We initially wrote this for a less technical audience (where we spelled out MCP), then edited it to post here - it's not AI, it's just bad editing from my part. Fixed now.
The only Beam-specific part are the sandboxes, but those can easily be swapped out for the vendor of your choice. The architecture we described isn't exclusive to our product.
Beam is an ultrafast AI inference platform. We built a serverless runtime that launches GPU-backed containers in less than 1 second and quickly scales out to thousands of GPUs. Developers use our platform to serve apps to millions of users around the globe.
We’re working on challenging problems, including:
* Low-level systems development: working with container runtimes, OCI image formats, and lazy-loading large files from content addressable storage
* Efficiently packing thousands of workloads into GPUs across multiple clouds
* Working with cutting-edge technologies, like GPU checkpoint restore and CRIU
You don’t need prior experience with AI/ML, only an interest in working on hard problems and shipping quickly.
You should look into beam.cloud (I'm the founder, but it's pretty great)
It lets you quickly run long-running jobs on the cloud by adding a simple decorator to your Python code:
from beam import function
# Some long training function
@function(gpu="A100-80")
def handler():
return {}
if __name__ == "__main__":
# Runs on the cloud
handler.remote()
You should checkout beam.cloud (I'm the founder). It's a modern FaaS platform for Python, with support for REST endpoints, task queues, scheduled jobs, and GPU support.
The major clouds don't support serverless GPU because the architecture is fundamentally different from running CPU workloads. For Lambda specifically, there's no way of running multiple customer workloads on a single GPU with Firecracker.
A more general issue is that the workloads that tend to run on GPU are much bigger than a standard Lambda-sized workload (think a 20Gi image with a smorgasbord of ML libraries). I've spent time working around this problem and wrote a bit about it here: https://www.beam.cloud/blog/serverless-platform-guide
beam.cloud | Founding Software Engineer, Infrastructure | Full-time | REMOTE | New York, NY USA
Beam is building a cloud runtime for running remote containers on GPUs. We’re used by thousands of developers for powering their generative AI apps, including companies like Coca-Cola, and we’re backed by great investors like YC and Tiger.
We’re building gnarly low-level distributed systems. You’ll have a major impact on the product and ship features directly to users. If working on a new Pythonic cloud runtime sound exciting, you might really like it here.
beam.cloud | Founding Software Engineer, Infrastructure | Full-time | REMOTE | New York, NY USA
Beam is building a cloud runtime for running remote containers on GPUs. We’re used by thousands of developers for powering their generative AI apps, including companies like Coca-Cola, and we’re backed by great investors like YC and Tiger.
We’re building gnarly low-level distributed systems. You’ll have a major impact on the product and ship features directly to users. If working on a new Pythonic cloud runtime sound exciting, you might really like it here.
There are a number of good options here. The different axes are cost of GPUs, performance, and ease of use / developer experience. You might consider beam.cloud (I'm one of the founders), which is oriented strongly on the performance and developer experience angle.
Cloudflare AI and Replicate are great for running off-the-shelf models, but anything custom is going to incur a 10+ minute cold start.
For running custom fine-tuned models on serverless, you could look into https://beam.cloud which is optimized for serving custom models with extremely fast cold start (I'm a little biased since I work there, but the numbers don't lie)
Serverless only works if the cold boot is fast. For context, my company runs a serverless cloud GPU product called https://beam.cloud, which we've optimized for fast cold start. We see Whisper in production cold start in under 10s (across model sizes). A lot of our users are running semi-real time STT, and this seems to be working well for them.
Sharing GPUs only really makes sense for GPUs that are large enough to share. MIGs can work for 80Gi A100s but won't work with smaller cards like T4s. It also adds latency to the GPU operations. Unfortunately there's not yet a silver bullet for this stuff.
What are you using for K8s autoscaling? We initially tried a few standard K8s scaling mechanisms and found that they didn't work well for GPU workloads. For example, if we were serving a low-RAM Huggingface model on GPU, it wouldn't trigger autoscaling. But since the GPU can only process one request at a time, the system would get bottlenecked while it waited to process requests one-by-one.
Slai is a tool to quickly build machine learning-powered applications. Our browser-based sandbox is the easiest way to build, deploy, and share machine learning models with zero setup [1]. We’re currently a team of four, and we’re looking to hire someone to help us with SRE / DevOps work.
You should have experience setting up and maintaining infrastructure at scale - ideally, you’ll have fluency with Docker/Kubernetes, EKS, KNative, Terraform, Terragrunt, Gunicorn, and Python. You should be able to communicate clearly in English, and you can work from anywhere (although if you prefer to work in-person, we have an office in Cambridge, MA)
We have a hackathon-inspired culture – the team works in one week sprints and has very few meetings besides daily standup and a Friday afternoon all-hands. If this interests you, please send a brief email with your resume to eli at slai dot io.
On paper, Sagemaker does everything, but they don’t do many of those things well. I think Sagemaker is a great product for enterprises who want to maximize the products procured from a single vendor — it’s easy to buy Sagemaker when all of your infra is already on AWS.
It’s fairly painful to productionize a model on Sagemaker — they make you think about a lot of things and fit into AWS primitives. Besides the code for the model, we don’t force users to think about anything. Our focus is helping engineers get models into production, not reading documentation.
Using our tool, you can fork a model and deploy it to production right away — there’s no time spent battling AWS primitives. We’re focused on developer experience above everything else - which means we enforce sandboxes on our platform to be consistent and reproducible.