app.py

import os
from fastapi import FastAPI
import subprocess
import wandb
from huggingface_hub import HfApi

TOKEN = os.environ.get("DATACOMP_TOKEN")
API = HfApi(token=TOKEN)
wandb_api_key = os.environ.get('wandb_api_key')
wandb.login(key=wandb_api_key)

random_num = 10.0
subset = 'frac-1over32'
experiment_name = f"ImageNetTraining{random_num}-{subset}"
experiment_repo = f"datacomp/{experiment_name}"
app = FastAPI()

@app.get("/")
def start_train():
    os.system("echo '#### pwd'")
    os.system("pwd")
    os.system("echo '#### ls'")
    os.system("ls")
    # Create a place to put the output.
    os.system("echo 'Creating results output repository in case it does not exist yet...'")
    try:
        API.create_repo(repo_id=f"{experiment_repo}", repo_type="dataset",)
        os.system(f"echo 'Created results output repository {experiment_repo}'")
    except:
        os.system("echo 'Already there; skipping.'")
        pass
    os.system("echo 'Beginning processing.'")
    # Handles CUDA OOM errors.
    os.system(f"export PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True")
    os.system("echo 'Okay, trying training.'")
    os.system(f"cd pytorch-image-models; ./train.sh 4 --dataset hfds/datacomp/imagenet-1k-random-{random_num}-{subset} --log-wandb --experiment {experiment_name} --model seresnet34 --sched cosine --epochs 150 --warmup-epochs 5 --lr 0.4 --reprob 0.5 --remode pixel --batch-size 256 --amp -j 4")
    os.system("echo 'Done'.")
    os.system("ls")
    # Upload output to repository
    os.system("echo 'trying to upload...'")
    API.upload_folder(folder_path="/app", repo_id=f"{experiment_repo}", repo_type="dataset",)
    API.pause_space(experiment_repo)
    return {"Completed": "!"}