What needs improvement with Amazon SageMaker?

The user interface (UI) and user experience (UX) of SageMaker and AWS, in general, need improvement as they are not intuitive and require substantial time to learn how to use specific services. Additionally, the platform struggles with tuning large models, which is a significant limitation.

There are two areas in SageMaker that need improvement. Firstly, when starting a new session, the waiting time can be quite long, ranging from two to five minutes. Secondly, the integration with Snowflake is not optimal. I face challenges in authentication, requiring multiple steps to achieve single sign-on. This process feels somewhat cumbersome.

While not specific to SageMaker, I've observed a trend where companies are more attracted to the marketing of Gemini and GPT, leading some to migrate their pipelines to these platforms. Amazon might need to emphasize its capabilities in generative models more effectively.

While integration is available, there are concerns about how secure this integration is, particularly when exposing data to SageMaker. Enhancing encryption and overall security could uplift the platform and build more trust among users. Another area of improvement is the cost, which can become quite high during prolonged use.

One area where Amazon SageMaker could improve is its pricing. The high costs can drive companies to explore other cloud options. Additionally, while generally good, the updates sometimes come with bugs, and the documentation could be much better. More examples and clearer guidance would be helpful.

For any cloud provider, the cost has to be substantially reduced, especially in the case of Amazon SageMaker, which is extremely expensive for huge workloads. In EC2, you have spot instances that cut costs tremendously, but you don't have that in Amazon SageMaker. You pay for the local usage. I would like to see better integration with GPUs. GPUs are very expensive for AWS or any cloud provider. NVIDIA has introduced options with Databricks for GPUs, so it would be interesting to see how Amazon SageMaker can parallelize GPU usage. I haven't used it to scale multiple GPUs automatically for model training. The key points are the cost and how effectively they integrate GPUs into the workload for training machine learning models. We want to see how seamless it is and how it can work. I haven't used multiple GPUs scaled automatically. For model training, the first concern is cost, and the second is how effectively they want to integrate GPUs into the workload for training machine learning models.

Amazon SageMaker can make it simpler to manage the data flow from start to finish, such as by integrating data, usingthe machine, and deploying models. This process could be more user-friendly compared to other tools. I would also like to improve integration with Bedrock and the LLM connection for AWS.

In my opinion, one improvement for Amazon SageMaker would be to offer serverless GPUs. Currently, we incur costs on an hourly basis. It would be beneficial if the tool could provide pay-as-you-go pricing based on endpoints. In the three months I've been using it, I've noticed that higher GPU instances can be quite costly. To mitigate this cost impact, serverless GPUs would be beneficial.

The product must provide better documentation. I don't see a lot of documentation, particularly on the Studio feature. In general, there is not a lot of information about how to use Feature Store. I can see it there, but the documents are not very explanatory.

The training modules could be enhanced. We had to take in-person training to fully understand SageMaker, and while the trainers were great, I think more comprehensive online modules would be helpful. Additionally, the user manuals can be difficult to navigate without prior knowledge. We often test new features for clients in small groups, and I've heard feedback that the documentation could be more user-friendly.

I feel that the area around the interface in AWS is overall confusing. The payment and monitoring metrics are a bit confusing not only for Amazon SageMaker but also for the range of other products that fall under AWS, especially for a new user of the product. The tool is not simplified enough for beginners to use. From an improvement perspective, the tool needs to be simplified enough for beginners to use.

The solution is complex to use. Some additional functionality would be for them to provide sample end-to-end card formation templates and try to unify the setup. At the moment, as you move from one set of documentation to the next, some of the documentation is for bringing your model, some of the documentation is for SDK, some of it's for API, some of it's for command lines, and some of it's for step functions. None of the documentation seems to be end-to-end. There are gaps in the documentation. It proves to require a lot of digging from the user to figure it out. I did get through the AWS machine learning specialty certification, but that proved to be a bit superficial. Though it covered a lot of ground, it didn't have the detail one would need for Sagemaker. I am self-taught in a lot of the stuff. I could dive deeper into some code and take time to get examples running. But I was consulting a startup, and they needed to move quickly. I was hoping SageMaker would be easier to work with because I was expecting there would be examples we could repurpose that were more complete. The new functionality I'd like to see is Amazon tuning attention to the documentation sets and the templates.

Creating notebook instances for multiple users is pretty expensive in Amazon SageMaker.

Amazon SageMaker should concentrate and get the performance of the ensemble model to be good enough for its users. Improvements are needed in terms of performance for not all but some of the models, especially whenever we use the product for image classification or something. In general, improvements are needed on the performance side of the product's graphical user interface-related area since it consumes a lot of time for a user.

Amazon SageMaker is expensive. It is more expensive than Databricks because it has a pay-as-you-go model with SageMaker. The pricing of the solution is an issue. In SageMaker, the monitoring does not support data captured in a protocol called protobuf since it only accepts JSON and CSV. We use protobuf to exchange messages. We have to write our own logic to serialize protobuf, which is more work, and it'd be better if SageMaker provided out-of-the-box support for that. In SageMaker, monitoring could be improved by supporting more data types other than JSON and CSV. Amazon SageMaker can start with Protobuf since it is a popular messaging protocol.

The solution needs to be cheaper since it now charges per document for extraction.

There are other better solutions for large data, such as Databricks.

SageMaker would be improved with the addition of reporting services. In addition, the models available in SageMaker are not enough for most of our use cases and require customization to be useful.

The product has come a long way and they've added a lot of things, but in terms of improvement I would like to probably have features such as MLflow embedded into it. Additional features I would like to see would include, as mentioned, MLflow and ML Pipelines which are more of a feature rich support of machine learning pipelines as well as scheduling machine learning pipelines, and visualization of machine learning pipelines.

AI is a new area and AWS needs to have an internship training program available. This is one place where I see this solution lagging. There is high-level training available, but when you consider that people have been working with Windows, Linux, and various applications for the past 20 years, they know those products inside and out. SageMaker, on the other hand, is a completely new tool. It can be very hard to digest. AWS needs to provide more use cases for SageMaker. There are some, but not enough. They should collect or create more use cases and then distribute them free of charge to the customers. I would like to see a more graphical, low-code interface that can be used to customize SageMaker.

The interface and the IDE are in need of improvement. For example, including drag and drop functionality would be helpful. If the ETL can be made a little better then that would be good for us. The entire machine learning project flow, or data science project flow, can be a little better. It is good but it would benefit from more machine learning options, making it really good. Scalability to handle big data can be improved by making integration with networks such as Hadoop and Apache Spark easier. Adding some AI functionality, similar to what DataRobot or Azure AI has, would be really great.

The pricing is complicated and should be simplified. I would suggest that Amazon SageMaker provide free slots to allow customers to practice, such as a free slot to try out working with a Sandbox. This would be beneficial for newcomers, especially those who are getting into the cloud space. They could explore this area and get all of the aspects including data engineering, data recognition, and data transformation.

I would say the IDE is quite immature, but it is still in its infancy, so I expect it to get better over time.