GLM 4.5 air running on 4x 3090 with 28 tokens/sec Pipeline parallelism over PCIe(no nvlink)

People still haven’t caught up on GLM-4.5V > Same size as GLM-4.5-Air > Can identify instructions in text in an image and respond with correct code > Can identify a multi GPU rig even guessing 4090s (close enough) and AI usecases as the reason > Can identify a specific type

Having been around most corners of the world I can confidently say that Japan is the best country. It’s clean, organised, advanced, comfortable, and beautiful. Its people are kind, respectful, disciplined, intelligent and hardworking. It has given us some of the best products,

> Be Karl Bushby > Decide life is not difficult enough > “I will walk home, back to the UK” > Goes to the southern edge of Chile > Packs his bags, sets off in 1998 > No transport, his only rule > Should be home in 10 years > 2 years later he’s already walked through the entire

I really love GLM, having something so accurate run at home is wild. I’m working on a project to get a vision enable cat door to let only my cats into the house now that it’s getting cold. This might work, I wonder

They’re putting up GPUs in space, least you can do is start using the hell out of yours. There’s a reason this is all happening so fast.

I asked the same question to GLM-4.5V (self hosted) and GPT-5-High This is a question that’s been bumping around in my skull for years. When I was a child, I was at a park somewhere in New York with my baby sitter, and I remember as soon as the sun started setting hundreds of

Python is a shit language, fucking environments garbage

You should never be satisfied with where you are.

Unless you're a world class dev like Linus, or a Karpathy don't lecture me about how AI coding is worse than human coding. How it sucks, how it's not really that good, how you can't trust it, etc.. Most people suck, most people are bad at their jobs. They can be lazy, they

When you're against me this is who you have to go through

Saturn devouring his son. - Goya

If you ever feel satisfied with where you left something off you should reconsider. If you are not obsessed with refining and improving your work, your self, your family, your financial status, and your mind you are turning your back against god.

I ran all the most important benchmarks against: - GLM-4.5-Air-reap-82b at Q8 - GLM-4.5-Air Q4_K_S Reap did horribly, like sub-8B param horrible.. vs the q4 which did decent enough.. It's crazy