Skip to content

Instantly share code, notes, and snippets.

View bquast's full-sized avatar
🌴
On vacation, but I love this stuff, so still working 😄

Bastiaan Quast bquast

🌴
On vacation, but I love this stuff, so still working 😄
225 followers · 85 following
View GitHub Profile
@bquast
bquast / qwen3-1-7b-sft-simplified-technical-english.ipynb
Created August 10, 2026 18:45
qwen3-1.7b-SFT-Simplified-Technical-English.ipynb
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
@bquast
bquast / qwen3-1-7b.ipynb
Created August 10, 2026 17:45
qwen3-1.7b.ipynb
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
@bquast
bquast / sft-smollm2.ipynb
Created August 10, 2026 00:54
SFT-SmolLM2.ipynb
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
@bquast
bquast / self-learning-adaptive-grapling-system.py
Created May 21, 2026 13:55
import time
import numpy as np
from lerobot.common.robot_devices.robots.so101 import SO101Follower
from lerobot.common.robot_devices.cameras import get_cameras
# initialize robot and cameras
robot = SO101Follower(port="/dev/ttyACM0", id="my_so101") # Adjust port
robot.connect()
@bquast
bquast / SmolLM-135M-quantize-int8-mlx.py
Created May 18, 2026 09:38
MLX script to quantize SmolLM-135M from bf16 to int8
import mlx.core as mx
from mlx.utils import tree_flatten
from mlx_lm import load
model, _ = load("HuggingFaceTB/SmolLM2-135M")
def absmax_int8(w):
scale = mx.max(mx.abs(w)) / 127.0
w_q = mx.round(w / scale).astype(mx.int8)
w_hat = w_q.astype(w.dtype) * scale
@bquast
bquast / big-files-list.sh
Last active May 12, 2026 18:51
Saving disk space: below script lists the large files and folders, to be copy-pasted into an LLM, for space-saving suggestions
#!/bin/zsh
# cleanup.sh
# Usage: ./cleanup.sh [directory]
# Default to HOME if no directory is provided
TARGET="${1:-$HOME}"
echo "### macOS SSD Analysis for: $TARGET"
echo "### Date: $(date)"
echo "---"
@bquast
bquast / llama.go
Last active May 10, 2026 20:09
llama inference engine, written in Go, builds to wasm, downloads SmolLM2-135M and runs inference in the browser
package main
import (
"encoding/binary"
"fmt"
"math"
"math/rand"
"regexp"
"strings"
"syscall/js"
@bquast
bquast / rsa_homomorphic_demo.html
Last active April 29, 2026 15:37
LIVE: https://bquast.github.io/RSA_homomorphic_demo/
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>RSA — Multiplicative Homomorphism</title>
<link href="https://fonts.googleapis.com/css2?family=Archivo+Black&family=Work+Sans:wght@400;600&family=Space+Mono&display=swap" rel="stylesheet">
<style>
*, *::before, *::after { box-sizing: border-box; margin: 0; padding: 0; }
@bquast
bquast / claude-export-markdown.js
Created April 12, 2026 10:33
Export Claude conversation to markdown
(() => {
const title = document.title.replace(' - Claude', '').trim();
const lines = [`# ${title}\n`];
// Grab every turn in the conversation
const turns = document.querySelectorAll('[data-testid="user-message"], .font-claude-response-body, .standard-markdown');
// Better approach: walk the full message list
// User messages
const userMsgs = [...document.querySelectorAll('[data-testid="user-message"]')];
@bquast
bquast / calmplot.js
Last active April 11, 2026 12:28
calm animation for blog charts live, example in the comments
(function () {
function parseDuration(value) {
var text = String(value || "").trim();
if (text.endsWith("ms")) {
return parseFloat(text.slice(0, -2));
}
if (text.endsWith("s")) {
return parseFloat(text.slice(0, -1)) * 1000;