# Models is not saved in S3 bucket location **URL:** **Author:** ![amtha24](https://avatars.discourse-cdn.com/v4/letter/a/a6a055/32.png) [@amtha24](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Fu%2Famtha24)%5C **Post date:** [April 9, 2024, 9:46am UTC](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Fmodels-is-not-saved-in-s3-bucket-location%2F80861%2F1 "2024-04-09T09:46:35Z") import time import json from sagemaker.huggingface import HuggingFace # define Training Job Name job\_name = f’huggingface-donut-{time.strftime(“%Y-%m-%d-%H-%M-%S”, time.localtime())}’ # stingify special tokens special\_tokens = ‘,’.join(map(str, processor.tokenizer.special\_tokens\_map\_extended[“additional\_special\_tokens”])) training\_input\_path = f’s3://processed/donut-sagemaker/train’ # hyperparameters, which are passed into the training job hyperparameters = { ‘model\_id’: model\_id, # pre-trained model ‘special\_tokens’: json.dumps(special\_tokens), # special tokens which will be added to the tokenizer ‘dataset\_path’: training\_input\_path, # path where sagemaker will save training dataset ‘epochs’: 1, # number of training epochs ‘per\_device\_train\_batch\_size’: 8, # batch size for training ‘gradient\_checkpointing’: True, # batch size for training ‘lr’: 4e-5, # learning rate used during training } # create the Estimator huggingface\_estimator = HuggingFace( entry\_point = ‘train.py’, # train script source\_dir = ‘./scripts’, # directory which includes all the files needed for training instance\_type = ‘ml.p3.2xlarge’, # instances type used for the training job output\_path = “s3://processed/donut-sagemaker/path/”, code\_location = “s3://processed/donut-sagemaker/code/”, instance\_count = 1, # the number of instances used for training base\_job\_name = job\_name, # the name of the training job role = role, # Iam role used in training job to access AWS ressources, e.g. S3 volume\_size = 100, # the size of the EBS volume in GB transformers\_version = ‘4.26’, # the transformers version used in the training job pytorch\_version = ‘1.13’, # the pytorch\_version version used in the training job py\_version = ‘py39’, # the python version used in the training job hyperparameters = hyperparameters ) # define a data input dictonary with our uploaded s3 uris data = {‘training’: training\_input\_path} # starting the train job with our uploaded datasets as input huggingface\_estimator.fit(data, wait=True) ``` model.save_pretrained("s3://processed/donut-sagemaker/model/") processor.save_pretrained("s3://processed/donut-sagemaker/model/") ``` I am trying run this training on sagemaker pipeline , the training is success but my models is not saved in that location can anyone help me to resolve it?