Spaces:
Runtime error
Runtime error
File size: 1,886 Bytes
cef9e84 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 |
import os
import boto3
from pytorch_lightning.callbacks import ModelCheckpoint
import glob
class S3ModelCheckpoint(ModelCheckpoint):
def __init__(self, bucket_name, s3_folder, *args, **kwargs):
super().__init__(*args, **kwargs)
self.bucket_name = bucket_name
self.s3_folder = s3_folder
if self.bucket_name is not None:
self.s3_client = boto3.client('s3')
self.last_checkpoint_path = None
def on_save_checkpoint(self, trainer, pl_module, checkpoint):
# Save the checkpoint locally as usual
filepath = self.last_model_path
super().on_save_checkpoint(trainer, pl_module, checkpoint)
if pl_module.global_rank == 0 and self.bucket_name is not None:
list_of_files = glob.glob(os.path.join(self.dirpath, '*.ckpt')) # Get list of all .ckpt files
if list_of_files:
filepath = max(list_of_files, key=os.path.getctime) # Find the most recently created file
# If this isn't the last checkpoint or it's not one of the top_k, upload and delete
if filepath != self.last_checkpoint_path and not self._is_last_checkpoint(filepath):
self.upload_to_s3(filepath)
os.remove(filepath)
else:
# Update the path of the last saved checkpoint
self.last_checkpoint_path = filepath
def upload_to_s3(self, filepath):
s3_path = os.path.join(self.s3_folder, filepath)
self.s3_client.upload_file(filepath, self.bucket_name, s3_path)
print(f"[INFO] Model checkpoint uploaded to {self.bucket_name}/{s3_path}")
def _is_last_checkpoint(self, filepath):
# Determine if this is the last checkpoint based on file naming
# Assuming your last checkpoint is named like 'last.ckpt'
return 'last' in filepath |