Save and load Keras models to and from AWS S3

keras_model_s3_wrapper.py

import s3fs
import zipfile
import tempfile
import numpy as np
from tensorflow import keras
from pathlib import Path
import logging


AWS_ACCESS_KEY="aws_access_key"
AWS_SECRET_KEY="aws_secret_key"
BUCKET_NAME="bucket_name"


def get_s3fs():
  return s3fs.S3FileSystem(key=AWS_ACCESS_KEY, secret=AWS_SECRET_KEY)


def zipdir(path, ziph):
  # Zipfile hook to zip up model folders
  length = len(path) # Doing this to get rid of parent folders
  for root, dirs, files in os.walk(path):
    folder = root[length:] # We don't need parent folders! Why in the world does zipfile zip the whole tree??
    for file in files:
      ziph.write(os.path.join(root, file), os.path.join(folder, file))

            
def s3_save_keras_model(model, model_name):
  with tempfile.TemporaryDirectory() as tempdir:
    model.save(f"{tempdir}/{model_name}")
    # Zip it up first
    zipf = zipfile.ZipFile(f"{tempdir}/{model_name}.zip", "w", zipfile.ZIP_STORED)
    zipdir(f"{tempdir}/{model_name}", zipf)
    zipf.close()
    s3fs = get_s3fs()
    s3fs.put(f"{tempdir}/{model_name}.zip", f"{BUCKET_NAME}/{model_name}.zip")
    logging.info(f"Saved zipped model at path s3://{BUCKET_NAME}/{model_name}.zip")
 

def s3_get_keras_model(model_name: str) -> keras.Model:
  with tempfile.TemporaryDirectory() as tempdir:
    s3fs = get_s3fs()
    # Fetch and save the zip file to the temporary directory
    s3fs.get(f"{BUCKET_NAME}/{model_name}.zip", f"{tempdir}/{model_name}.zip")
    # Extract the model zip file within the temporary directory
    with zipfile.ZipFile(f"{tempdir}/{model_name}.zip") as zip_ref:
        zip_ref.extractall(f"{tempdir}/{model_name}")
    # Load the keras model from the temporary directory
    return keras.models.load_model(f"{tempdir}/{model_name}")
  


inputs = keras.Input(shape=(32,))
outputs = keras.layers.Dense(1)(inputs)
model = keras.Model(inputs, outputs)
model.compile(optimizer="adam", loss="mean_squared_error")

# Save the model to S3
s3_save_keras_model(model, "my_model") 

# Load the model from S3
loaded_model = s3_get_keras_model("my_model")

how you able to load tensorflow and s3fs in lambda layer because size will be higher than 250 mb ? i think tensorflow 10 inclused the tf.keras and its size is big.
can you help me here i am little lost .

So I didn't do this on a lambda function, this was a plain Python function that I ran on my machine. Yes, loading tf in lambdas is a pain, but you could try using Lambda containers.

When I am implementing this I am getting the following error :
"SavedModel file does not exist at: /var/folders/cb/ns18k3051f35p2jr32r2t4vr0000gp/T/tmpbv0lstar/yolo_v3/{saved_model.pbtxt|saved_model.pb}"
Any ideas on what can be done?

@rahuja23, This seems to be a keras error based on yolo, which you seem to be using. I'd say try to print out the files at the temp folder path. You could also maybe get some help from this question: https://stackoverflow.com/questions/60071818/savedmodel-file-does-not-exist-at-model-h5-saved-model-pbtxtsaved-model-pb

@rahuja23
I countered same problem too. Zip file extract doesn't work.
Have you figure it out?

@joyspark What's the error you encountered? Was it the same one as @rahuja23 ?

@ramdesh Yes, same one as @rahuja23 but I just give up and find other way. Thank you.

@ramdesh this worked for saving/loading my keras SavedModel from S3 when everything I tried before wasn't working. Thanks so much!

This worked perfectly for me. Thanks