rs-gpt
- 7 Devlogs
- 8 Total hours
A Rust implementation of Andrej Karpathy's microgpt
Ship #1
Pending review
I rewrote Andrej Karpathy’s microgpt in rust. Also, added a beautiful tui. You classic mode and tui mode. The classic mode behaves exactly as the original microgpt behaves, but the tui mode is a classic mode wrapper with a tui. You can use it now by installing it from crate.io or building it from the source using github.
30m logged
DevLog#7 — rs-gpt is finally published on github and crate.io
2h 37m 26s logged
DevLog#6 — I added a tui that really makes the training and inference visually impressive. Also I published the project.
31m 12s logged
DevLog#5 — Implemented the inference, ran the model and it works!!!!!!! Now, I can say that I have finished rewriting Andrej Karpathy’s microgpt in rust. I will add more things to it, maybe a tui and make the code more modular.
1h 6m 44s logged
DevLog#4 — Implemented the training loop, the model parameters, and the optimizer. Also, removed the dataset() function and just added the logic in main(). And what is left? Is the inference, and by finishing the inference I will be finished with the rust rewrite.
1h 11m 55s logged
DevLog#3 — Implemented the gpt-2 like architecture, actually it was tense, the rust rewrite for it made my mind want explode, but yeah I rewrote it and now it’s here. the training and inference are the next goal.
1h 6m 14s logged
DevLog#2 — Implemented the Autograd for rs-gpt, it took a while trying to find a way to implement the original python code to rust, but once I figured out (inspired from stochastical/microgpt-rs implementation) everything became easy.
31m 24s logged
DevLog#1 — Implemented the dataset function. Currently, it reads the input.txt and randomize it into a list, I didn’t check for the file existence, but I will fix that in the future.
