Reference Sheet

Python

Python Environments, Packages, Modules, Cache

# Conda environment management
conda create -n transcriptions python=3.10.20
conda activate transcriptions
conda deactivate
conda remove --name transcriptions --all          # --all removes all packages inside that environment

# venv virtual environment management (windows)
python -m venv venv               # create
venv\Scripts\activate             # activate
deactivate                        # deactivate
rmdir /s /q <venv_folder_name>    # delete

# venv virtual environment management (posix)
sudo apt install python3-venv     #! extra install may be required for venv on Linux/MacOS
python3 -m venv venv              # create
source venv/bin/activate          # activate
source deactivate                 # deactivate
rm -rf venv/                      # delete

# PIP: Install regular and specific package versions with PIP
pip install numpy openai websockets
pip install torch==2.8.0 torchvision==0.23.0 torchaudio==2.8.0 --index-url https://download.pytorch.org/whl/cu128
pip install -r requirements.txt

# MISC
python -m module                      # runs a module as a script
python -m pip install numpy           # -m to guarantee running pip tied to exact python interpreter
python -m pip install --upgrade pip   # upgrade pip itself
pip list --outdated --format=columns  # list upgradable packages
pip install --upgrade package_name    # upgrade specific package

pip cache dir                         # where cache is
pip cache info                        # size, file count, location
pip cache list                        # list names of cached packages
pip cache remove package_name         # remove specific package from cache
pip cache purge                       # clear entire cache

Python Imports: modules, absolute, relative

# Assumed structure
project/
├── main.py
├── utils/
│   ├── __init__.py
│   ├── math_utils.py
│   └── io_utils.py
└── models/
    ├── __init__.py
    └── user.py

# import entire module + objects their object refs
import utils.math_utils
utils.math_utils.add(1, 2)

# import specific objects (func/class)
from utils.math_utils import add, subtract
add(1, 2)

# import and rename
from utils.math_utils import add as add_numbers
add_numbers(1, 2)

# import everying
from utils.math_utils import *

# absolute imports
from utils.math_utils import add
from models.user import User

# relative imports inside packages (!not in top-level script)
from .math_utils import add         # from same directory
from ..utils.math_utils import add  # from parent directory
from ..utils import io_utils        # from sibling via parent
from ...core.config import settings # from two levels up
python -m project.main              #! relative imports require running as module

Python paths

# easy mode (any OS)
from pathlib import Path
p = Path("dir") / "subdir" / "file.txt"           # safely join path
p = Path(r"C:\Users\masly\file.txt")              # handle Windows path
p = Path(r"C:\Users\masly\file.txt").as_posix()   # normalize to forward slash

print(p.resolve())                                # absolute path
p.exists(); p.is_file(); p.is_dir()               # exist; type; type
p.name; p.parent; p.suffix                        # base name; dir name (full if resolved); extension

here = Path(__file__).parent                      # script location
home = Path.home()                                # ~/

import glob
csv_files = glob.glob("root/**/*.csv", recursive=True)  # find all CSV in subdirectories of root and get their full paths
csv_files = list(Path("root").rglob("*.csv"))           # same thing but Path objects
csv_files = list(Path("root").glob("**/*.csv"))         # same thing but Path objects

# Manual Windows Paths handling
windows_path = r"C:\Users\masly\projects\file.txt"
windows_path_fixed = windows_path.replace("\\", "/")
wsl_path = windows_path_fixed.replace("C:/", "/mnt/c")
windows_path_fixed = wsl_path.replace("/mnt/c/", "C:/")

Environment Variables

Linux/macOS (bash/zsh)

export API_KEY="sk-..."                                         # set in current session (shell/terminal)
echo $MY_VAR                                                    # read
API_KEY="sk-..." python main.py                                 # only set env var for this command
echo 'export API_KEY="sk-..."' >> ~/.bashrc && source ~/.bashrc # Linux persist in shell cfg
echo 'export API_KEY="sk-..."' >> ~/.zshrc && source ~/.zshrc   # macOS persist in shell cfg

