Luvata

import os
import datetime
import random
from typing import Callable, Dict, List, Tuple, Any
import logging
import warnings

import torch
import torch.nn.functional as F
from transformers import AutoModelForCausalLM, AutoTokenizer, PreTrainedModel

Luvata

#!/bin/bash

cd /content/sapiens/lite || exit
SAPIENS_CHECKPOINT_ROOT=/content/sapiens_0.3b/sapiens_lite_host

MODE='torchscript' ## original. no optimizations (slow). full precision inference.
# MODE='bfloat16' ## A100 gpus. faster inference at bfloat16
# MODE='float16' ## V100 gpus. faster inference at float16 (no flash attn)

SAPIENS_CHECKPOINT_ROOT=$SAPIENS_CHECKPOINT_ROOT/$MODE

Luvata

A simple message queue application

How to run

1. First run the server

uvicorn app:app --reload

2. Interacting with the server using cli.py

Luvata

import os
import shutil
import tarfile
import zipfile
import requests
import re
import sys


def remove_preamble(latex):

Luvata

import os
import shutil
import tarfile
import zipfile
import requests
import fnmatch
import re
import sys

Luvata

Summarize this paper

The summary is derived from the current web page context.

This paper presents DragGAN, a method for interactive point-based manipulation of images generated by GANs. The user can click on any points on the image and drag them to desired positions, and the method will deform the image accordingly. DragGAN consists of two components: a feature-based motion supervision that guides the points to move towards their targets, and a point tracking approach that uses the discriminative features of the generator to locate the points on the image. DragGAN can control various spatial attributes such as pose, shape, expression, and layout of different object categories. DragGAN produces realistic results even for challenging scenarios such as occluding or deforming parts of the object. The paper compares DragGAN with prior methods and shows its advantages in image manipulation and point tracking. The paper also demonstrates how DragGAN can be applied to real images using GAN inversion.

Source: Co

Luvata

Tải file lớn từ google drive

https://stackoverflow.com/a/67550427

1. Vào https://developers.google.com/oauthplayground/
2. "Select the scope", chọn https://www.googleapis.com/auth/drive.readonly
3. Chọn "Authorize APIs" và chọn acc google
4. Chọn "Exchange authorization code for tokens"
5. Copy Access token

Luvata

Cách apply TRC và sử dụng TPU-VM

1. Đăng ký TRC

1. Apply TRC: https://sites.research.google/trc/
2. Đợi mail, trong thời gian đó mở cloud console (cùng mail đăng ký TRC), tạo project mới
3. Khi có mail, điền project number vào form (Nhìn phần dashboard -> Project info -> Project number)
4. Đợi confirm (khoảng 1-2 tiếng), sẽ có thông tin là được những TPU nào tại Zone nào

2. TPU - VM

• Dựa theo talk này https://youtu.be/fuAyUQcVzTY để tạo TPU-VM. Cụ thể cần làm các bước sau:

Luvata

"""
This file uses code from 
https://github.com/t-vi/pytorch-tvmisc/blob/master/hacks/visualize-jit-models.ipynb


MIT License

Copyright (c) 2017 Thomas Viehmann <tv.code@beamnet.de>

Permission is hereby granted, free of charge, to any person obtaining a copy

Luvata

from load_data import DataGenerator
import matplotlib.pyplot as plt


N = 10
K = 1
B = 3
generator = DataGenerator(num_classes=N, num_samples_per_class=K*2)

all_imgs, all_labels = generator.sample_batch("train", B) # (3, 2, 10, 784)