tcapelle

Linux

Setup

When setting up a fresh VM.

sudo apt update
sudo apt-get install vim tmux wget

tcapelle

My Personal Top Wines: by Region

This is an arbitrary good list of wines I like, don't be offended if your favorite wine is not there. I have limited tasting time, and have mostly choose little family owned producers

Everything here is Organic, most are biodynamic and some are natural.

I have visited most of the winerys myself, and now I order directly from them or by my local wine-shop that actually helps me find more wine of this quality.

Loire

Probably my fav white wine region, I love the floral and sometimes minerality of these wines. There is always a good reason to open any of these bottles. It is also a very beautiful region to visit where you find accomodation on old castles like Chteau de Beaujeu.

tcapelle

import wandb
from wandb import Api

api = Api()
ENTITY = "fastai"
PROJECT = "fine_tune_timm"

project = api.project(PROJECT, entity=ENTITY)

sweeps = project.sweeps()

tcapelle

import os
import openai
from rich.console import Console

console = Console()

openai.api_key = os.getenv("OPENAI_API_KEY")
    
history = [{"role": "system", "content": "You are a helpful assistant."},]

tcapelle

tcapelle

import wandb
import timm
import argparse
from fastai.vision.all import *
from fastai.callback.wandb import WandbCallback
from torchvision import models

def parse_args():
    parser = argparse.ArgumentParser()
    parser.add_argument('--batch_size', type=int, default=64)

tcapelle

tcapelle

Setup NGC Docker

you will need to setup docker with nvidia runtime, containers

• Install docker and nvidia-docker:

sudo apt-get install -y docker nvidia-container-toolkit

tcapelle

from torch import nn
from functools import partial
from einops.layers.torch import Rearrange, Reduce

class PreNormResidual(nn.Module):
    def __init__(self, dim, fn):
        super().__init__()
        self.fn = fn
        self.norm = nn.LayerNorm(dim)

tcapelle

def _get_col_idxs(df, cols):
    "return cols index to perform iloc"
    return [df.columns.get_loc(c) for c in L(cols) if c in df]

def _iloc(df, rows, cols=None):
    "Iloc that supports col names"
    if isinstance(cols, (tuple, list, str)):
        cols = _get_col_idxs(df, cols)
        return df.iloc[rows, cols]
    return df.iloc[rows, slice(None)]