Windows

# CMD
set API_KEY=sk-...      # set ! no double quotes
echo %API_KEY%          # read
setx API_KEY "sk-..."   # persist (user-level)

# PowerShell
$env:API_KEY="sk-..."     # set
$env:API_KEY              # read
[System.Environment]::SetEnvironmentVariable("API_KEY","sk-...","User") # persist (user-level)

Python

import os
import subprocess

if "API_KEY" not in os.environ: print("nu-uh")  # check existence
os.environ["API_KEY"] = "sk-..."                # set (current process)
value = os.environ.get("API_KEY")               # read

subprocess.run(["python", "script.py"]) # subprocess inherits!
# Copy and overwrite before subprocessing
env = os.environ.copy()
env["API_KEY"] = "sk-2..."
subprocess.run(["python", "script.py"], env=env)

Linux

Linux PIDs & Open Files

# List active processes: a = all users; u = user-formatted columns; x = include non-terminal processes (e.g., daemons)
ps aux 

# List python programs from current user running in terminals
ps -au | grep python
pgrep -a -u $USER python

# Kill specific process ID
kill 76531
kill -9 76531 # SIGKILL

# Kill all python apps running from this user
pkill -u $USER python

lsof -i :8081 # list open files on port 8081 -> what is serving

Apt package manager

sudo apt update # update sources (refresh package lists)
sudo apt upgrade # upgrade installed packages
sudo apt install nginx # install package
sudo apt install nginx=1.18.0-0ubuntu1 # install specific package version
sudo apt install ./package.deb # install manually downloaded package
sudo apt reinstall nginx # reinstall package

sudo apt full-upgrade # handle dependencies/removals
apt search nginx # search for specific packages
apt show nginx # package details
apt list --installed
apt list --upgradable
sudo apt remove nginx # keeps config files
sudo apt purge nginx # removes configs too
sudo apt autoremove # autoremove unused depenencides
sudo apt clean # remove all cached .deb files
sudo apt autoclean # remove only obsolete
sudo apt --fix-broken install
sudo nano /etc/apt/sources.list # edit sources
sudo nano /etc/apt/sources.list.d/myrepo.list # add new repo file (example entry: `deb http://archive.ubuntu.com/ubuntu jammy main universe`)
sudo add-apt-repository --remove ppa:deadsnakes/ppa # remove repository: specific one
sudo rm /etc/apt/sources.list.d/myrepo.list # remove repository: delete repository list file

# repo keys & MISC
sudo apt-key list
sudo mkdir -p /etc/apt/keyrings # update repo key
curl -fsSL https://example.com/key.gpg | sudo tee /etc/apt/keyrings/example.gpg > /dev/null # udpate repo key
deb [signed-by=/etc/apt/keyrings/example.gpg] https://example.com repo main # reference key in source
sudo do-release-upgrade # update sources

TODO: systemctl/systemd

Bash shell settings

# ~/.bashrc: executed by bash(1) for non-login shells.

# ...

# HISTORY SETTINGS
# Don't put duplicate lines or lines starting with space in the history.
HISTCONTROL=ignoreboth
# Append to the history file, don't overwrite it
shopt -s histappend
# History size
HISTSIZE=1000
HISTFILESIZE=2000
# Immediate history writes (not on exit)
PROMPT_COMMAND='history -a; history -n'

# PS1 PROMPT (PS1)
# Format: [Time] User@Host:Directory$ 
# Colors: \[\e[32m\] is Green, \[\e[34m\] is Blue, \[\e[m\] is Reset
PS1="[\[\e[33m\]\t\[\e[m\]] \[\e[32m\]\u@\h\[\e[m\]:\[\e[34m\]\w\[\e[m\]\$ "

Shell text editing

VIM

• [sudo] vi file.txt # open file.txt in vim
• i # insert mode (type)
• Esc # normal mode (command input)
• :w # save
• :wq or :x or ZZ # save and quit
• :q # quit (no save)
• :q! or ZQ # force quit (discard changes)

