aws-ai-ml-workshop-kr: A Korean-Language AWS AI/ML Lab Collection

What problem a Korean-language AWS lab collection solves

Most AWS machine learning workshops are written in English, and the gap is not vocabulary. It is the service console labels, the error messages learners hit while a training job fails, and the instructor's ability to explain a hyperparameter while the notebook is still running. This repository is AWS's attempt to close that gap for Korean-speaking audiences, and it is aimed at a narrow group: instructors, solutions architects, and internal enablement teams who run hands-on sessions and need the notebooks to already be in Korean. The README states the repository holds AWS AIML code and workshop examples, organized under three headings: Generative AI, SageMaker AI, and AWS Neuron. That structure tells you the intended audience is broad within machine learning but specific in platform, since all three tracks assume you are working inside AWS. A learner who wants a general PyTorch tutorial will find nothing here. A team preparing a two-day internal workshop on SageMaker training jobs, or on deploying models to Inferentia and Trainium, is the actual target.

Three tracks, and what each directory contains

The repository is a container for three separate curricula rather than one coherent course. The generative AI track lives under genai/aws-gen-ai-kr and is split into three numbered subdirectories: 20_applications for service application examples, 30_fine_tune for training and fine-tuning, and 40_inference for model deployment. Numbering them in that order (applications before fine-tuning before inference) is an editorial choice worth noting: the workshop appears to lead with what a model can do before showing how to adapt or host it. The SageMaker track sits under sagemaker/ and documents two levels, 01-sagemaker-101 described in the README as an introduction to SageMaker training, and a hyperpod directory for model training on SageMaker HyperPod. The third track, neuron/, covers AWS Neuron, which the README glosses as Inferentia, Inferentia2, and Trainium, the custom silicon AWS uses for inference and training acceleration. Each track has its own Readme.md, and the top-level README explicitly directs readers there: it says to refer to the Readme file in each directory. That is the honest reading of the repository. The root README is an index, not documentation.

How the labs are actually delivered

The primary language is Jupyter Notebook, which sets the delivery model. Learners open a notebook, execute cells in order, and read Korean markdown between code blocks. That format works well for instructor-led sessions where a facilitator can pause on a failing cell, and poorly for self-paced study, because notebook cells that call AWS APIs depend on account state: an IAM role that exists, a bucket that has been created, a service quota that has been raised. The repository's contribution section acknowledges the maintenance problem indirectly. It says the maintainers are still working on the best mechanism to accept examples from external sources and asks contributors to bear with them if pull requests take longer than expected or are closed. For a workshop repository, that is a reasonable position, since a merged notebook that no longer runs is worse than a closed pull request. It also means you should treat the notebooks as snapshots. There are no releases listed for this repository, so there is no versioned artifact to pin. The default branch is master, and the last push recorded is 2026-08-04, which tells you the repository is active but gives no signal about which notebooks were touched in that push.

Getting a lab running: what the material specifies

The material does not provide a single setup path, and this is a real friction point. There is no top-level requirements.txt, no Dockerfile, and no bootstrap script described in the README. What you get instead is a pointer: clone the repository, then open the README inside the track you intend to teach. The commands implied by the structure are ordinary ones. You would clone with git clone https://github.com/aws-samples/aws-ai-ml-workshop-kr, change into the track directory, for example cd sagemaker/01-sagemaker-101 or cd genai/aws-gen-ai-kr/40_inference, and then follow that folder's Readme.md for the specific steps. Because the notebooks target SageMaker, HyperPod, and Neuron, the realistic execution environment is SageMaker notebook instances or SageMaker Studio rather than a laptop, since the labs assume an execution role with access to those services. The repository does not state which Python version, which SageMaker SDK version, or which AWS region the notebooks were validated against, and I cannot confirm those from the supplied material. Treat environment discovery as the first task of any pilot session, not an afterthought.

