interesting articles on gpt

a collection of interesting articles i have been reading on gpt
intro i've been reading a lot of articles on gpt, and i've been interested in the abilities, limitations, and future of gpt and similar large language models.
more (100 words) →

comparsion of gpt models

a table of evaluation metrics for gpt models
updated 2022-02-06: added entry for gpt-neox-20B intro i put together a table of various metrics for all the main gpt models.
more (200 words) →

micro-finetuning gpt2

notes on finetuning gpt2 with a tiny dataset
introduction disclaimer i am primarily a software engineer, and thus am quite an amateur at machine learning.
more (1900 words) →

irre-tools devlog

working project notes collection
Goals TODO completely rewrite compiler to replace clang-leg TODO test device driver functionality Plans Compiler rewrite (clang-leg to vbcc) Problems with clang-leg (why rewrite?
more (500 words) →

duckmind ai engine, part 1

overview and core concepts of ducia
overview duckmind is a collection of projects that are based on Sor's AI engine, and provides its game-playing AI functionality as a reusable library.
more (900 words) →

keras on amd with plaidml

setup for amd gpu acceleration for keras
intro this week i wanted to revisit one of my old machine learning projects. that project used the excellent keras library for building the model.
more (300 words) →

sōr devlog

original devlog notes for sor
Progress Graphical juice An interesting way to make the graphics look cooler that I discovered accidentally; mixing the color channels (for example, swapping red and blue) will give it a cool effect.
more (3800 words) → on linux

using in rider on linux
nowadays, the latest pipelines can be found on the azure dev page: in particular, the oss package. from there, go to linux artifacts (example).
more (200 words) →

better native monogame

using .net core tools to build smaller, better native bundles
intro since .NET Core was released, its native publishing options have improved significantly. it is now possible to build a MonoGame application and build it into a fully native, self-contained release that is less (sometimes significantly less!
more (500 words) →

remote borg access

accessing a borg repo stored on an rclone remote
i have been experimenting with accessing a remotely stored borg repo. since these repos can get fairly large, i would like to be able to fully access backup data using only an rclone remote storage.
more (500 words) →