Nano

• [sudo] nano file.txt # open file.txt using nano

• Ctrl + O # save

• Ctrl + X # exit

• Replace entire contents:

  Ctrl + / -> 1                         # go to start of line 1
  Ctrl + ^                              # set mark
  Ctrl + / -> large line (or Ctrl + V)  # go to end of last line
  Ctrl + K                              # cut selection
  Ctrl + Shift + V                      # Paste system copy buffer
  

Find executables based on $PATH

# Linux, macOS -- find executable that will run for a command (first match in PATH)
which python    # may miss aliases
type python     # better
whereis python  # find bin + src + man

# Windows CMD -- show all matches of executable in PATH
where python

Linux & MacOS commands

history | grep python   # show lines containing "python"

<command> &             # run job in background
CTRL + Z                # put in background
fg                      # bring job into foreground

nohup <command> &       # ignore hangup command (shell close / ssh disconnect) and start in background

TODO “screen”

• List directory

  • ls -lahF -l = long; -a = show hidden; -h = readable sizes; -F = classify (file type)

  • ls -ltr list sorted by timestamp in reverse order

  • ls -ltc list sorted by change time

  • ls -lS list sorted by size

  • ls -lX list sorted by extension

  • ls -1 short format, separated by lines

  • Add alias to ~/.bashrc

    echo 'alias ll="ls -lahF --color=auto"' >> ~/.bashrc && source ~/.bashrc # linux
    echo 'alias ll="ls -lah --color=auto"' >> ~/.zshrc && source ~/.zshrc # mac
    

  • ls -1A | wc -l count num files + dirs + hidden

• less, grep, find
• disk free df, disk usage du
• uname,
• create user,
• change password,
• chmod, ch permissions
• .bashrc + create aliases
• File management
  • Move mv
    • mv file.txt /dest/ -> move file keep fname
    • mv file.txt newname.txt -> move file rename
    • mv folder/ /dest/ -> move directory
  • Copy cp
    • cp file.txt /dest/ -> copy file keep fname
    • cp file.txt /dest/newname.txt -> copy file rename
    • cp -r folder/ /dest/ -> copy dir recursive
    • cp -a folder/ /dest/ -> copy dir preserve metadata
    • cp -a /source/folder/*.png /dest/folder/ -> copy all PNGS keep filenames and metadata
  • Remove rm
    • rm file.txt -> remove single file
    • rm -r ./dirname -> remove dir contents
    • rm -rf ./dirname -> remove dir without questions
• Redirection operators (>, >>, |)
  • > write stdout to file: echo "hello" > file.txt (overwrites if file exists)
  • >> append redirect stdout: echo "hello" >> file.txt
  • | pipe stdout -> stdin: cat file.txt | grep hello
  • STDERR (2)
    • redirect stderr: command 2> err.txt
    • redirect both: command > all.txt 2>&1
    • pipe strout,stderr -> stdin: command 2>&1 | grep error
  • Frequent combinations
    • ls -1 > files.txt # save output
    • echo "log" >> app.log # append log
    • make > build.log 2>&1 # capture everything
    • cat file | sort | uniq # chain commands

Windows

winget (Windows package manager)

winget list                                   # list installed packages
winget list --source winget                   # only list manageable with winget

winget search vscode                          # search by package name
winget show Microsoft.VisualStudioCode        # package details (versions, installer info)

winget install voidtools.Everything           # install package via lookup
winget install -e --id WinDirStat.WinDirStat  # install specific package via ID

# Must haves
winget install voidtools.Everything WinDirStat.WinDirStat FFmpeg Klocman.BulkCrapUninstaller 
winget install VideoLAN.VLC Rufus.Rufus IrfanSkiljan.IrfanView IrfanSkiljan.IrfanView.PlugIns OBSProject.OBSStudio Notepad++.Notepad++

winget upgrade                                # show packages with available upgrades
winget upgrade WinSCP.WinSCP                  # upgrade a specific page
winget upgrade --id <Package.ID>              # via specific package ID (more precise)
winget upgrade --all                          # upgrade all packages

winget uninstall --id <Package.ID>            # uninstall specific package

winget source list                            # list available sources (repos)
winget source update                          # update sources
winget settings                               # open config file

