highway VS llamafile

The bf16 dot instruction replaces 6 instructions: https://github.com/google/highway/blob/master/hwy/ops/x86_12...

[0] for those interested in Highway.

It's also mentioned in [1], which starts off

> Today we're sharing open source code that can sort arrays of numbers about ten times as fast as the C++ std::sort, and outperforms state of the art architecture-specific algorithms, while being portable across all modern CPU architectures. Below we discuss how we achieved this.

[0] https://github.com/google/highway

[1] https://opensource.googleblog.com/2022/06/Vectorized%20and%2..., which has an associated paper at https://arxiv.org/pdf/2205.05982.pdf.

Thanks so much!

Everyone working on this self-selected into contributing, so I think of it less as my team than ... a team?

Specifically want to call out: Jan Wassenberg (author of https://github.com/google/highway) and I started gemma.cpp as a small project just a few months ago + Phil Culliton, Dan Zheng, and Paul Chang + of course the GDM Gemma team.

C++ users can enjoy Highway [1].

[1] https://github.com/google/highway/

At that point it is better to have some kind of DSL that should not be in the main language, because it would target a much lower level than a typical program. The best effort I've seen in this scene was Google's Highway [1] (not to be confused with HighwayHash) and I even once attempted to recreate it in Rust, but it is still distanced from my ideal.

[1] https://github.com/google/highway

Interesting, thanks for sharing :)

At the time we open-sourced Highway, the standardization process had already started and there were some discussions.

I'm curious why stdlib is the only path you see to default? Compare the activity level of https://github.com/VcDevel/std-simd vs https://github.com/google/highway. As to open-source usage, after years of std::experimental, I see <200 search hits [1], vs >400 for Highway [2], even after excluding several library users.

But that aside, I'm not convinced standardization is the best path for a SIMD library. We and external users extend Highway on a weekly basis as new use cases arise. What if we deferred those changes to 3-monthly meetings, or had to wait for one meeting per WD, CD, (FCD), DIS, (FDIS) stage before it's standardized? Standardization seems more useful for rarely-changing things.

1: https://sourcegraph.com/search?q=context:global+std::experim...

2: https://sourcegraph.com/search?q=context:global+HWY_NAMESPAC...

Thanks for the link. We were previously using GFNI for bit reversal and 8-bit shifts, and I just extended that to our 8-bit BroadcastSignBit (https://github.com/google/highway/pull/1784).

You could study Google's Highway library [1].

[1] https://github.com/google/highway

Runtime feature detection need not be rare nor hard, it's a few dozen lines of boilerplate. You can even write your code just once: see https://github.com/google/highway#examples.

I think the llamafile[0] system works the best. Binary works on the command line or launches a mini webserver. Llamafile offers builds of Mixtral-8x7B-Instruct, so presumably they may package this one up as well (potentially a quantized format).

You would have to confirm with someone deeper in the ecosystem, but I think you should be able to run this new model as is against a llamafile?

[0] https://github.com/Mozilla-Ocho/llamafile

Thermostats: https://www.sinopetech.com/en/products/thermostat/

I haven't tried running a local text-to-speech engine backed by an LLM to control Home Assistant. Maybe someone is working on this already?

TTS: https://github.com/SYSTRAN/faster-whisper

LLM: https://github.com/Mozilla-Ocho/llamafile/releases

LLM: https://huggingface.co/TheBloke/Nous-Hermes-2-Mixtral-8x7B-D...

It would take some tweaking to get the voice commands working correctly.

While I did not succeed in making the matmul code from https://github.com/Mozilla-Ocho/llamafile/blob/main/llamafil... work in isolation, I compared eigen, openblas, and mkl: https://gist.github.com/Dobiasd/e664c681c4a7933ef5d2df7caa87...

In this (very primitive!) benchmark, MKL was a bit better than eigen (~10%) on my machine (i5-6600).

Since the article https://justine.lol/matmul/ compared the new kernels with MLK, we can (by transitivity) compare the new kernels with Eigen this way, at least very roughly for this one use-case.

Yes, they're just ZIP files that also happen to be actually portable executables.

https://github.com/Mozilla-Ocho/llamafile?tab=readme-ov-file...

have you seen llamafile[0]?

[0] https://github.com/Mozilla-Ocho/llamafile

llama.cpp has integrated gemma support. So you can use llamafile for this. It is a standalone executable that is portable across most popular OSes.

https://github.com/Mozilla-Ocho/llamafile/releases

So, download the executable from the releases page under assets. You want either just main or just server. Don't get the huge ones with the model inlined in the file. The executable is about 30MB in size,

https://github.com/Mozilla-Ocho/llamafile/releases/download/...

The improvements in ease of use for locally hosting LLMs over the last few months have been amazing. I was ranting about how easy https://github.com/Mozilla-Ocho/llamafile is just a few hours ago [1]. Now I'm torn as to which one to use :)

1: Quite literally hours ago: https://euri.ca/blog/2024-llm-self-hosting-is-easy-now/

Slightly off topic, here is the best local llama.cpp wrapper I've run into:

https://github.com/Mozilla-Ocho/llamafile

You can download any .gguf model (not just the ones in their examples) and run it locally (as long as you have the ram for it). I was running 7B models with ease on an old FX8350 and now 13B models on a 5600X (32GB RAM on both machines).

This wrapper spins up a local web server that runs a simple web frontend to use immediately with no code, but also exposes an OpenAI compatible API for dev work and alt frontends (like SillyTavern).