Loading...
Loading...
So, this week has been pretty busy for me in terms of work, and it brings no joy in admitting this but I had to do plenty of overtime work because of the current status of my team's project. I was handed out the biggest task I have ever been since I started working here, and it's been quite exciting but -- given the scale -- also quite challenging.
One takeaway from this: I love programming, but I despise writing tests.
For my weekly learning, I did try what I could, but personally, I think this has been an underwhelming one. Here's what I can all think of:
I've been trying to code a simple transformer using pure Tinygrad. I have gotten into this habit lately of watching George Hotz streams and the way he programs, and I've kinda wanted to contribute to Tinygrad for a bit, so thought this would be a good way to get used to the library. Read some docs, watched some streams. Haven't completed writing the transformer yet (current progress), but should be done soon. Frankly, think I've been a bit lazy. Not good enough.
Solved 2 Deep-ML problems. Been very fun, but I was supposed to do one each day... not good enough.
Studied KV Cache via the latest KV Cache pull request I saw in the nanoVLM repository. They're doing pretty good out there. Their KV Cache blog is pretty nice.
Coded a tic-tac-toe game in C++. This... does not sound very impressive but well, it did take me a little bit to craft the storing & pattern matching logic in this language. Was super fun.
... and that's pretty much it.
I saw that Sebastian Raschka dropped a new Substack post about coding KV Cache from scratch, so I'll do that this upcoming week. And among other things. See you next week!