CMD

dir /a | findstr "sandbox"      # list all and pipe to string match (like ls -la | grep sandbox)
dir /a | findstr /i "sAnDboX"   # case insensitive

Powershell

Get-ChildItem -Force | Where-Object { $_.Name -match "sandbox" } # list subdiles/subdirs with "sandbox" substring in name

macOS

~/.zshrc

# Prompt customization
PROMPT='%F{242}[%*]%f %F{33}%d%f %# '

Homebrew (MacOS package management)

/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)" # install

brew update           # update brew itself + formulae
brew install wget     # install specific package
brew uninstall wget   # uninstall specific package
brew reinstall wget   # reinstall specific package
brew upgrade wget     # upgrade specific package
brew list             # list installed
brew leaves           # list top-level explicit installs

# cask (GUI apps)
brew install --cask google-chrome
brew list --cask

brew search python
brew info python
brew upgrade          # upgrade all installed
brew cleanup          # remove old versions
brew doctor           # check for issues
brew outdated         # list outdated packages

# Brew services
brew services start postgresql
brew services stop postgresql
brew services list

Other & Tools

TODO: regex

find . | grep "\.mp4"   # find all files/dirs recursively with .mp4 as substr
ll | grep "\.mp4"       # same but just pwd
grep '\.mp4' file.txt   # find all lines with .mp4 substr in file.txt
rg "\.mp4$" file.tct    # sudo apt install ripgrep, -N to hide linenums
grep -vE '^(\./survey/|\./demo/)' "files_2026-03-18.txt" > files_noextra.txt # filter down to files NOT containing one of those two strings

TODO: ffmpeg

ffprobe --hide-banner input.mp4           # full media info
ffprobe -v error -show_streams input.mp4  # stream info, codecs, bitrate
ffprobe -v error -show_format input.mp4   # format and container info
ffmpeg -i input.mp4                       # quick info less structured

ffmpeg -i input.mkv -c:v h264_nvenc -qp 23 -profile:v high -pix_fmt yuv420p -preset:v p5 -rc:v vbr output.mp4 # reencode as mp4 using h264 nvidia encoder

ffmpeg -i URL -c copy output.mp4                              # download stream as local file

ffmpeg -i input.mp4 output.wav                                # extract audio (no resampling)
ffmpeg -i input.mp4 -ac 1 -ar 16000 output.wav                # extract audio + convert to mono 16 kHz
ffmpeg -i input.mp4 -ac 1 -ar 16000 -c:a pcm_s16le output.wav # PCM for common ASR format

ffmpeg -ss 00:01:00 -to 00:02:30 -i input.mp4 -c copy output.mp4                # cut video between 1 min and 2 min 30 sec (no reencode)
ffmpeg -ss 00:01:00 -to 00:02:30 -i input.mp4 -c:v libx264 -c:a aac output.mp4  # cut and reencode (slower)
ffmpeg -ss 00:01:00 -t 30 -i input.mp4 -c copy output.mp4                       # cut 30 seconds from 1-min timestamp

# burn subtitles onto video
ffmpeg -i "video.mkv" -vf "subtitles='subs.srt'" -c:v libx264 -crf 20 -preset medium -c:a copy video_subs.mkv # CPU
ffmpeg -i "video.mp4" -vf "subtitles='subs.srt'" -c:v h264_nvenc -pix_fmt yuv420p -cq 25 -preset p5 -c:a copy video_subs.mkv # nvidia
ffmpeg -i "video.mkv" -vf "scale=-2:540,subtitles='subs.srt'" -c:v h264_nvenc -pix_fmt yuv420p -cq 25 -preset p5 -c:a copy video_subs.mkv # nvidia + rescale

