Loc Vinh Giang raspiduino

Fix PCIe allocation error when connecting RTX50xx series eGPU to Linux laptop via Thunderbolt

UPDATE: Ubuntu 25.10 completely fixed this issue. This fix is NOT required for Ubuntu 25.10

TLDR

Skip to solution

Introduction

https://www.youtube.com/watch?v=_36yNWw_07g

How to download undownloadable PDF from Heyzine flipbook (as of June 2025)

(Seriously?)

The following instructions are for Chrome, user may perform similar actions when using other browsers

Step 1: F12 inspect the flipbook page
Step 2: Go to Application then Session storage and click on https://heyzine.com and you should see this:
Step 3: Get PDF url from the Key column. Original key should be something like hz_cache_1201_p_10_normal_https://cdnc.heyzine.com/files/uploaded/v3/[REDACTED].pdf.
Step 4: Your PDF is at https://cdnc.heyzine.com/files/uploaded/v3/[REDACTED].pdf
Step 5: Enjoy!

Collection of stupid methods to stream H.265 RTSP from an IP camera to Youtube without video transcoding

Why?

I have a IP cam that output RTSP stream with:

H.265 video codec
Weird resolution of 1920x2160 (which is the result of stacking two 1920x1080 streams vertically into one frame)
Super unstable timing (pts + dts is not even provided)
15fps when stable (yes that's what the camera output)

and I want to stream that to Youtube (using ffmpeg)

About MemGPT's free endpoint

MemGPT provides you a free endpoint for trying. It's at https://inference.memgpt.ai/chat/completions From official docs it's claimed that the free endpoint is running on variants of Mixtral 8x7b:

MemGPT Free Endpoint: select this if you'd like to try MemGPT on a top open LLM for free (currently variants of Mixtral 8x7b!)

However, after I manually try running Mixtral 8x7b on my own machine, I saw that the model cannot be compared to the free endpoint in term of accuracy (function calling + response). This makes me want to find out the real model behind this endpoint.

TL;DR:

Free endpoint just forwards your call to OpenAI's ChatGPT 3.5. However, I cannot be sure if they log our requests or not.

	import os
	import shutil
	import json
	import logging
	import re
	from abc import ABC, abstractmethod
	from typing import BinaryIO, Tuple, Dict

	from open_webui.config import (
	S3_ACCESS_KEY_ID,

	#!/usr/bin/python3
	import os
	import base64
	import hashlib
	from datetime import datetime, timezone
	import requests
	from http.client import RemoteDisconnected

	# This function is literally imported from Gemini, I can't write comments this detailed and format this properly :)
	def generate_soap_header_values(plain_password: str) -> dict:

	root@cado:~# mkdir crack
	root@cado:~# cd cado-nfs/
	root@cado:~/cado-nfs# ./cado-nfs.py 7071142888328001159781190611496955124398513403641364137666269609876069075501385806782438994383629333720075608597235677504440240558571591145514673066702347 server.address=192.168.1.9 server.ssl=no server.whitelist=REDACTED workdir=/root/crack/
	Info:root: Using default parameter file ./parameters/factor/params.c155
	Info:root: No database exists yet
	Info:Database: Database URI is db:sqlite3:///root/crack//c155.db
	Info:Database: Opened connection to database /root/crack//c155.db
	Info:root: Set tasks.linalg.bwc.threads=20 based on detected physical cores
	Info:root: Set tasks.threads=20 based on detected logical cpus
	Info:root: tasks.threads = 20 [via tasks.threads]

	D:\dev>git clone --recursive https://github.com/mapillary/OpenSfM
	D:\dev>cd OpenSfM
	D:\dev\OpenSfM>git clone https://github.com/microsoft/vcpkg
	Cloning into 'vcpkg'...
	remote: Enumerating objects: 247401, done.
	remote: Counting objects: 100% (59827/59827), done.
	remote: Compressing objects: 100% (4389/4389), done.
	remote: Total 247401 (delta 55699), reused 55438 (delta 55438), pack-reused 187574 (from 1)
	Receiving objects: 100% (247401/247401), 75.41 MiB \| 10.16 MiB/s, done.
	Resolving deltas: 100% (164296/164296), done.

	# Import
	import cv2
	from deepsparse.pipeline import Pipeline
	from deepsparse.yolo.schemas import YOLOInput
	from deepsparse.yolo.utils import COCO_CLASSES
	import time

	# Model settings
	task = "yolo"
	model_path = "zoo:yolov5-l-coco-pruned.4block_quantized"