Demonstrates that Apple Neural Engine (ANE) achieves significantly higher throughput with INT8 W8A8 quantization vs FP16, consistent with native INT8 datapath support.
| Method | FP16 | INT8 W8A8 | Ratio |
|---|
Demonstrates that Apple Neural Engine (ANE) achieves significantly higher throughput with INT8 W8A8 quantization vs FP16, consistent with native INT8 datapath support.
| Method | FP16 | INT8 W8A8 | Ratio |
|---|
| magick input.png -resize "2880x1800>" -background transparent -gravity center -extent 2880x1800 output.png |
| # Examples: | |
| # python cubemap_stitcher.py /path/to/your/images | |
| # or | |
| # python cubemap_stitcher.py ./my_cubemap_folder --output ./final_skybox.jpg --width 4096 --ext .jpg | |
| from pathlib import Path | |
| from typing import Optional | |
| import typer | |
| import numpy as np |
| import bpy | |
| import math | |
| def create_3d_rotation_arrow(): | |
| # Parameters to tweak the look of your arrow | |
| radius = 2.0 # Radius of the overall circle | |
| thickness = 0.2 # Thickness of the curved pipe | |
| arrow_radius = 0.5 # Width of the arrowhead | |
| arrow_length = 1.0 # Length of the arrowhead | |
| start_angle = 0.0 |
| import bpy | |
| import bmesh | |
| import math | |
| # ============================================================ | |
| # Toon / technical-diagram outline setup using Freestyle | |
| # ============================================================ | |
| # | |
| # What this does: |
| import bpy | |
| def create_toon_shader(): | |
| mat_name = "Toon_3_Tone" | |
| # Create a new material and enable use of nodes | |
| mat = bpy.data.materials.new(name=mat_name) | |
| mat.use_nodes = True | |
| # Get the node tree |
| name: Publish Debian Package to R2 | |
| on: | |
| push: | |
| tags: | |
| - 'v*.*.*' # Triggers only when you push a version tag (e.g., v1.0.0) | |
| jobs: | |
| build-and-publish: | |
| runs-on: ubuntu-latest |
A pattern for building personal knowledge bases using LLMs.
This is an idea file, it is designed to be copy pasted to your own LLM Agent (e.g. OpenAI Codex, Claude Code, OpenCode / Pi, or etc.). Its goal is to communicate the high level idea, but your agent will build out the specifics in collaboration with you.
Most people's experience with LLMs and documents looks like RAG: you upload a collection of files, the LLM retrieves relevant chunks at query time, and generates an answer. This works, but the LLM is rediscovering knowledge from scratch on every question. There's no accumulation. Ask a subtle question that requires synthesizing five documents, and the LLM has to find and piece together the relevant fragments every time. Nothing is built up. NotebookLM, ChatGPT file uploads, and most RAG systems work this way.
| import torch | |
| from transformers import Sam3Processor, Sam3Model | |
| from PIL import Image | |
| import requests | |
| # Target your Nvidia GPU | |
| device = "cuda" if torch.cuda.is_available() else "cpu" | |
| # Automatically download and load SAM 3 to the GPU | |
| print("Loading model...") |
| """Simple CLI example: convert a PDF file to Markdown using pdf_oxide.""" | |
| from pathlib import Path | |
| import typer | |
| from pdf_oxide import PdfDocument | |
| app = typer.Typer( |