• TODO: mac-specific convert videos and audio
• TODO: resize image

yt-dlp

python3 -m pip install -U "yt-dlp[default]"                       # install/upgrade with pip
yt-dlp -F https://www.youtube.com/watch?v=n8X9_MgEdCg             # list available formats
yt-dlp -f 244+140 https://www.youtube.com/watch?v=n8X9_MgEdCg     # download video format 244 (HD 480p) + audio format 140 (m4a 128kbps) !requires ffmpeg for merging
yt-dlp -o yt-dlp -o "%(title)s - %(upload_date)s.%(ext)s" "URL"   # -o options: %(title)s: video title; %(uploader)s: channel; %(upload_date)s: YYYYMMDD; %(id)s: unique video ID.

yt-dlp -f 244+140 https://www.youtube.com/watch?v=n8X9_MgEdCg --print after_move:filepath # force output path on last line

Satisfy JS requirements

curl -fsSL https://deno.land/install.sh | sh # linux path: ~/.deno/bin/deno
winget install DenoLand.Deno # win path -> C:\Users\masly\.deno\bin\deno.exe

yt-dlp -F URL --js-runtimes "path/to/deno" --remote-components ejs:github   # hardcode deno path
yt-dlp -F URL --js-runtimes deno --remote-components ejs:github             # if deno path on OS is properly configured

Communication and file transfer

ssh -i ~/.ssh/id_rsa -p PORT user@host

scp -i ~/.ssh/id_rsa -P PORT file.txt user@host:/remote/path/     # upload (local -> remote)
scp -i ~/.ssh/id_rsa -P 2222 user@host:/remote/path/file.txt .    # download (remote -> local)
scp -i ~/.ssh/id_rsa -P 2222 -r folder/ user@host:/remote/path/   # recursive copy

# -a = archive recursive + preserve metadata; -v = verbose; -z = compress during transfer; -P (--partial) = resume on same command run
rsync -avz -e "ssh -i ~/.ssh/id_rsa -p 2222" folder/ user@host:/remote/path/            # upload directory
rsync -avz -e "ssh -i ~/.ssh/id_rsa -p 2222" user@host:/remote/path/ folder/            # download directory
rsync -avz --delete -e "ssh -i ~/.ssh/id_rsa -p 2222" folder/ user@host:/remote/path/   # delete extra files
rsync -avz --progress -e "ssh -i ~/.ssh/id_rsa -p 2222" folder/ user@host:/remote/path/ # progress

TODO: FRP (fast reverse proxy)

TODO: netcat, simple ports comms tests

zip

sudo apt update && sudo apt install zip unzip
zip -r assets_backup_date.zip assets/           # zip directory recursively
unzip -l backup.zip                             # list contents
unzip assets_backup_date.zip -d path-to-assets/ # unzip to specific directory

wget

wget -O ~/Downloads/name.zip https://example.com/remote.zip     # download + rename + location
wget -c -O ~/Downloads/name.zip https://example.com/remote.zip  # resume download after interrupt
wget -P ./data/raw/ https://example.com/remote.json             # keep original filename but save elsewhere
wget --limit-rate=500k https://example.com/remote.zip           # limit download speed
wget -b https://example.com/massive-remote.tar.gz               # download in background (-b)
tail -f wget-log                                                # check progress

Git

Install, update, set identity

sudo add-apt-repository ppa:git-core/ppa -y && sudo apt update && sudo apt install git -y # Debian/Ubuntu install/update
git update-git-for-windows # update on windows

# plaintext. win: C:/Users/user/.gitconfig. mac/linux: ~/.gitconfig
git config --global user.name "Mykola Maslych"
git config --global user.email "maslychm@gmail.com"
git config --global --edit # open global config file to edit

# to keep email private, set it to private one provided by GitHub
# https://github.com/settings/emails
# https://docs.github.com/en/account-and-profile/how-tos/email-preferences/setting-your-commit-email-address
git config --global user.email XXXXXXXX+username@users.noreply.github.com

