TE
TechEcho
Home24h TopNewestBestAskShowJobs
GitHubTwitter
Home

TechEcho

A tech news platform built with Next.js, providing global tech news and discussions.

GitHubTwitter

Home

HomeNewestBestAskShowJobs

Resources

HackerNews APIOriginal HackerNewsNext.js

© 2025 TechEcho. All rights reserved.

Show HN: less than 650 LOC trainable GPT only using NumPy

90 pointsby joennlaeover 1 year ago

3 comments

cuuupidover 1 year ago
I think people are forgetting that transformer architectures are a wider field from GPT and predate GPT3 by 3+ years. Referring to transformer architectures using a branded commercial nomer (GPT) is just going to help cement OpenAI’s brand exposure and soon regulatory capture.<p>For comparison this would be like referring to convonets as Inception architectures back during the CV boom (or VGGnets before that)
评论 #38307668 未加载
评论 #38314428 未加载
评论 #38307907 未加载
评论 #38307565 未加载
评论 #38307736 未加载
评论 #38312050 未加载
评论 #38312030 未加载
gfaureover 1 year ago
Nice! The README mentions `LayerNorm` is implemented here, but while it&#x27;s in the equivalence tests with PyTorch, I don&#x27;t see it in the implementation.
评论 #38309767 未加载
p1eskover 1 year ago
I wonder how easy it would be to port this library from numpy to cupy.
评论 #38312145 未加载
评论 #38310283 未加载