Getting an FileNotFoundError: [Errno 2] No such file or directory: 'inference.py'

[Nuwan1654]

Describe the bug Probably this is not a Bug, but when I try to deploy the Sagemaker Pytorch model, I am getting a FileNotFoundError: [Errno 2] No such file or directory: 'inference.py' error

To reproduce

1. Create 'my_model.tar.gz' file according to the instructions on this page
2. upload to s3 My folder structure as follow

   
   .
   ├── best.pt
   ├── coco.yaml
   └── code
   ├── common.py
   ├── experimental.py
   ├── general.py
   ├── inference.py
   ├── loss.py
   ├── requirements.txt
   ├── torch_utils.py
   └── yolo.py

4. Run `predictor = pytorch_model.deploy(instance_type='ml.c4.xlarge', initial_instance_count=1)`

**Expected behavior**
successful deploy

**Screenshots or logs**

---

FileNotFoundError Traceback (most recent call last) Cell In[14], line 1 ----> 1 predictor = pytorch_model.deploy(instance_type='ml.c4.xlarge', initial_instance_count=1)

File ~/.local/lib/python3.8/site-packages/sagemaker/model.py:1260, in Model.deploy(self, initial_instance_count, instance_type, serializer, deserializer, accelerator_type, endpoint_name, tags, kms_key, wait, data_capture_config, async_inference_config, serverless_inference_config, volume_size, model_data_download_timeout, container_startup_health_check_timeout, inference_recommendation_id, explainer_config, **kwargs) 1257 if self._base_name is not None: 1258 self._base_name = "-".join((self._base_name, compiled_model_suffix)) -> 1260 self._create_sagemaker_model( 1261 instance_type, accelerator_type, tags, serverless_inference_config 1262 ) 1264 serverless_inference_config_dict = ( 1265 serverless_inference_config._to_request_dict() if is_serverless else None 1266 ) 1267 production_variant = sagemaker.production_variant( 1268 self.name, 1269 instance_type, (...) 1275 container_startup_health_check_timeout=container_startup_health_check_timeout, 1276 )

File ~/.local/lib/python3.8/site-packages/sagemaker/model.py:693, in Model._create_sagemaker_model(self, instance_type, accelerator_type, tags, serverless_inference_config) 671 def _create_sagemaker_model( 672 self, instance_type=None, accelerator_type=None, tags=None, serverless_inference_config=None 673 ): 674 """Create a SageMaker Model Entity 675 676 Args: (...) 691 not provided in serverless inference. So this is used to find image URIs. 692 """ --> 693 container_def = self.prepare_container_def( 694 instance_type, 695 accelerator_type=accelerator_type, 696 serverless_inference_config=serverless_inference_config, 697 ) 699 if not isinstance(self.sagemaker_session, PipelineSession): 700 # _base_name, model_name are not needed under PipelineSession. 701 # the model_data may be Pipeline variable 702 # which may break the _base_name generation 703 self._ensure_base_name_if_needed( 704 image_uri=container_def["Image"], 705 script_uri=self.source_dir, 706 model_uri=self.model_data, 707 )

File ~/.local/lib/python3.8/site-packages/sagemaker/pytorch/model.py:298, in PyTorchModel.prepare_container_def(self, instance_type, accelerator_type, serverless_inference_config) 290 deploy_image = self.serving_image_uri( 291 region_name, 292 instance_type, 293 accelerator_type=accelerator_type, 294 serverless_inference_config=serverless_inference_config, 295 ) 297 deploy_key_prefix = model_code_key_prefix(self.key_prefix, self.name, deploy_image) --> 298 self._upload_code(deploy_key_prefix, repack=self._is_mms_version()) 299 deploy_env = dict(self.env) 300 deploy_env.update(self._script_mode_env_vars())

File ~/.local/lib/python3.8/site-packages/sagemaker/model.py:626, in Model._upload_code(self, key_prefix, repack) 611 self.uploaded_code = fw_utils.UploadedCode( 612 s3_prefix=repacked_model_data, script_name=os.path.basename(self.entry_point) 613 ) 615 LOGGER.info( 616 "Repacking model artifact (%s), script artifact " 617 "(%s), and dependencies (%s) " (...) 623 repacked_model_data, 624 ) --> 626 utils.repack_model( 627 inference_script=self.entry_point, 628 source_directory=self.source_dir, 629 dependencies=self.dependencies, 630 model_uri=self.model_data, 631 repacked_model_uri=repacked_model_data, 632 sagemaker_session=self.sagemaker_session, 633 kms_key=self.model_kms_key, 634 ) 636 self.repacked_model_data = repacked_model_data

File ~/.local/lib/python3.8/site-packages/sagemaker/utils.py:516, in repack_model(inference_script, source_directory, dependencies, model_uri, repacked_model_uri, sagemaker_session, kms_key) 513 with _tmpdir(directory=local_download_dir) as tmp: 514 model_dir = _extract_model(model_uri, sagemaker_session, tmp) --> 516 _create_or_update_code_dir( 517 model_dir, 518 inference_script, 519 source_directory, 520 dependencies, 521 sagemaker_session, 522 tmp, 523 ) 525 tmp_model_path = os.path.join(tmp, "temp-model.tar.gz") 526 with tarfile.open(tmp_model_path, mode="w:gz") as t:

File ~/.local/lib/python3.8/site-packages/sagemaker/utils.py:577, in _create_or_update_code_dir(model_dir, inference_script, source_directory, dependencies, sagemaker_session, tmp) 575 os.mkdir(code_dir) 576 try: --> 577 shutil.copy2(inference_script, code_dir) 578 except FileNotFoundError: 579 if os.path.exists(os.path.join(code_dir, inference_script)):