Basics

git status
git add .                       # stage all local changes
git commit -m "commit message"  # commit staged changes

git restore --staged <file_name>  # unstage file after "git add <file_name>"
git restore --staged .            # unstage all staged files
git restore .                     #! discard unstaged changes

git reset HEAD~1        # undo last local commit, keep changes ustaged
git reset --soft HEAD~1 # undo last local commit, keep changes staged
git reset --hard HEAD~1 # undo last local commit, discard tracked changes

git branch -a           # list all branches
git branch -vv          # show relationships between local and origin branches
git log                 # interactive scrollable commit list. "q" to exit

git clean -n            # preview untracked changes that will be deleted
git clean -fdx          # delete all untracked files (-f), directories (-d), ignored (-x)
git reset --hard HEAD && git clean -fxd # nuke all local tracked and untracked changes

Rewrite commit history

git commit --amend --no-edit              # combine staged with last commit (--amend), keep message (--no-edit)
git push origin main --force-with-lease   # safe, fails if remote has moved on
git push origin main --force              # overwrite remote history with local history

SQUASH last 3 commits into 1

• Using soft reset: keep changes from last 3 commits as staged -> commit all changes as 1 commit

  git reset --soft HEAD~3
  git commit -m "Single commit for the feature"
  

• Using interactive REBASE

  git rebase -i HEAD~3 
  # -> keep "top" one as "pick" and and others as "s" -> save -> close. Example:
  ---
  pick a1b2c3d Initial feature logic
  s e5f6g7h Fix typo in logic
  s i9j0k1l Final touches for feature
  ---
  git push
  

Merge new branch into main (main stayed as of local ‘feature’ branch creation)

git switch -c feature       # + make changes -> commit
git push -u origin feature  # --set-upstream (-u) creates link between local feature
git checkout main
git merge feature
git push origin main

Merge new branch into main (main advanced after local ‘feature’ branch creation)

git switch -c feature # + make changes -> commit
git checkout main
git pull origin main
git checkout feature
git merge main
git checkout main
git merge feature
git push origin main

Rebase for linear history: merge new branch into main (main advanced after ‘feature’ branch creation)

## switch -> change -> commit -> push
git checkout main
git pull origin main
git checkout feature
git rebase main
git checkout main
git merge feature

Git terminology

• remote - non-local version(s), can be multiple servers

• origin - primary remote server.

• upstream - source

  • With forks: origin is the fork, upstream is the original repo. Typical flow: pull changes from upstream into local, push them to origin (fork), submit pull request to upstream.
  • With branches: local main tracks upstream origin/main

• Common cases

  • origin is automatically set with git clone <url>

  • Manually set origin for new local project with

    git init
    git remote add origin https://github.com/org/project.git
    git remote -v # verify, show fetch and push origins
    

  • Manually set origin for a new feature branch

    git switch -c feature
    git push -u origin feature
    

  • Manually link a local branch feature to origin/feature branch

    git branch --set-upstream-to=origin/feature feature
    

  • Change remote URL

    git remote set-url origin https://github.com/org/different-project.git
    

• TODO: change commiter info and rewrite history (ex case: wrong email -> no account link)

• TODO: configuring keys

SLURM

sbatch jobname.sh               # start a job
squeue -u mmaslych              # check jobs queue for user
scancel -u mmaslych             # cancel all jobs
scancel -t RUNNING -u mmaslych  # cancel only running jobs
srun --partition=gpu --gres=gpu:1 --cpus-per-task=4 --mem=16G --pty bash -i # interactive bash shell

sreport cluster UserUtilizationByAccount Users=mmaslych Start=2026-01-01 -t Minutes -T cpu,gres/gpu # check all utilization for user
sreport -T gres/gpu cluster AccountUtilizationByUser mmaslych Start=3/1/26  # check GPU minutes
sreport cluster AccountUtilizationByUser mmaslych Start=3/1/26              # check CPU minutes

latexdiff

brew install latexdiff          # macOS install
sudo apt-get install latexdiff  # Linux install

latexdiff v1/main.text v2/main.tex > diff.tex

• Highlight changes between PDF versions

• More instructions: https://www.overleaf.com/learn/latex/Articles/How_to_use_latexdiff_on_Overleaf

