Alexis Gallagher algal

Serving Nemotron-3-Super-120B-A12B-NVFP4 on a single DGX Spark

Tested 2026-03-27. Uses sparkrun with a community-patched vLLM container that works around the MIXED_PRECISION whitelist bug in the NGC vLLM container (vllm-project/vllm#37854).

Prerequisites

DGX Spark with DGX OS (Ubuntu 24.04, aarch64)
Docker installed (ships with DGX OS)
Your user in the docker group
A HuggingFace account and access token

Notebook Formats & Conversion Tools

This document summarizes some popular, public notebook formats (Org, Jupyter, Markdown, Python variants) and the tools which interconvert them.

Notable omissions:

Wolfram notebooks. My impression is the proprietary platform offer mature export to many formats but not ipynb. There is no mature, widely used too for export to ipynb, not are there great import tools afaict.
SolveIt platform. Not publicly released yet. But I suspect it will support excellent ipynb interop. 😉

Epistemic status: Moderate confidence. This is based mostly on AI research on 2025-08-30 but it passes spot checks and matches my prior background knowledge, which covers many of these technologies.)

Help me understand the Student's t-distribution.

I've studied it but I don't remember it.

Of course, I do want to understand the procedure to use when trying to estimate the variance of a population from a single sample. I understand that this is its main use.

But I mainly want to have a terse, but 100% conceptually crisp, understanding of the key concepts necessary to understand why this procedure is what it is. FOr me to remember things, it is important that they be very coherent. So I care a lot about precision in terminology, e.g., to distinguish what is treated as a random variable, what is a parameter of a distribution, over what range of values an expectation is calculated, etc..

🤖🤖🤖

How do I find out about the e2scrub_all.timer on my system?

To learn more about the e2scrub_all.timer on your system, you can use these systemctl commands:

View timer details:

systemctl status e2scrub_all.timer

	#!/usr/bin/env python3
	# /// script
	# requires-python = ">=3.9"
	# ///
	"""Checkpoint: save, then squash consecutive solveit commits into one with given name."""

	import subprocess, sys
	from datetime import datetime, UTC

	def git(*args, check=True):

	import AVFoundation
	import Foundation
	import os

	// Microphone permission (supports macOS, fallback)
	func requestMicrophonePermission() async -> Bool {
	await withCheckedContinuation { continuation in
	AVCaptureDevice.requestAccess(for: .audio) { granted in
	continuation.resume(returning: granted)
	}

	;; This is miraculously all you need to add decent support for python.
	;; It requires that you have uv installed. Then run `eglot` in a python buffer.
	(use-package eglot
	:defer t
	:config
	(add-to-list 'eglot-server-programs
	`(python-mode . ,(split-string "uvx --quiet --from pyright pyright-langserver --stdio"))))

	;; this is handy for dumping web pages into buffers, to then add them to context for AI with gptel
	(defun alg/insert-markdown-from-url (url)