binsquare

przedwczorajdiscuss

it's incredibly repairable. I still use the t480 for modern workloads because I've been able to switch out the battery, the fan, the keyboard with fairly abundant aftermarket parts.

binsquare

4 dni temudiscuss

given that i couldn't even use fable without it downgrading to Opus, this is just a straight upgrade for me

binsquare

9 dni temudiscuss

I also have a light weight cross platform virtual machine with egress filtering if you're looking for an even more batteries included approach.

https://github.com/smol-machines/smolvm

binsquare

14 dni temudiscuss

Yes, been a lot of heads down technical work since.

Hope I can share some interesting things I've worked on and hopefully useful to others

binsquare

14 dni temudiscuss

Yep, there are inference stacks where it just does not work without cuda in any meaningful performance

binsquare

14 dni temudiscuss

I don't think it's rust

binsquare

15 dni temudiscuss

I agree with this, virtual machines are invented to solve the sandboxing/multi-tenant issues.

This is why ec2 and the likes all sell you access to virtual machines (dividing up their underlying hardware).

binsquare

15 dni temudiscuss

If you back it through smolvm, I can take care of the isolation story for you across windows, macOS

https://github.com/smol-machines/smolvm

SDKs available https://github.com/smol-machines/smol

binsquare

15 dni temudiscuss

smol machines: just in time cloud infrastructure

https://www.smolmachines.com

I am building it on top of a new primitive called smolvm: a hybrid that combines isolation of VM with speed and flexibility of containers.

https://github.com/smol-machines/smolvm

binsquare

20 dni temudiscuss

I use a tool called linzumi for this

binsquare

25 dni temudiscuss

yes to both, here's the example for running docker(indirectly, devcontainers) inside smolvm:

https://github.com/smol-machines/smolvm/blob/main/examples/d...

binsquare

25 dni temudiscuss

the underlying vmm is libkrun: https://github.com/libkrun/libkrun is battle tested and used for podman.

It provides kernel isolation for running untrusted code which is a security boundary that traditional containers can't guarantee.

I'm engaged with a third party security penetration company for their review, and will be happy to share it publicly when it is available.

binsquare

25 dni temudiscuss

smol machines is a virtual machine, akin to firecracker but designed to run locally as well

It also has container-inages support built-in with crun so you can create a VM with a container running by default.

You can also just... run docker inside of it.

binsquare

25 dni temudiscuss

author of smol machines here, it has no relation to firecracker.

It runs ontop of the libkrun vmm forked with optimizations, which is the underlying lib powering podman as well.

open source, will contribute upstream when possible: https://github.com/smol-machines/libkrun

binsquare

27 dni temudiscuss

Sony also used to mock digital only approaches.

https://www.youtube.com/watch?v=kWSIFh8ICaA

Now at a time when memory and storage prices are at an all time high - they pull a move that directly hurts the customer.

binsquare

27 dni temudiscuss

Yep, I needed to go deeper down the stack in order to get to the lightweight optimizations I needed.

binsquare

27 dni temudiscuss

I build a local, cross platform virtual machine for this purpose: https://github.com/smol-machines/smolvm

binsquare

29 dni temudiscuss

Fixed and released in v1.3.1: https://github.com/smol-machines/smolvm/pull/497

binsquare

30 dni temudiscuss

smolpi looks great!

and smolvm does clean up ephemeral runs if the machine run exits gracefully. I'll take a deeper look into this edge case and fix it today.

binsquare

30 dni temudiscuss

I made a lightweight vm specifically for this use case: https://github.com/smol-machines/smolvm

Submissions

Smol Machines: Kubernetes-in-a-Microvm

Show HN: Smol machines – subsecond coldstart, portable virtual machines

Microsoft update breaks internet access to Windows 11 Teams, Edge, OneDrive

The ArXiv is separating from Cornell University, and is hiring a CEO for 300k/yr

Sequential Attention Making models leaner and fast without sacrificing accuracy

Qwen3-coder-next: SOTA open source coding model

Qwen launched new open source TTS models

[untitled]

Containrrr/watchtower is now unmaintained

Lessons from building a content scanner for multiple social platforms

New research shows RL may not help a model learn new basic skills

Show HN: Inferbench, collect/share datapoints on GPU's inference performance

Show HN: Env files aren't meant for storing secrets

Show HN: StandbyBro – friends for rent (platonic companionship)

Show HN: Open-source library to get detailed power and system info for Macs

Show HN: Secure Code Execution for AI

Show HN: Play with our Agentic Agentic builder

Show HN: Unified access to top AI models, supporting GPT4, Claude and more

comments