H3-metal – Native MiniMax-H3 inference for Apple Silicon • Hacker News Board

Re: H3-metal – Native MiniMax-H3 inference for Apple Silicon

#101

Post by tracerbulletx » Wed, Aug 12, 2026, 1:18 AM UTC

thedangler wrote: post #78

I'm looking to setup a way to create images for my own instagram marketing. I do not care how long it takes to make 10 variations of a post as that speed would still be faster than me making it. Does this model work with ComfyUI easily? Can I just download it?

Yes, there's a default workflow template in Comfy for that matter now.

Re: H3-metal – Native MiniMax-H3 inference for Apple Silicon

#102

Post by Manfrednotfunny » Wed, Aug 12, 2026, 8:59 AM UTC

antirez wrote: post #40

Manfrednotfunny wrote: post #36

Earlier quoted context omitted.

Memory bandwidtih between pro and max is double. 300gb/s vs. 600gb/s btw.

Does not matter much in this case. GPU bound.

Seems to be true, but also seems hard t obenchmark with max having more GPU cores too.

Re: H3-metal – Native MiniMax-H3 inference for Apple Silicon

#103

Post by Nekit1234007 » Wed, Aug 12, 2026, 9:50 AM UTC

embedding-shape wrote: post #89

Nekit1234007 wrote: post #77

Earlier quoted context omitted.

One thing I read on this topic on reddit is never EVER use the word "girl" when prompting H3. So CSAM probably.

Did you try this yourself? Of course you wouldn't, because not wanting to produce SCAM sorry I meant CSAM. And no, including the word "girl" in H3 does not lead to CSAM in any way, shape or form, but it's a great example how FUD quickly spreads.

I don't have hardware to runs this, so no, I didn't.

One thing that I had experience with, that led me to believe this might be true: it seems one of the earlier llama models was over-tuned to resist generating CSAM. Once I tried a somewhat sensitive prompt containing the word "girl" in it. Llama only ever generated refusals for this prompt, citing I was prompting for CSAM. GPTs and Claudes of that era had no issues with the same prompt.

Re: H3-metal – Native MiniMax-H3 inference for Apple Silicon

#104

Post by mrbonner » Wed, Aug 12, 2026, 3:07 PM UTC

robotresearcher wrote: post #25

abhinai wrote: post #4

How similar are Jeff Dean and Salvatore Sanfilippo?

People are really good at stuff. I noticed on a bar TV the other day that some of the Chromecast screensaver landscape photo credits were to Peter Norvig. They were really lovely pictures.

I have had 2 encounters with Dr Norvig, all online. One, I randomly reach out for him to get a referral to a hiring manager at Google. He gave me a couple of email leads. Two, I emailed ask him about a writing piece regarding fuzzy logic and why it was not popular in the US academic landscape. I got a very long, specific and detail response back. I need to print that email out and frame it lol.

Re: H3-metal – Native MiniMax-H3 inference for Apple Silicon

#105

Post by mrbonner » Wed, Aug 12, 2026, 3:22 PM UTC

iamyatin wrote: post #59

Noob question to all, is there any open source coding model that I can run on Mac mini 16gb?

I use Bonsai 27B ternary on my 24GB MBP. But, I believe you definitely can run it with 18 or even 8GB.

Re: H3-metal – Native MiniMax-H3 inference for Apple Silicon

#106

Post by ChristmasTomer » Wed, Aug 12, 2026, 4:38 PM UTC

I mean yeah, a 5090 is obviously gonna smoke a Mac lol. But that’s not really the point. If I can run this thing locally and get a video in a few minutes instead of an hour, that’s already pretty sick.

Re: H3-metal – Native MiniMax-H3 inference for Apple Silicon

#107

Post by benbojangles » Fri, Aug 14, 2026, 5:03 AM UTC

running this on an m1 max 64gb, it produces some really nice clips with music and audio, stitching the clips together after makes a nice short story generated entirely using a local ai system. It takes quite a while to generate each clip though, But at least it is working. I would like to know a few things:

-Optimal settings/configs examples for h3.c to help speed things up -Optimal recommended generation settings for each mac product, i am sure it is easy to do -Prompt generator assistant

Great work from the github author. The more i use it the more i realise i don't need a gui to generate video, just terminal.

My setup: Macbook #1 as a client Macbook #2 as server

-use macbook #1 terminal + ssh -run mactop in terminal tab to monitor Macbook #2's hardware during ai video generation -run h3.c on Macbook #2 via ssh session -transfer the output video file from macbook #2 to macbook #1 using terminal scp -view the video

Re: H3-metal – Native MiniMax-H3 inference for Apple Silicon

#108

Post by throw1234567891 » Mon, Aug 17, 2026, 1:42 PM UTC

embedding-shape wrote: post #90

throw1234567891 wrote: post #84

Earlier quoted context omitted.

An RTX6000 is a completely different class of hardware.

Really? No wonder I keep trying to type on it like a laptop but it doesn't work and doesn't even have a display!

Right? Do you also type on a Mac Studio without a keyboard plugged in? Like tap the ethernet port 3 times in a row then this sequence of sticking your fingers into TB5 ports? I mean, it’s clear you stick your RTX into a computer.