Michael Moser MoserMichael

llama.cpp - reading the source code

looking through the llama.cpp source code, to learn about language models. Disclaimer: it's january 2026, so things will likely change, at some point.

llama.cpp repo link
build instructions
- contains instructions on how to build the debug version
- default build creates devices: 'libggml-base.so' and 'libggml-cpu.so' - probably need to set up additional requirements for the other backends.
- if you need additional backends: documentation on each backend includes build instructions
looking at main.cpp for llama-simple - a CLI program to continue with a given prompt, specified on the command line. command line: ./llama-simple -m model.gguf [-n n_predict] [-ngl n_gpu_layers] {prompt-text}

not very politically correct gist

I am taking the json from the following gist original rsdb gitst - local copy It contains a json with a set of records on racial and ethnical slurs.

Wrote a python script that does the following things

create a json where all slurs are grouped by the name of the group, which is the subject of the slur slurs sorted by group name

learning queuing theory basics

Bikey Bonn Kleiford Seranilla - Queuing Theory

Liz Thompson - Queueing theory (simple)

SREcon24 Americas - System Performance and Queuing Theory - Concepts and Application

Learning about the concept

Wikipedia on Meditation

The objectives of mediation

Reach a "mentally clear and emotionally calm and stable state".
other objectives:
- "improve focus by decreasing self-referential chatter"
- minimize "mind-wandering (aka spontaneous drift, shift of attention to 'something personal')
minimize rumination (repetitive thought pattern where the mind becomes "stuck" on negative content) and

	# -------------------------
	# Bash stuff
	# -------------------------

	Never understood arrays in bash (even though they were added in bash2.0 - 2009)
	This explains them:

	(can you compare bash to perl5?
	No, when declaring a data structure in bash, then you cannot nest a map inside an array or map.
	Also runs slower - bash interpreted line by line, instead of bytecode translation/execution of perl5

	# -------------------------
	# Windows stuff
	# -------------------------

	You may find yourself to wonder, why the mouse cursor disappeared on what seems to be your laptop, while you do see a mouse cursor in the windows login screen, prior to login.

	after login:

	press Windows and I key, simultaneously, to jump to the 'Windows Settings' dialogue

	# --------------------------------------------
	# video / audio manipulation with command line
	# --------------------------------------------

	# downloading youtube videos via command line...

	# yt-dlp (best to use it via virtual env)

	# installing yt-dl
	# - (never do apt-get install - this doesn't work here)

	Alternatives:

	https://search.brave.com/ - they maintain their own crawlers & index. It is surprisingly good
	(some people say that regular (non ai mode) google search is getting worse and worse)

	---

	Link to google in AI mode, with the option to chat with a LLM.
	Probably uses Gemini of various versions, the version can probably vary - according to task.

	# -------------------------
	# tcpdump packet capture
	# -------------------------

	# capture tcp to and from port 9393 (-n tells not to translate port numbers)
	# just logs a message for each packet with tcp flags, sequence number, without the data

	# Note! for SYN / SYN-ACK the real sequence number/ack number is shown, all other packets get relative sequence number
	# (relative to initial sequence number)

	# use brew on the mac, this installs ollama that has been compiled with the metal library
	# metal is a library that uses the native GPU on the M3 processor
	brew install ollama

	# check that metal is in use
	# if metal is installed correctly, you will see a line like this on the screen:
	# "inference compute" ... library=Metal ...
	OLLAMA_DEBUG=1 ollama serve

	# next thing we want to do: load the claude model