NanoEuler
AI
language model training from scratch
SHOW HN2026
How HN took it
3 positive, 5 neutral or asking, 0 mixed, 4 critical, out of 12 comments about the project. Thread score -0.04.
Docs: 1 for, 1 againstExecution: 0 for, 1 againstPerformance: 0 for, 1 againstDesign: 0 for, 1 againstIdea: 1 for, 0 against
The thread, comment by comment
Each row is a top-level comment and its replies, in HN's order. Select a cell to open that comment on HN.
positive neutral or mixed critical about another commenter or something else the maker
What HN said
What people liked
Interesting and educational project
From 3 comments
this is super interesting. Looking forward to trying this out!
ali_chherawalla on HNAmazing, i’ll try to learn from your code. Thanks!
memezed on HNWow that's really cool i'll definitely check it out! have played around with machine learning algorithms built from scratch in c / cuda too, but once i hit the cuda part of it i…
AndReics on HN
Objections
Training is slower than PyTorch, about 1.5-2.5x
From 1 comment
Hi, in nanoeuler I use cuBLAS (NVIDIA's super-optimized library) for all matrix multiplications, with the tensor cores in TF32 mode. It's the same thing PyTorch uses underneath,…
vforno on HN
Code style is odd and README appears AI-generated
From 2 comments
Very weird coding style, did you run astyle --style=python on C code? Also, your LLM left a comment in the cuda source that it is untested, does the cuda stuff work?
Chu4eeno on HNMentioning neural ODE doesn't make sense here, as this is unrelated. Basically any implementation of transformer uses residuals, but you're not really training a neural ODE here.…
tdesilva on HN
Training data from Gutenberg and Shakespeare is a poor corpus for modern English
From 1 comment
Not sure, but the code is quite dense and lacking in comments. `nanoeuler` & `nanoeuler_check` is itself the binary checked straight into git with the `.log` file? All of the…
bArray on HN
Commit history is sparse and uploads are rushed
From 1 comment
Not sure, but the code is quite dense and lacking in comments. `nanoeuler` & `nanoeuler_check` is itself the binary checked straight into git with the `.log` file? All of the…
bArray on HN
Requests
Disclose LLM usage in the project
From 1 comment
Consider adding a rule that an author must disclose (in their own words) for what parts and to what extent LLMs have been used to assist their project.
gaflo on HN
Add more comments and clearer README
From 2 comments
Not sure, but the code is quite dense and lacking in comments. `nanoeuler` & `nanoeuler_check` is itself the binary checked straight into git with the `.log` file? All of the…
bArray on HNMentioning neural ODE doesn't make sense here, as this is unrelated. Basically any implementation of transformer uses residuals, but you're not really training a neural ODE here.…
tdesilva on HN
Use a modern English corpus such as Kiwix ZIM archives
From 1 comment
Not sure, but the code is quite dense and lacking in comments. `nanoeuler` & `nanoeuler_check` is itself the binary checked straight into git with the `.log` file? All of the…
bArray on HN
Questions
How much of the code is LLM generated
From 1 comment
I'm genuinely curious how much of this is LLM generated?
isatty on HN
Does the CUDA code work, since a comment says it is untested
From 1 comment
Very weird coding style, did you run astyle --style=python on C code? Also, your LLM left a comment in the cuda source that it is untested, does the cuda stuff work?
Chu4eeno on HN
How long was training and how many tokens
From 1 comment
How long was it trained for? How many tokens?
ericb on HN
Is training faster or slower than PyTorch
From 1 comment
Wow that's really cool i'll definitely check it out! have played around with machine learning algorithms built from scratch in c / cuda too, but once i hit the cuda part of it i…
AndReics on HN
Alternatives named
PyTorch with torch.compile
From 2 comments
Hi, in nanoeuler I use cuBLAS (NVIDIA's super-optimized library) for all matrix multiplications, with the tensor cores in TF32 mode. It's the same thing PyTorch uses underneath,…
vforno on HNYes, because it has many separate kernels instead of aggressive merges like PyTorch (with Torch Compile). Each pass (norm, matmul, residual, RoPE, etc.) launches its own kernel,…
vforno on HN
nanoGPT-style minimal training code
From 1 comment
Wow that's really cool i'll definitely check it out! have played around with machine learning algorithms built from scratch in c / cuda too, but once i hit the cuda part of it i…
AndReics on HN