• To only show additions in blue, change these lines in DIF preamble below (removes \uwave from added, remove #1 from deleted):

  -\providecommand{\DIFaddtex}[1]{{\protect\color{blue}\uwave{#1}}}  %DIF PREAMBLE
  +\providecommand{\DIFaddtex}[1]{{\protect\color{blue}#1}}          %DIF PREAMBLE
  -\providecommand{\DIFdeltex}[1]{{\protect\color{red}\sout{#1}}}    %DIF PREAMBLE
  +\providecommand{\DIFdeltex}[1]{{\protect\color{red}\sout{}}}      %DIF PREAMBLE
  

LLMs, Neural Nets, Applications

Huggingface

https://huggingface.co/docs/huggingface_hub/en/guides/manage-cache

pip install -U huggingface_hub

hf cache ls                                               # list cached models
hf cache rm model/drbaph/OmniVoice-bf16                   # remove specific ID
hf cache rm $(hf cache ls --filter "accessed>1y" -q) -y   # remove all models not accessed for a year

Gaussian Splatting: WSL + CUDA

• Requirements: WSL + nvidia toolkit 12.6; build colmap; build opensplat

mkdir PROJNAME & cd PROJNAME
mkdir images & mkdir sparse

# Extract frames at 2 FPS (make sure to delete blurry)
ffmpeg -i ~/videos/IMG_0467.MOV -vf "fps=2" -qscale:v 2 images/frame_%04d.jpg

# colmap-style features/points (takes ~30 seconds on RTX4070 for 350 frames)
colmap feature_extractor --database_path database.db --image_path images --SiftExtraction.use_gpu 1

# match points between frames (takes ~4 minutes on RTX 4070 for 350 frames)
colmap exhaustive_matcher --database_path database.db --SiftMatching.use_gpu 1
# exhaustive_matcher -> match every frame with every frame
# sequential_matcher -> match only in order
# vocab_tree_matcher -> use .bin vocab to search for images and only match similar ones with each other

# actual mapping (takes ~15 minutes on RTX 4070 for 350 frames)
colmap mapper --database_path database.db --image_path images --output_path sparse

# train splat (takes anywhere between 20 minutes and 2 hours)
~/OpenSplat/build/opensplat . --colmap-image-path images -d 2 -o PROJNAME.ply -n 30000 -s 5000
~/OpenSplat/build/opensplat . --colmap-image-path images -d 2 -o hec208_d2_controlled.ply -n 45000 --sh-degree 2 --num-downscales 2 --resolution-schedule 2500 --warmup-length 1500 --refine-every 250 --densify-grad-thresh 0.001 --split-screen-size 0.12 --stop-screen-size-at 2000

# TODO splat viewer

llama.cpp

Build latest on Ubuntu with CUDA support

sudo apt install build-essential git cmake libcurl4-openssl-dev -y
sudo apt install nvidia-cuda-toolkit -y
sudo apt install openssl
sudo apt install gcc-10 g++-10 -y

git clone https://github.com/ggerganov/llama.cpp
cd llama.cpp
mkdir build && cd build

export FORCE_CMAKE=1
export CMAKE_ARGS="-DGGML_CUDA=on -DLLAMA_OPENSSL=ON -DCMAKE_CUDA_HOST_COMPILER=/usr/bin/g++-10 -DCMAKE_CXX_STANDARD=17"
cmake ..
cmake --build . --config Release -j $(nproc)

Run models

cd llama.cpp
./build/bin/llama-server -m ~/Downloads/Meta-Llama-3.1-8B-Instruct-Q4_K_M.gguf -ngl 999 --port 8082 --host 0.0.0.0 --ctx-size 16384 -lv 2
./build/bin/llama-server -hf ggml-org/gemma-4-E4B-it-GGUF

• TODO: continuous batching, prompt caching, async, logprobs

TODO: Ollama

• install on platforms
• set custom port