g023’s gists

g023 / ex1.htm

Last active June 19, 2025 01:12

3d Kart Game boilerplate raw examples (rough outs, may not work 100%)

	<!DOCTYPE html>
	<html lang="en">
	<head>
	<meta charset="UTF-8" />
	<title>Mini 3D Kart Game</title>
	<style>
	html, body {
	margin: 0;
	overflow: hidden;
	background: #000;

g023 / python_fps_movement.py

Last active June 7, 2025 01:29

Basic Python 3D FPS movement (OpenGL)

	import glfw
	from OpenGL.GL import *
	from OpenGL.GLU import *
	import math

	# pip install glfw PyOpenGL PyOpenGL_accelerate

	class FPSCamera:
	"""
	An FPS-style camera that processes input and calculates the corresponding Euler Angles,

g023 / glfw_imgui

Created May 29, 2025 04:24

Enhanced File Browser in Python using GLFW and ImGui

	"""
	Enhanced File Browser using GLFW and ImGui
	A modern, intuitive file browser with proper error handling, visual feedback,
	navigation with expandable folder tree and content pane.
	"""
	import os
	import sys
	import stat
	from pathlib import Path
	from typing import Dict, Set, Optional, Tuple

g023 / ocr_ollama.py

Last active May 16, 2025 05:53

OCR with Ollama - Using Local Vision Models to OCR an Image

	# Author: https://gist.github.com/g023
	# Link for this file: https://gist.github.com/g023/1ec7723745b17fec80f665951bca3d75
	# Convert text in an image to a text string.
	import requests
	import base64
	import json

	server_url = "http://localhost:11434/api/generate"
	model = "qwen2.5vl:7b-q4_K_M"
	prompt = "Extract all the text from this image and present it as a single string."

g023 / manual_webcam_control.py

Created April 21, 2025 05:31

Manual Webcam Control

	# Author: g023 - https://gist.github.com/g023
	import cv2
	import numpy as np
	from datetime import datetime
	import os
	import logging

	# Configure logging
	logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
	logger = logging.getLogger(__name__)

g023 / think_deep_explorer.py

Last active April 9, 2025 06:29

Ollama chat with a model's thoughts directly and force answer when desired.

	# Thinking Deep Explorer - Communicate with Thoughts
	# This program is a chat interface for the Ollama server.
	# In this chat, you communicate in the model's thoughts, and the model responds with its thought process.
	# Answers are forced by requesting "answer: <further answer instructions>" for a generic non-thinking response.
	# When in thinking response mode, only the thoughts will be presented.
	# Thoughts are passed forward as the conversation continues.
	# You can save by typing "save" and load by typing "load".
	# You can also clear the conversation by typing "clear"
	# Remove the last thought by typing "remove".
	# Chat with a model directly in its thoughts.

g023 / human-in-the-loop-llm-trainer.py

Created March 27, 2025 23:21

Human-In-The-Loop LLM fine tuning using Unsloth

	from unsloth import FastLanguageModel
	import torch
	from trl import SFTTrainer
	from transformers import TrainingArguments
	import datasets
	import os

	# description: This script allows users to interact with a language model,
	# either querying it or fine-tuning it with new data. The model is loaded
	# from a specified directory, and the user can choose to keep or replace

g023 / simple_text_predictor.py

Created March 26, 2025 02:51

This script is a simple n-gram text predictor that uses a frequency-based approach to predict the next token in a sequence. It can be trained on text files and can generate text based on the learned frequencies.

	# Description: This script is a simple n-gram text predictor that uses a frequency-based approach to predict the next token in a sequence. It can be trained on text files and can generate text based on the learned frequencies. The script also includes functions for processing text files, cleaning data, and generating text based on user input.

	# generates from all .txt files in a ./docs/ directory that is a directory in the same file this script is run.
	# will reduce power of model on multiple passes to help reduce and trims some of the bottom feeders.

	# Uses: Tiktoken for tokenization, multiprocessing for parallel processing, and chardet for encoding detection.

	# pip install tiktoken chardet

g023 / simple_text_predictor.py

Last active March 25, 2025 05:41

Simple Text Prediction - Read all .txt from ./docs directory.

	import random

	def predict_next_token(sequence, ngram_freq, max_n=3):
	for n in range(min(max_n, len(sequence) + 1), 0, -1):
	context = tuple(sequence[-(n - 1):]) if n > 1 else ()
	if context in ngram_freq[n]:
	next_tokens = ngram_freq[n][context]
	tokens, weights = zip(*next_tokens.items())
	return random.choices(tokens, weights=weights, k=1)[0]
	# Fallback to most frequent unigram

g023 / turbo_vector_rag_batch.py

Last active March 30, 2025 04:34

Turbo Vector RAG - Lean mean, multiple-instance with single LLM control with batch loading and Chunk Loading Strategies

	# Just a simple class to handle a vector db instance for RAG with a manager class to handle multiple instances at once while sharing the model between instances.
	# ability to save/load (saves as a pickle file) the vector db instance to disk (in folder ./_data/{md5}.pkl)
	# ability to share model between instances for memory efficiency and speed of loading
	# reference target RAG db with an id.
	# -- now in turbo (batch) mode for adding segments to the db instance (much faster)
	# -- automatically uses GPU if available
	# -- added ability to load a file
	# -- added multiple chunking strategies for documents (by sentences or by words using a sliding window with overlap)
	# -- Just query like an LLM

g023 g023