dot-aliases

A simple, cross-shell repository for keeping your aliases in one place and installing them on any machine in a single command.

📖 AI Agent Guide: See AGENT_BENCHMARK_GUIDE.md for comprehensive benchmark instructions for AI agents.

Files

• aliases.sh — Your shell aliases and functions (works in bash, zsh, and POSIX-compatible shells)
• bashrc.snippet — Loader snippet for Bash (~/.bashrc)
• zshrc.snippet — Loader snippet for Zsh (~/.zshrc)
• install.sh — Idempotent installer script
• python/vllm_test_infra/ — Python infrastructure for vLLM testing
• scripts/ — Python test scripts (gsm8k_eval.py, benchmark_compare.py)
• bash-scripts/ — Legacy bash scripts (deprecated, use Python scripts instead)

Quick Install

One-liner for any machine:

bash -c "$(curl -fsSL https://raw.githubusercontent.com/LucasWilkinson/dot-aliases/main/install.sh)"

This will:

1. Create ~/.config/aliases/ directory
2. Download aliases.sh to ~/.config/aliases/aliases.sh
3. Add a loader line to your ~/.bashrc or ~/.zshrc (idempotent—safe to run multiple times)
4. Load the aliases in your current session

Open a new shell or run:

source ~/.config/aliases/aliases.sh

Machine-Specific Overrides

Create ~/.config/aliases/local.sh for host-specific aliases or secrets. It will automatically be sourced at the end of aliases.sh if it exists:

# ~/.config/aliases/local.sh (not tracked in Git)
alias myhost='ssh user@myhost'
export MY_SECRET_TOKEN="..."

Auto-Update

Add this alias to your aliases.sh to keep aliases up-to-date:

alias aliases-update='curl -fsSL https://raw.githubusercontent.com/LucasWilkinson/dot-aliases/main/aliases.sh -o ~/.config/aliases/aliases.sh && source ~/.config/aliases/aliases.sh && echo "Aliases updated."'

Included Aliases

Navigation & Lists

• ll — ls -lah
• la — ls -la
• l — ls -la

Git

• gs — git status
• gp — git pull --rebase --autostash && git push
• gd — git diff
• gc — git commit
• ga — git add

Editors

• v — nvim
• vi — vim

System

• ports — sudo lsof -i -P -n | grep LISTEN (show listening ports)
• gpu — nvidia-smi or "No NVIDIA GPU"
• gpumem — Show GPU memory usage

vLLM Development

• venv — Activate vLLM virtual environment
• vllm-code — Jump to vLLM code directory
• vllm-build — Rebuild vLLM with CUDA kernels
• vllm-test — Run pytest with verbose output
• cleanup-vllm — Kill vLLM processes

vLLM Test Infrastructure

• benchmark — Benchmark current branch (standalone, simplified)
• benchmark-compare — Compare benchmarks across git branches
• gsm8k-eval — Run GSM8K evaluation (from any directory)
• vllm-test-infra-install — Install Python dependencies for test infrastructure

Development

• py / py3 — Python shortcuts
• reload — Reload ~/.bashrc
• aliases-edit — Edit aliases with $EDITOR
• aliases-reload — Reload aliases immediately

Safety

• cp — cp -i (interactive)
• mv — mv -i (interactive)
• rm — rm -i (interactive)

Customization

Edit aliases.sh and commit:

cd ~/dot-aliases
# Edit aliases.sh
git add aliases.sh
git commit -m "Update aliases"
git push

Then on any machine, run:

aliases-update

Or force a fresh pull on your current machine:

curl -fsSL https://raw.githubusercontent.com/LucasWilkinson/dot-aliases/main/install.sh | bash

vLLM Test Infrastructure

A reusable Python infrastructure for running vLLM benchmarks and evaluations with robust server management, process monitoring, and user interfaces.

Features

• Robust Server Management: Auto-detects GPU requirements, uses chg for reservation, monitors logs for fast-fail
• Process Monitoring: Signal handling, zombie cleanup, timeouts
• Dual UI Modes: Textual-based TUI for interactive use, simple stdout for AI agents
• Git Integration: Branch comparison, automatic builds
• Extensible: Easy to add new test scripts

Quick Start

Install dependencies:

vllm-test-infra-install
# Or manually: pip install -r ~/.config/aliases/dot-aliases/python/requirements.txt

Benchmark current branch (from any directory):

benchmark --model deepseek-ai/DeepSeek-V3.2-Exp -tp 8 --rates "1,5,10"

Run GSM8K evaluation (from any directory):

gsm8k-eval --model deepseek-ai/DeepSeek-R1 --limit 100

Compare benchmarks across branches (from any directory):

benchmark-compare \
  --model meta-llama/Meta-Llama-3-8B-Instruct \
  --rates 1,5,10 \
  --variants 'base::;full::-O.cudagraph_mode=FULL'

See python/README.md for full documentation.

Security Notes

• The installer is simple and readable—review it before running
• For extra safety, pin a specific commit:

  bash -c "$(curl -fsSL https://raw.githubusercontent.com/LucasWilkinson/dot-aliases/<COMMIT_SHA>/install.sh)"

• Use ~/.config/aliases/local.sh to keep secrets out of Git

License

MIT