brunosan

./finetune/classify/classify.py: """ Command line interface to run the neural network model! From the project root directory, do: python classify.py fit --config configs/classify_eurosat.yaml References: - https://lightning.ai/docs/pytorch/2.1.0/cli/lightning_cli.html - https://pytorch-lightning.medium.com/introducing-lightningcli-v2-supercharge-your-training-c070d43c7dd6 """ from lightning.pytorch.cli import LightningCLI from finetune.classify.eurosat_datamodule import EuroSATDataModule # noqa: F401 from finetune.classify.eurosat_model import EuroSATClassifier # noqa: F401 # %% def cli_main(): """ Command-line inteface to run Clasifier model with EuroSATDataModule. """ cli = LightningCLI(EuroSATClassifier, EuroSATDataModule) return cli # %% if __name__ == "__main__": cli_main() print("Done!")
./finetune/classify/factory.py: import re import torch from torch import nn from src.model import Encoder class Classifier(nn.Module): """ Classifier class uses Clay Encoder for feature extraction and a head for classif

brunosan

brunosan

#/bin/bash

rm -rf a_temp_images
mkdir -p a_temp_images
i=1
for f in $(ls Images_*.png | sort -V ); do
  ln -s "$PWD/$f" "a_temp_images/$(printf "%05d.png" $i)"
  i=$((i + 1))
done

brunosan

All files used to train Clay model v0, using the sampling strategy v2.

The full .txt file lists 1045224 files (and summary at the end), and is 88Mb, so I've split them to fit on a gist in 10mb chunks.

Generated using:

aws s3 ls s3://clay-tiles-02/ --recursive --human-readable --no-paginate --summarize > s3_listing_data_v2_model_v0.txt
split -b 10m s3_listing_data_v2_model_v0.txt -d s3_listing_data_v2_model_v0-part- --additional-suffix=.txt
gh gist create s3_listing_data_v2_model_v0-part-0* -p

brunosan

brunosan

import openai

def recursive_summarization(text, chunk_size=32000, summary_length=20000, final_length=2000):
    openai.api_key = "ADD YOURS"

    print("Starting summarization...")

    if len(text) <= final_length:
        print("Text is short enough. No summarization needed.")
        return text

brunosan

Conceptual explanation of the vanilla Transformer.

Overall, it's a pipeline that takes a sequence of items represented by learnable vectors, and then create parallel attempts ("heads") to understand how each sequence item relates to the others ("self-attention"), to create an output sequence. There can be many layers of heads, making a transformer very deep. Sometimes the task is to purely predict the next item, and sometimes is to create a version of the list, e.g. translate languages, summarize, ...

There are extensions like [[Vision Transformer ViT]] but these are mostly wrappers to convert image patches into digestible tokens for the same architecture.

1. The input (e.g. a text snippet) of length L is tokenized. I.e. every token (word, or subword) assigned the number of that token in the dictionary set. (e.g. {sun}=0,{moon}=1, ...). There are special tokens to indicate things like "start of sequence", "new line", ... Additionally,, all uncommon tokens are replaced with the same extra token.
2. Each

brunosan

“Bruno is one of the most creative and thoughtful scientists that I know. In this work, he examines how to use science and technology to make this world a better place—critically important on a changing and increasingly complex planet.” 
Dr. Ellen Stofan
Former NASA Chief Scientist and
    Director of the Smithsonian National Air and Space Museum
 
“If you liked ‘Freakonomics,’ you will love this one!"
Najat Vallaud-Belkacem
Former France Minister of Education and Research

brunosan

import os
import openai
import requests
import tiktoken
from typing import List
import numpy as np
import matplotlib.pyplot as plt


# Set your OpenAI API key from the environment variable

brunosan