment questions exceeded the 100MB GitHub limit, so the assignment datasets were moved to Baidu Cloud hosting. The practical consequence for anyone using this archive is that the code in the repository is not self-contained, and you have to fetch the data from a separate Chinese cloud service to run the experiments.

The second is from the ModelSim simulation lab, experiment six, updated on 2024-05-17. The warning is that in the last installation step you must not let the installer put the hardware lock in place, because it caused the author's own machine to blue screen repeatedly, and if it happens to you, boot into safe mode and delete hardlock.sys from the machine. This is the kind of note that only exists because somebody hit the problem and wrote it down.

There is also an entry for experiment seven, the intelligent programming language operator lab, noting that the evaluation could not be run on the course platform because of a platform failure, although local tests all passed. So for that lab, the submitted results are unverified by the graders' own infrastructure.

A community-contributed archive with an unresolved licence

The acknowledgements section names ten contributors who raised issues and pull requests, with links to their profiles, and the README asks readers to leave a star if the repository helped them. The Star History chart is embedded at the bottom.

The licence situation is the one thing the repository does not settle. The licence field is one GitHub cannot classify, and unlike the hardware and firmware projects in this set, there is no LICENSE file in the top level directory listing, and the README carries no licence paragraph. What is present instead is the two standard as-is distribution sentences, including the disclaimer that the software is distributed on an AS IS BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND.

For a student using this to pass a course, that is academic. For anyone who intends to reuse the code, repackage the manuals, or build a course from the same structure, it is not, and the right move is to ask the repository owner rather than assume.

The last push to the main branch was on 2026-06-03, one day after the most recent log entry, and there are no GitHub releases. The archive is therefore a moving snapshot tied to a course calendar rather than a versioned series, which is the right shape for its purpose and the wrong shape for anything that needs reproducibility.

Editorial conclusion

Adopt this repository if you are taking the AICS intelligent computing system course, because it is the course's own archive of manuals, assignments and reference answers, or if you are designing a comparable lab sequence and want to see the structure of one that has been through a grading cycle. Do not adopt it as a production codebase, because every experiment is a graded assignment rather than a maintained library, and the datasets are hosted outside the repository rather than versioned with it. Before you rely on it, read the project status notes, which state the content follows the course schedule and was updated to the large model experiments as of 2025-06-11, and note that the repository has no releases and an unrecognised licence field, so check with the repository owner before redistributing any of it.

Frequently asked questions

Official sources

1. Issues
2. README
3. Yuichi1001/2024-AICS-EXP on GitHub