Soul Player C64 – A real transformer running on a 1 MHz Commodore 64
A transformer model with approximately 25,000 parameters is implemented on an unmodified Commodore 64, utilizing hand-written 6502/6510 assembly. This 2-layer decoder-only architecture features real multi-head causal self-attention, softmax, and RMSNorm, and can be loaded from a floppy disk.
github.com
5 min
4/21/2026
We Are the Last People Who Know How It Works
In the 1990s, playing a computer game often required understanding the underlying computer system, including editing files like autoexec.bat. Users sometimes created boot disks specifically configured to launch a single game.
unix.foo
4 min
6/30/2026
Soul Player C64 – A real transformer running on a 1 MHz Commodore 64
A transformer model with approximately 25,000 parameters is implemented on an unmodified Commodore 64, utilizing hand-written 6502/6510 assembly. This 2-layer decoder-only architecture features real multi-head causal self-attention, softmax, and RMSNorm, and can be loaded from a floppy disk.
github.com
5 min
4/21/2026
Soul Player C64 – A real transformer running on a 1 MHz Commodore 64
A transformer model with approximately 25,000 parameters is implemented on an unmodified Commodore 64, utilizing hand-written 6502/6510 assembly. This 2-layer decoder-only architecture features real multi-head causal self-attention, softmax, and RMSNorm, and can be loaded from a floppy disk.
github.com
5 min
4/21/2026
No more articles to load