Skip to content

Instantly share code, notes, and snippets.

View howespt's full-sized avatar

Phil Howes howespt

3 followers · 3 following
  • San Francisco
View GitHub Profile
@howespt
howespt / README.md
Last active July 15, 2024 02:10
Sticky model

‎‎​

@howespt
howespt / unattended-upgrade.yaml
Last active April 20, 2024 04:05
unattended-upgrades.yaml
apiVersion: apps/v1
kind: DaemonSet
metadata:
name: no-nvidia-unattended-upgrades
namespace: gpu-operator
spec:
selector:
matchLabels:
name: no-nvidia-unattended-upgrades
template:
@howespt
howespt / OPENSSL.md
Last active August 22, 2023 23:34
SSL handshakes

Prod app.baseten.co

openssl s_client -connect app.baseten.co:443

CONNECTED(00000006)
write W BLOCK
---
Certificate chain
 0 s:/CN=app.baseten.co
@howespt
howespt / README.md
Last active August 25, 2023 22:57
L4 Benchmarking on GCP

Benchmarking L4 on GCP

This benchmark is designed to simulate the way our users typically load/run models in Baseten without additional optimisations. The goal is to get a baseline for how our users will experience Baseten.

Setup

Create a g2-standard-8 VM with running ubuntu-2004-focal-v20230715 image, 200GB nvme. This machine has 1x nvidia L4. SSH into the VM and run the following commands to install cuda:

wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64/cuda-ubuntu2004.pin
@howespt
howespt / README.md
Last active August 27, 2023 02:27
GPU benchmarking
@howespt
howespt / main.go
Last active May 31, 2023 03:01
beefeater
package main
import (
"bytes"
"encoding/json"
"errors"
"fmt"
"io/ioutil"
"log"
"net/http"
@howespt
howespt / loadtest.ts
Last active April 7, 2023 20:04
BasetenStabilityLoadtest
import http from "k6/http";
import { group, check } from "k6";
const BASE_URL = "https://app.baseten.stability.ai";
let accept = "application/json";
export const options = {
tags: {
host: "https://app.baseten.stability.ai",
},
scenarios: {
@howespt
howespt / README.md
Created August 16, 2021 21:04
FastAI example

  1. Export the model as model.pkl

  2. Deploy to baseten

import baseten

baseten.deploy_custom(
  model_name='FastAI demo',
  model_class='FastaiModel',
 model_files=['fai_model.py', 'model.pkl'],
@howespt
howespt / index.html
Created May 25, 2017 23:47
circle circles
<meta charset="utf-8">
<style>
body {
background-color: #77f2ff;
}
circle {
fill: none;
stroke: white;
@howespt
howespt / index.html
Last active May 25, 2017 18:30
circles on circles
<!DOCTYPE html>
<meta charset="utf-8">
<style>
path {
cursor: pointer;
fill: #eee;
stroke: #666;
stroke-width: 1.5px;
}