I write code that works, sometimes. runtime @modal

It's only a sandbox if it comes from the gVisor region of userspace, otherwise it's just a sparkling container
2
2
111
5,297
Crazy to me that you can get 500+ cores on a single system these days
3
1
15
1,629
FYI
The secret to Modal's speed is our custom lazily loaded container filesystem, ImageFS. ImageFS is built on a massive tiered distributed filesystem, layering local NVMe disks, regional and p2p caches, and global blob stores. When we launched VM Sandboxes this month, we wanted to make sure they could take advantage of ImageFS in the same way our gVisor containers do. To do that, we built out a new filesystem stack that tricks a VM's ext4 driver into thinking it's talking to a normal disk.
1
2
50
2,936
If I give a model access to a Pangram MCP would it successfully iterate until Pangram thought the text was human 🧐
9
73
8,171
If you wanna do cool things with BPF, Linux kernel, and/or networking I got a guy, at a place (im the guy, modal.jobs is the place)
Scaling compute across global sites requires precise, process-level network attribution. In our latest post, Eric breaks down how we combined Linux eBPF kernel tracing, service logs, and hardware counters to map every byte back to the right workflow. Read on HRT Beat: hudsonrivertrading.com/hrtbe…
12
12
404
27,147
I'm not exaggerating in the slightest when I say we're doing some of the coolest systems work out there. There's more where this came from:
The secret to Modal's speed is our custom lazily loaded container filesystem, ImageFS. ImageFS is built on a massive tiered distributed filesystem, layering local NVMe disks, regional and p2p caches, and global blob stores. When we launched VM Sandboxes this month, we wanted to make sure they could take advantage of ImageFS in the same way our gVisor containers do. To do that, we built out a new filesystem stack that tricks a VM's ext4 driver into thinking it's talking to a normal disk.
2
386
The secret to Modal's speed is our custom lazily loaded container filesystem, ImageFS. ImageFS is built on a massive tiered distributed filesystem, layering local NVMe disks, regional and p2p caches, and global blob stores. When we launched VM Sandboxes this month, we wanted to make sure they could take advantage of ImageFS in the same way our gVisor containers do. To do that, we built out a new filesystem stack that tricks a VM's ext4 driver into thinking it's talking to a normal disk.
13
30
543
52,153
Our original (naive) implementation of this -- simply sharing ImageFS to a guest VM -- totally blew up our performance. If you know anything about Modal, this was unacceptable. By rethinking ImageFS in blocks we went two orders of magnitude faster, making for an ultra-fast backbone for our VM Sandboxes.
1
32
4,423
Amit retweeted
Replying to @5rypto
holy fuck i had no idea this dashboard existed we're measuring in petabits/s now???
1
4
631
New goal: one concurrent sandbox for every dollar we raise Who’s in?
Today we're announcing that Modal has crossed $1B in annualized run rate, as well as raised a $750M Series D at a $15.75B valuation, led by @Accel. Since our Series C, usage has tripled and we’ve grown more than 15x year-over-year. We are deeply grateful to the developers and companies building on Modal. There's still so much more to build, and we can't think of a better time to be building software.
1
37
1,368
Amit retweeted
Vera CPUs allow us to offer higher workload concurrency, without sacrificing throughput or latency. More juice per squeeze.
3
10
134
11,170
can't get enough of these visuals 🤤
Training large models with RL typically runs on a fixed homogeneous cluster of the latest accelerators for numerical stability and faster throughput. This throws away inference optimization work done for other vendors and generations, limits the capacity you can utilize, and is unnecessarily expensive. So we trained a model on agentic coding tasks using fully asynchronous RL and mixed-precision, mixed-accelerator samplers. With off-policy correction, training stayed stable through convergence, on par with homogeneous samplers, at a lower cost.
1
1
23
1,022
gosh I hate the stupid slow performance of Go standard library's networking stuff
8
620
The best infra is infra you don't think about. VMs for Sandboxes helps immensely with this. Looking forward to sharing some eng deep dives on the cool tech that the runtime team is 🍳 up
Replying to @modal
VM Sandboxes give your agents access to a full Linux kernel. Reach for VMs when your agents reach the limits of limits of user space: working with Docker, FUSE filesystems, and certain Linux kernel features. modal.com/blog/vm-sandboxes-…
37
2,538
rdma=True is such a high Aura option (also, clusters on @modal are GA!!)
3
5
65
4,627
so
9
1,058
Giving this restaurant a 10.0 on beli because they’re playing Daft Punk
2
12
867