bwasti

# This is a test (not implementation) of the impact bucketMul has on matrix multiplications
# https://kolinko.github.io/effort/bucketmul.html
import torch
import torch.nn.functional as F
import math
torch.manual_seed(1337)

B = 2
N = 8
M = 16

bwasti

Here's the API interface to bun:sqlite,

class Database {
  constructor(
    filename: string,
    options?:
      | number
      | {
 readonly?: boolean;

bwasti

import time
import multiprocessing

def test_lock(lock, iterations, shared_value):
    for _ in range(iterations):
        with lock:
            shared_value.value += 1

def benchmark(lock_type, num_processes, iterations_per_process):
    shared_value = multiprocessing.Value('i', 0)

bwasti

import torch

def cudagraph(f):
    _graphs = {}
    def f_(*args):
        key = hash(tuple(tuple(a.shape) for a in args))
        if key in _graphs:
            wrapped, *_ = _graphs[key]
            return wrapped(*args)
        g = torch.cuda.CUDAGraph()

bwasti

console.log("Oh no! bun hit an error\n")
const report = 'bun report blah blah blah'
const url = `https://duckduckgo.com/?q=${report}`
const str = '[click to upload report]'
console.log(`\u001b]8;;${url}\u001b\\${str}\u001b]8;;\u001b\\`)

bwasti

read -p "this script will remove libarrayfire from your system and install all requirements to build from source. continue? [Y/n]" -n 1 -r
echo
if [[ $REPLY =~ ^[Yy]$ ]]
then
  sudo apt remove libarrayfire-dev libarrayfire-cpu3 libarrayfire-cpu-dev
  sudo apt install -y libblas-dev liblapack-dev liblapacke-dev libfftw3-dev libboost-all-dev cmake make g++
  cd /tmp
  rm -rf arrayfire
  git clone https://github.com/arrayfire/arrayfire.git
  cd arrayfire

bwasti

# conv bwd implemented with fwd functions

import torch
import torch.nn.functional as F


def dconv2d(grad, x, w, stride, padding, groups):
    batch = grad.shape[0]
    channel_out = grad.shape[1]
    channel_in = x.shape[1]

bwasti

# example of backward pass implemented with only forward functions

import torch
import torch.nn.functional as F


def dconv(grad, x, w, stride, padding, groups):
    batch = grad.shape[0]
    channel_out = grad.shape[1]
    channel_in = x.shape[1]

bwasti

# examples of backward passes implemented with fwd functions

import torch
import torch.nn.functional as F


def simple():
    print("simple")
    x = torch.randn(1, 1, 4, 4)
    x.requires_grad = True

bwasti

Shumai's io namespace

Inference

Want to expose your machine learning model to the network? Use io.serve:

import * as sm from '@shumai/shumai'
import { model } from './model'