scaled dot-product + multi-head attention implemented in raw numpy, no PyTorch. Runs a toy sentence through it, prints an ASCII attention heatmap, and saves a real PNG visualization.
Python Updated 4 months ago 1 commits 1 stars
prism-dev|Mirror: slymp3/attention-from-scratch (⭐1)
4 months agoE2B SandboxFirecracker microVMpython
No executable code files found in this repository.
About
scaled dot-product + multi-head attention implemented in raw numpy, no PyTorch. Runs a toy sentence through it, prints an ASCII attention heatmap, and saves a real PNG visualization.
Readme
1 stars