Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

GLM-5 was trained entirely on Huawei chips

glm5.net

Post reply on HN
Print view View original

1–10 of 17 posts

Re: GLM-5 was trained entirely on Huawei chips

#2

Post by readitalready »

Welp. NVidia had a good run while it lasted. RIP.

BTW so far on my GLM-5 evals it's performing qualitatively as well as Opus 4.5/4.6. Only issue is maybe speed. I will likely incorporate it into daily use. The previous versions were trash, filled with syntax errors and instruction following mistakes.

Re: GLM-5 was trained entirely on Huawei chips

#3

Post by chasing0entropy »

readitalready wrote: post #2

Welp. NVidia had a good run while it lasted. RIP. BTW so far on my GLM-5 evals it's performing qualitatively as well as Opus 4.5/4.6. Only issue is maybe speed. I will likely incorporate it into daily use. The previous versions were trash, filled with syntax errors and instruction following mistakes.

I wouldn't say nvda is completely out,but the chess moves in response is tough - Release a new chip, obsoleting the existing lines and take hits on billions of defaults on hardware

Re: GLM-5 was trained entirely on Huawei chips

#4

Post by keyle »

chasing0entropy wrote: post #3
readitalready wrote: post #2

Welp. NVidia had a good run while it lasted. RIP. BTW so far on my GLM-5 evals it's performing qualitatively as well as Opus 4.5/4.6. Only issue is maybe speed. I will likely incorporate it into daily use. The previous versions were trash, filled with syntax errors and instruction following mistakes.

I wouldn't say nvda is completely out,but the chess moves in response is tough - Release a new chip, obsoleting the existing lines and take hits on billions of defaults on hardware

Let's say nvidia has been de-moated.

Re: GLM-5 was trained entirely on Huawei chips

#6

Post by stingraycharles »

mindwok wrote: post #5

This doesn’t appear to be an official website. The official release only mentions Huawei is supported for inference. So… pretty sure this is not true.

Which would make a whole lot more sense as a stepping stone. I believe Anthropic also only used Google TPUs for inference until the last generation.

Re: GLM-5 was trained entirely on Huawei chips

#10

Post by mindwok »

stingraycharles wrote: post #6
mindwok wrote: post #5

This doesn’t appear to be an official website. The official release only mentions Huawei is supported for inference. So… pretty sure this is not true.

Which would make a whole lot more sense as a stepping stone. I believe Anthropic also only used Google TPUs for inference until the last generation.

Inferencing with Huawei chips is not new either, they’ve had DeepSeek running on them since last year.
Post reply on HN