File /usr/lib/python3.8/shutil.py:435, in copy2(src, dst, follow_symlinks) 433 if os.path.isdir(dst): 434 dst = os.path.join(dst, os.path.basename(src)) --> 435 copyfile(src, dst, follow_symlinks=follow_symlinks) 436 copystat(src, dst, follow_symlinks=follow_symlinks) 437 return dst

File /usr/lib/python3.8/shutil.py:264, in copyfile(src, dst, follow_symlinks) 262 os.symlink(os.readlink(src), dst) 263 else: --> 264 with open(src, 'rb') as fsrc, open(dst, 'wb') as fdst: 265 # macOS 266 if _HAS_FCOPYFILE: 267 try:

FileNotFoundError: [Errno 2] No such file or directory: 'inference.py' '''

System information A description of your system. Please provide:

• SageMaker Python SDK version: 2.155.0
• Framework name:: PyTorch
• Framework version: 2.0.1+cu117
• Python version: 3.8.10
• CPU or GPU: GPU
• Custom Docker image (Y/N): Y

Any help on this would be highly appreciated.

[hiyamgh]

Facing the same issue!

[hooNpk]

Is there anyone who solve this problem? I have a same issue

[kenny-chen]

I conjecture that you are compiling segamaker using a local environment. The reason for the issue is the absence of inference.py.

When executing pytorch_model.deploy(), it will initiate the Docker server image. Since I am using TensorFlow, it will launch the server image 763104351884.dkr.ecr.us-east-1.amazonaws.com/tensorflow-inference:2.11-gpu.

However, this image lacks "inference.py", resulting in a FileNotFoundError: [Errno 2] No such file or directory: 'inference.py' error.

Solution:

Firstly, create ./code/inference.py. Refer to the following link for the code: https://github.com/aws/sagemaker-tensorflow-serving-container/blob/master/test/resources/examples/test1/inference.py

import json
from collections import namedtuple

Context = namedtuple('Context',
                     'model_name, model_version, method, rest_uri, grpc_uri, '
                     'custom_attributes, request_content_type, accept_header')

def input_handler(data, context):
    if context.request_content_type == 'application/json':
        d = data.read().decode('utf-8')
        return d if len(d) else ''

    if context.request_content_type == 'text/csv':
        return json.dumps({
            'instances': [float(x) for x in data.read().decode('utf-8').split(',')]
        })

    raise ValueError('{{"error": "unsupported content type {}"}}'.format(
        context.request_content_type or "unknown"))

def output_handler(data, context):
    if data.status_code != 200:
        raise ValueError(data.content.decode('utf-8'))

    response_content_type = context.accept_header
    prediction = data.content
    return prediction, response_content_type

Next, deploy the model using the inference approach. The steps below demonstrate how to locally deploy the model using TensorFlow:

import os
from sagemaker.tensorflow import TensorFlowModel

model = TensorFlowModel(
    entry_point='inference.py',
    source_dir='./code',
    role=os.environ['AWS_ROLE'],
    model_data=f'{output}/model.tar.gz',
    framework_version='2.11'
)

predictor = model.deploy(
    initial_instance_count=1,
    instance_type='local_gpu',
)

1. The ./code/inference.py will be automatically loaded into the server image.
2. {output}/model.tar.gz will be automatically loaded as the local model.
3. The model.deploy() command will successfully initiate the server image.

[nehyaeeg3]

whats the solution?

[kitty2121]

I think i've figured it out ; you MUST create the app.py file WITHIN here :

file browser pane, browse to "./lab1/packages/{account_id}-lab1_code-1.0/src/

Some tutorials say here: "In the file browser pane, browse to ./lab1/packages/{account_id}-lab1_code-1.0/. Find descriptor.json." BUT the right tutorial (https://catalog.workshops.aws/panorama-immersion-day/en-US/20-lab1-object-detection/21-lab1) correctly states to create and save the app.py txt file in ""./lab1/packages/{account_id}-lab1_code-1.0/src/" i.e. the SRC folder found in Lab 1.

Wrong version - https://explore.skillbuilder.aws/learn/course/17780/play/93251/aws-panorama-building-edge-computer-vision-cv-applications Right version - https://catalog.workshops.aws/panorama-immersion-day/en-US/20-lab1-object-detection/21-lab1

[evankozliner]

Also facing this issue. I think the docs on https://sagemaker.readthedocs.io/en/stable/frameworks/pytorch/using_pytorch.html#bring-your-own-model may be faulty as well?

[Itto1992]

I also encountered the same problem. I tried from notebook instance. Here is my implementation:

import sagemaker
from sagemaker.pytorch import PyTorchModel
from sagemaker.serverless import ServerlessInferenceConfig

sagemaker_session = sagemaker.Session()
role = sagemaker.get_execution_role()

model = PyTorchModel(
    entry_point='inference.py',
    role=role,
    model_data='s3://***/model.tar.gz',
    framework_version='2.1',
    py_version='py310',
)

serverless_config = ServerlessInferenceConfig(
    max_concurrency=1,
    memory_size_in_mb=3072,
)

deploy_params = {
    'instance_type': 'ml.t3.medium',
    'initial_instance_count': 1,
    'serverless_inference_config': serverless_config,
}

predictor = model.deploy(**deploy_params)

[Itto1992]

I solved this problem by changing how to create tar.gz file. The key point is to use tar command in the same directory as the model file like:

$ tar czvf ../model.tar.gz *
code/
code/requirements.txt
code/inference.py
model.pth

I failed when I run tar command in the parent directory of the model file:

$tar czvf model.tar.gz model
model/
model/model.pth
model/code/
model/code/requirements.txt
model/code/inference.py

As you can see, these results are different each other.