m1n1
tinygrad
Our great sponsors
m1n1 | tinygrad | |
---|---|---|
17 | 58 | |
3,376 | 17,800 | |
2.2% | - | |
8.9 | 9.7 | |
20 days ago | 10 months ago | |
Python | Python | |
MIT License | MIT License |
Stars - the number of stars that a project has on GitHub. Growth - month over month growth in stars.
Activity is a relative number indicating how actively a project is being developed. Recent commits have higher weight than older ones.
For example, an activity of 9.0 indicates that a project is amongst the top 10% of the most actively developed projects that we are tracking.
m1n1
-
Asahi Linux project's OpenGL support on Apple Silicon officially surpasses Apple
One of the coolest things (IMO) about the entire Asahi effort, and why I'm not at all surprised that they surpassed Apple, was the dedicated effort to build bespoke developer-friendly Python tooling early in the reverse engineering process.
https://asahilinux.org/2021/08/progress-report-august-2021/
> Since the hypervisor is built on m1n1, it works together with Python code running on a separate host machine. Effectively, the Python host can “puppeteer” the M1 and its guest OS remotely. The hypervisor itself is partially written in Python! This allows us to have a very fast test cycle, and we can even update parts of the hypervisor itself live during guest execution, without a reboot.
> We then started building a Python implementation of this RPC protocol and marshaling system. This implementation serves a triple purpose: it allows us to parse the DCP logs from the hypervisor to understand what macOS does, it allows us to build a prototype DCP driver entirely in Python, and it will in the future be used to automatically generate marshaling code for the Linux kernel DCP driver.
Code here: https://github.com/AsahiLinux/m1n1/blob/main/proxyclient/m1n...
If you watch any of Asahi Lina's streams from the time before they had working drivers, she's able to weave together complex bitflag-manipulating pipelines at the speed of thought with self-documenting code, all in Python running on the host machine, all while joking with viewers via her adorable avatar. I've never seen anything like it before. The whole workflow is a tremendous and unprecedented accomplishment by the entire Asahi team.
-
Strange (scheduling?) latency on the host when KVM guest runs something demanding
I wrote a m1n1 experiment to test IRQ delivery in EL1 and noticed something was weird. I already knew about that IRQ control register (3 masks IRQs entirely and is the default, that whole thing took like a day or two back when I first added M1 Pro/Max support), so I tried other values and 2 fixed it.
-
Apple Silicon - iBoot
They have not dropped hardly any official documentation on iBoot. The best safe documentation is in: https://github.com/AsahiLinux/m1n1/ There's a lot of tainted docs out there because the source to iBoot was illegally leaked a while back, but marcan is known for being a bit of a hardass when it comes to legal reverse engineering (thankfully).
- Everything we know about the Apple Neural Engine (ANE)
-
Dumb question - is it possible to install Windows on top of Asahi?
No its uefi boot manager “https://github.com/AsahiLinux/m1n1 “
-
Updates galore! November 2022 Progress Report
What makes you say there isn't community participation? The repo for m1n1, at least, has 42 contributors according to Github[1]. There's plenty more reporting bugs and such, and their IRC channel seems relatively active.
1: https://github.com/AsahiLinux/m1n1
-
A Secret Apple Silicon (M1) Extension to Accommodate an Intel 8080 Artifact
I can confirm that’s pretty much exactly what happens. I did some digging with the help of m1n1 a while back, and essentially yeah, this is almost exactly what it does. The only difference is that, rather than tracking which cores are running emulated code, the scheduler keeps track of which processes are running in emulation mode, and prior to returning to userland after a context switch, the kernel sets control register bits for features like TSO (which is what I was interested in looking into at the time; I believe that specifically is controlled somewhere in the actlr_el1 register). Although only the P-cores actually implement the necessary x86 emulation behavior, the scheduler of course wants to ensure that a process is not pinned to any one core, and that native and emulated processes alike can preempt and interleave with each other on the P-cores just as they would if there were no emulation.
-
Asahi Lina (Linux Developer VTuber) wants to write the new Apple Silicon GPU driver for Linux in Rust!
That shim/stub is m1n1. It is the bootloader for Asahi Linux, but also makes it possible to talk to the hardware over USB as just described by Lina. marcan even implemented a small hypervisor in m1n1, so it can be used to run MacOS and trace how MacOS is accessing the hardware.
-
questions about the new architecture
Windows 11 ARM dualboot will come to M1 macs in the near future. See m1n1 for progress.
- First triangle ever rendered on an M1 Mac with a fully open-source driver
tinygrad
- tinygrad: extreme simplicity, easiest framework to add new accelerators to
-
GGML – AI at the Edge
Might be a silly question but is GGML a similar/competing library to George Hotz's tinygrad [0]?
[0] https://github.com/geohot/tinygrad
-
Render neural network into CUDA/HIP code
at first glance i thought may its like tinygrad. but looks has many ops than that tiny grad but most maps to underlying hardware provided ops?
i wonder how well tinygrad's apporach will work out, ops fusion sounds easy, just a walk a graph, pattern match it and lower to hardware provided ops?
Anyway if anyone wants to understand the philosophy behind tinygrad, this file is great start https://github.com/geohot/tinygrad/blob/master/docs/abstract...
-
llama.cpp now officially supports GPU acceleration.
There are currently at least 3 ways to run llama on m1 with GPU acceleration. - mlc-llm (pre-built, only 1 model has been ported) - tinygrad (very memory efficient, not that easy to integrate into other projects) - llama-mps (original llama codebase + llama adapter support)
- George Hotz building an AMD competitor to Nvidia.
-
George Hotz ROCm adventures
Hopefully we will see now full support with AMD hardware on https://github.com/geohot/tinygrad. You can read more about it on https://tinygrad.org/
-
The Coming of Local LLMs
tinygrad
https://github.com/geohot/tinygrad/tree/master/accel/ane
But I have not tested it on Linux since Asahi has not yet added support.
llama.cpp runs at 18ms per token (7B) and 200ms per token (65B) without quantization.
- Everything we know about Apple's Neural Engine
- Everything we know about the Apple Neural Engine (ANE)
- How 'Open' Is OpenAI, Really?
What are some alternatives?
HelloSilicon - An introduction to ARM64 assembly on Apple Silicon Macs
Pytorch - Tensors and Dynamic neural networks in Python with strong GPU acceleration
unix-history-repo - Continuous Unix commit history from 1970 until today
llama.cpp - LLM inference in C/C++
pdp7-unix - A project to resurrect Unix on the PDP-7 from a scan of the original assembly code
openpilot - openpilot is an open source driver assistance system. openpilot performs the functions of Automated Lane Centering and Adaptive Cruise Control for 250+ supported car makes and models.
freebsd-src - The FreeBSD src tree publish-only repository. Experimenting with 'simple' pull requests....
llama - Inference code for Llama models
rss-proxy - RSS-proxy allows you to do create an RSS or ATOM feed of almost any website, just by analyzing just the static HTML structure.
tensorflow_macos - TensorFlow for macOS 11.0+ accelerated using Apple's ML Compute framework.
nixos-apple-silicon - Resources to install NixOS bare metal on Apple Silicon Macs
GPTQ-for-LLaMa - 4 bits quantization of LLaMA using GPTQ