The licence line contradicts the repository metadata

The README's License section says the library is licensed under the Apache 2.0 License and points to the LICENSE file. The repository metadata supplied for this review lists MIT-0. Those are different licences with different obligations, and the README also carries a badge reading LICENSE-MIT. This is not a subtle distinction: Apache 2.0 includes an explicit patent grant and requires preservation of notices, while MIT-0 is a permissive licence that drops the attribution requirement entirely. For a workshop repository the practical stakes are low, since you are copying teaching notebooks rather than shipping a product, but the discrepancy matters if your organization's policy engine reads licence metadata and blocks or flags repositories on that basis. The only reliable resolution is to open the LICENSE file in the repository root and read it directly. Do not rely on the README sentence, the badge, or the metadata field alone, and do not treat this paragraph as legal advice. It is a note that the three sources disagree and that the file itself is the one that governs.

Where this repository is the wrong tool

The clearest failure mode is using these notebooks as a starting point for production code. Workshop notebooks are written to be readable in sequence in front of an audience. They tend to inline configuration, hardcode resource names, and skip error handling because a stack trace in a live session is a teaching moment, not a defect. Lifting a fine-tuning notebook from genai/aws-gen-ai-kr/30_fine_tune into a pipeline without rewriting the configuration and IAM assumptions will produce something that runs once in your account and then breaks when a resource is renamed. The second mismatch is language. Every piece of prose in the repository is Korean, and the README, directory names, and notebook markdown all reflect that. If your team does not read Korean, the code cells are still usable but the explanatory value, which is the entire reason to choose this repository over a generic AWS sample, is gone. The third case is scope. A team that wants to learn machine learning concepts rather than AWS services will find the platform assumptions get in the way. The repository teaches you how to use SageMaker, HyperPod, and Neuron. It does not teach you why a transformer works.

What to use instead, and how the approach differs

The natural alternative is the English-language aws-samples ecosystem, and specifically the SageMaker example repositories and the Amazon SageMaker Examples collection that these Korean workshops are localized from. The difference is not quality but maintenance topology. The English repositories are the upstream source, so when an AWS API changes, the fix lands there first. A localized fork inherits that change only when a maintainer updates the Korean notebook, and this repository has no release cadence to signal when that happens. The second alternative is a documentation-first route: AWS's own workshop studio and the SageMaker developer guide, which are versioned and updated alongside the services. Those give you correct, current instructions but no runnable Korean notebooks and no classroom pacing. The trade-off is consistent across both alternatives. If you need Korean-language, hands-on, sequence-ready labs and you have an instructor who can absorb environment drift, this repository saves you weeks of translation. If you need something that tracks AWS API changes automatically, the upstream English material is the safer base, and you accept the translation work as your own cost.

Maintenance cost and what to verify before class

The maintenance burden here is front-loaded into preparation, not into ongoing dependency updates, because there is nothing to depend on. No package is published, no release is tagged, and no version is pinned. Your cost is the time to walk every notebook in the track you plan to teach against a live account before the session, which for a two-day workshop is a meaningful block of instructor hours. The contributing section's warning that pull requests may take longer than expected or be closed is a signal about how quickly your own fixes would flow back upstream, so plan to maintain a private fork if you need corrections. On licensing, verify the LICENSE file against your organization's policy before redistributing the notebooks internally, given that the README says Apache 2.0 while the metadata says MIT-0. The concrete next step is small and specific: clone the repository, open sagemaker/Readme.md and neuron/Readme.md, and check whether the instance types and services those files reference are available in the region where you intend to run the class. If they are not, the workshop does not run, regardless of how good the Korean prose is.

Editorial conclusion

Adopt this repository if you are delivering Korean-language AWS AI/ML training and want notebook labs that already map to SageMaker, Bedrock-era generative AI services, and Neuron chips, rather than translating English workshops yourself. Do not adopt it as a dependency, a production template, or a source of maintained APIs: there are no releases, the README's licence sentence contradicts the MIT-0 metadata, and the contribution section warns that pull requests may take longer than expected or be closed. Before scheduling a class, open the README inside the specific track directory you plan to teach (sagemaker/Readme.md or neuron/Readme.md) and confirm which services and instance types the notebooks in that folder actually reference, because the top-level README only names directories, not versions or regions.

Official sources

1. aws-samples/aws-ai-ml-workshop-kr on GitHub
2. Issues
3. License: MIT-0
4. README