Skip to content

README.md

Latest commit

 

History

History
146 lines (100 loc) · 4.33 KB

README.md

File metadata and controls

146 lines (100 loc) · 4.33 KB
Raw

PythonTools

Public drawer for small Python utilities, diagnostics, experiments, and script templates.

This is not a single polished application. It is a lightweight collection of tools that are useful enough to keep public, but not necessarily large enough to deserve their own repository.

Some tools that started here have graduated:

  • ArchiveUpdater moved to gitSpecOps.
  • github-org-duplicator lives in gitSpecOps.
  • PixelViewer moved into DesktopOpener.

LLM_Tools

Utilities for local LLM work, hardware context, and agent-friendly reporting. These scripts are intentionally small and can be run directly with Python; each file also has a top-of-file usage note for agents and humans reading the source.

generate_folder_structure.py

Generates a Markdown folder tree for a target project. Excludes common development artifacts such as .git, .venv, __pycache__, and build outputs.

python LLM_Tools/generate_folder_structure.py
python LLM_Tools/generate_folder_structure.py --path /path/to/project
python LLM_Tools/generate_folder_structure.py --org
python LLM_Tools/generate_folder_structure.py --path /path/to/project --out map.md

Generated maps are written to LLM_Tools/Data/folder_structure.md by default. Use --out to place the map somewhere else. The tool no longer depends on a Claude-specific notes folder.

hardware_detector.py

Detects hardware and formats output for humans, compact LLM context, or verbose debugging.

python LLM_Tools/hardware_detector.py
python LLM_Tools/hardware_detector.py --mode llm
python LLM_Tools/hardware_detector.py --mode verbose --save

No required third-party dependency. Uses psutil opportunistically if it is installed. Supports Windows, macOS, and Linux with platform-specific probes; missing optional tools simply leave some fields empty. --save writes to LLM_Tools/Data/settings.json by default.

sensor_monitor.py

Reads live hardware sensor data from existing monitor streams when available: HWiNFO64 shared memory, MSI Afterburner shared memory, LibreHardwareMonitor or OpenHardwareMonitor WMI, nvidia-smi, and Windows thermal fallbacks.

python LLM_Tools/sensor_monitor.py
python LLM_Tools/sensor_monitor.py --sources
python LLM_Tools/sensor_monitor.py --mode llm
python LLM_Tools/sensor_monitor.py --stream --out sensors.jsonl

Standard-library Python. Best results are currently Windows-first because the richest sensor integrations are Windows monitor tools. Linux and macOS can still report NVIDIA GPU metrics through nvidia-smi when it is installed.

Local_Benchmark

Ollama benchmark scripts for comparing local models across context sizes, prompt sizes, and launch profiles.

python Local_Benchmark/ollama_bench.py --models qwen3.5:4b,qwen2.5:14b --contexts 4096,16384
python Local_Benchmark/ollama_bench_orchestrator.py --profiles nvidia,cpu

The worker expects an Ollama server to already be running. The orchestrator can restart Ollama with profile-specific environment variables.

Explorations

Small standalone explorations that are polished enough to keep around, but are not general utilities yet.

collatz.py

Interactive Collatz conjecture explorer. Plots one or more sequences on a zoomable, log-scale tkinter canvas with animation and an interactive legend.

python Explorations/collatz.py

Uses only the Python standard library.

Scraps

Scratch scripts, templates, old experiments, and one-off probes. These are kept for reference and future salvage, not presented as stable public tools.

Several scripts in this folder have extra dependencies such as pandas, pyautogui, PyQt5, pyvda, pygetwindow, or screeninfo.

Dependency Notes

The repo does not currently define one authoritative environment for every script. Treat each folder as a small tool area:

  • LLM_Tools uses only the Python standard library at runtime.
  • hardware_detector.py can use psutil when available.
  • Local_Benchmark uses standard-library Python but requires Ollama for real benchmark runs.
  • Scraps contains mixed experiments with mixed dependencies.

For this reason, this repo does not currently ship one root dependency file.

Folder Map

Regenerate the repository map with:

python LLM_Tools/generate_folder_structure.py

Local output:

LLM_Tools/Data/folder_structure.md