coremltools VS msquic

Instead of trying to integrate the whole stack of, say, pytorch, Apple's primary approach has been converting models to work with Apple's stack.

https://github.com/apple/coremltools

Clearly no one is going to be doing training or even fine tuning on Apple hardware at any scale (it competes at the low end, but at scale you invariably will be using nvidia hardware), but once you have a decent model it's a robust way of using it on Apple devices.

There is one guy who was able to run it on iOS. See this thread for more information. Basically, the idea is to convert torch models to CoreMl. Only the CLIP tokenizer's implementation is currently missing. I guess this guy will keep modifications private, but he is trying to optimize model for lower RAM requirements.

Afaik (correct me if I’m wrong) both PyTorch and tensorflow only use the gpu when training and not the neural engine. I think the neural engines can be used for inference if the model is in the CoreML format (https://github.com/apple/coremltools)

CoreML Tools

And Python.

There's literally an Apple provided tool, called [coremltools[(https://github.com/apple/coremltools) to convert many common PyTorch and TensorFlow models to CoreML.

I referred to sockets as an API design, not to express an opinion on whether you should place your protocol implementations inside or outside the kernel. (Although that’s undeniably an interesting question that by all rights should have been settled by now, but isn’t.)

Even then, I didn’t mean you should reproduce the Berkeley socket API verbatim (ZeroMQ-style); multiple streams per connection does not sound like a particularly good fit to it (although apparently people have managed to fit SCTP into it[1]?). I only meant that with the current mainstream libraries[2,3,4], establishing a QUIC connection and transmitting bytestreams or datagrams over it seems quite a bit more involved than performing the equivalent TCP actions using sockets.

[1] https://datatracker.ietf.org/doc/html/rfc6458

[2] https://quiche.googlesource.com/quiche

[3] https://github.com/microsoft/msquic

[4] https://github.com/litespeedtech/lsquic

The documentation of MS QUIC says it is cross-platform, it should work on Linux, it has a CMake preset for Linux and you can download the prebuilt binary releases for Linux.

Hello HN, my name is Varun, and I am the co-founder of StepSecurity. Here is the backstory about Harden-Runner. We thoroughly researched past software supply chain security incidents. The devastating breaches of SolarWinds, Codecov, and others, have one thing in common – they attacked the CI/ CD pipeline or the build server.

These incidents made it clear that a purpose-built security agent was needed for CI/ CD. While there are numerous agents available for desktops and servers, such as from CrowdStrike and Lacework, none have been tailored specifically to address the unique risks present in CI/CD pipelines.

With the understanding that a specialized solution was needed to secure CI/CD environments, we developed Harden-Runner, an open-source solution tailored specifically for GitHub Actions hosted runners. It can be seamlessly integrated into your workflow by simply adding a step. The agent installation process is also lightning-fast, taking no more than 5 seconds to complete.

Harden-Runner's security agent is designed to closely monitor all aspects of the workflow run, including DNS, network, file, and process events. This allows for real-time identification of any potential security breaches. To prevent incidents like the Codecov breach, where exfiltration of credentials occurred, Harden-Runner allows you to set policies that restrict outbound traffic at both the DNS and network layers. Additionally, we are actively working on implementing further restrictions at the application layer, such as using HTTP verbs and paths, to provide an even more comprehensive security solution.

An excellent example of how Harden-Runner effectively blocks outbound traffic can be found in the following link: https://app.stepsecurity.io/github/microsoft/msquic/actions/.... As you can see, all traffic to unauthorized endpoints is highlighted in red, indicating that it has been blocked; this is because these endpoints are not included in the allowed list defined in the GitHub Actions workflow file, which can be viewed here: https://github.com/microsoft/msquic/blob/aaecb0fac5a3902dd24....

One of the key features of Harden-Runner's monitoring capabilities is its ability to detect any tampering or alteration of files during the build process, similar to the SolarWinds incident. To further enhance security and protect against potential malicious tools or attempts to disable the agent, Harden-Runner includes a disable-sudo mode. This mode effectively disables the use of 'sudo' on the hosted runner, providing an additional layer of protection

Harden-Runner has already been adopted by over 600 open-source repositories: https://github.com/step-security/harden-runner/network/depen.... To fully understand the capabilities of Harden-Runner and how it can protect against past supply chain attacks, please try out our attack simulator GitHub repository at https://github.com/step-security/attack-simulator. I would love to hear your feedback.

https://github.com/microsoft/msquic (QUIC / HTTP3)