Skip to content

Instantly share code, notes, and snippets.

Avatar
🏠
Working from home

Benjamin Bolte codekansas

🏠
Working from home
View GitHub Profile
@codekansas
codekansas / benchmark_self_attention.py
Last active March 11, 2023 18:34
Benchmarking script for attention
View benchmark_self_attention.py
import argparse
import contextlib
import logging
import math
import random
import time
from dataclasses import dataclass
from pathlib import Path
from typing import Callable
@codekansas
codekansas / camera_mount.scad
Last active February 9, 2023 00:06
Random camera mount part
View camera_mount.scad
// Mount for my ESP 32 above my workbench.
cam_diam = 7.7 + 1.0;
cam_offset = 2.1;
other_offset = 1.9;
rect_height = 39.6 + 1.0;
rect_width = 27.0 + 4.0;
cam_from_top = 10.4;
leng = 55;
@codekansas
codekansas / parse_webtime.py
Created December 3, 2020 20:50
Script to parse the Webtime Tracker CSV file and plot it
View parse_webtime.py
#!/usr/bin/env python
import argparse
from datetime import datetime
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
View bookend.scad
// Total dimensions.
height = 120;
width = 40;
padding = 10;
module col(h, sph=false) {
difference() {
cube([width, width, h]);
if (sph)
translate([width / 2, width / 2, h])
View combination_lock_riddle.scad
/*
A combination for a lock has 3 wheels, X, Y, and Z, each of which
can be set to eight different positions. The lock is broken and when
any two wheels of the lock are in the correct position, the lock
will open. Thus, anyone can open the lock after 64 tries (let A and
B run through all possible permutations). However, the safe can be
opened in fewer tries! What is the minimum number of tries that can
be guaranteed to open the lock?
*/
View two_students_riddle.py
#!/usr/bin/env python3
"""Problem statement:
There is a teacher and 2 students in a classroom. The students are A and B.
The teacher thinks of 2 positive integers and tells the sum of those numbers
to student A without student B hearing it. Then tells their product to student
B without student A hearing it. After this, the teacher asks the 2 students
what was the 2 numbers.
First student A says: I don't know.
View dishwasher-chopstick-holder.scad
thickness = 3;
padding = 1;
slat_size = 3;
short_length = 46.36;
long_length = 61.76;
height = 51.76;
first_indent = 15.92;
second_indent = 29.97;
View laptop-stand.scad
// Parametric OpenSCAD design for a laptop stand.
// Laptop dimensions.
lwid = 304.1;
lhei = 212.4;
ldep = 15;
// Stand long dimensions.
slen = 200;
sswid = 60;
@codekansas
codekansas / bert_pytorch.py
Created November 1, 2018 10:25
Implementation of the transformer block used by BERT
View bert_pytorch.py
#!/usr/bin/env python3
"""Implementation of the transformer block used by BERT.
I saw an excellent implementation of the complete BERT model here:
https://github.com/codertimo/BERT-pytorch
I re-wrote a simplified version of the transformer block below. This was mainly
for my own understanding (so that I could get a grasp of the dimensions and
how the whole attention mechanism works), but I tried to document it pretty
thoroughly so that other people can understand it without having to go too far
@codekansas
codekansas / binarized_nn_inference.cpp
Created November 1, 2017 02:25
Efficient binarized neural network inference
View binarized_nn_inference.cpp
/* Binarized neural network inference example.
This shows a simple C++ program for doing inference on
binarized neural networks. To do this efficiently, the code
below makes use of the "bitset" class, which uses the "popcnt"
instruction to count the number of 1's that show up in the
matrix product, in constant time. This means that a matrix
multiplication between a (A, B) and (B, C) matrix takes
O(A * C) time; in other words, each value in the output matrix
is computed in constant time.
*/