CODE

NanoChat Live Tokenization During Training Optimization

By

–

22 October 2025 17h11

atm nanochat does not pretokenize, instead tokenization is done live during training in the dataloader, which is ok and gives the CPU something to do while the GPUs are working. So I didn't try too hard to optimize this part.

→ View original post on X — @sumanth_077

Reflex Build: Open-Source Framework for Web Applications

By

@sumanth_077

–

22 October 2025 17h07

Checkout Reflex: https://
github.com/reflex-dev/ref
lex
… Try out Reflex Build here:

→ View original post on X — @sumanth_077

Reflex Build: Create LLM Apps in Python with Prompts

By

@sumanth_077

–

22 October 2025 17h07

Build LLM apps in Python with just a prompt! Reflex is a Python library for building full-stack web apps in pure Python. With the new Reflex Build, you can create internal tools and dashboards using AI with just a single prompt. Here’s why this is a game changer: Reflex Build

Regex bottleneck in tokenizer requires expert optimization

By

–

22 October 2025 16h57

Yep, the use of regex is both a huge dependency and huge bottleneck in the tokenizer. I think it's a beautiful project to try to do this correctly, but I'd need someone who is really familiar with regex to pitch in and also a large test suite to make sure. I'd love to merge such

→ View original post on X — @mathworks

MATLAB Copilot: AI Assistant Helps Overcome Development Roadblocks

By

@mathworks

–

22 October 2025 12h00

Hitting a roadblock? MATLAB Copilot has your back

→ View original post on X — @soumithchintala

Major Code Pushes Coming for Forge and Monarch Projects

By

@soumithchintala

–

22 October 2025 3h44

more on forge and monarch tomorrow. watch out for giant new code pushes

Learning Error Recovery Without Creating Errors in Testing

By

–

22 October 2025 0h38

That’s almost separate… You certainly want to learn recoveries in the event of errors at test time. You just don’t want to also learn to actively create errors.

GPT-10 Input Architecture: Pixels Over Tokenization

By

–

22 October 2025 0h33

Haha. I am afraid people interpreted my “delete tokenizer” as “use bytes directly without BPE”, the issue is you *still* need bytes encoding arbitrariness even for that! Pixels is the only way. Just like humans. It is written. If GPT-10 uses utf8 at the input I will eat a shoe.