clarifai 10.9.0__tar.gz → 10.9.1__tar.gz

{clarifai-10.9.0/clarifai.egg-info → clarifai-10.9.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: clarifai
-Version: 10.9.0
+Version: 10.9.1
 Summary: Clarifai Python SDK
 Home-page: https://github.com/Clarifai/clarifai-python
 Author: Clarifai
@@ -20,7 +20,7 @@ Classifier: Operating System :: OS Independent
 Requires-Python: >=3.8
 Description-Content-Type: text/markdown
 License-File: LICENSE
-Requires-Dist: clarifai-grpc>=10.8.7
+Requires-Dist: clarifai-grpc>=10.8.8
 Requires-Dist: clarifai-protocol>=0.0.6
 Requires-Dist: numpy>=1.22.0
 Requires-Dist: tqdm>=4.65.0

clarifai-10.9.1/clarifai/__init__.py

@@ -0,0 +1 @@
+__version__ = "10.9.1"

clarifai-10.9.1/clarifai/runners/models/model_run_locally.py

@@ -0,0 +1,194 @@
+import argparse
+import importlib.util
+import inspect
+import os
+import shutil
+import subprocess
+import sys
+import tempfile
+import traceback
+import venv
+
+from clarifai_grpc.grpc.api import resources_pb2, service_pb2
+from clarifai_grpc.grpc.api.status import status_code_pb2, status_pb2
+from clarifai_protocol import BaseRunner
+
+from clarifai.runners.models.model_upload import ModelUploader
+from clarifai.utils.logging import logger
+
+
+class ModelRunLocally:
+
+  def __init__(self, model_path):
+    self.model_path = model_path
+    self.requirements_file = os.path.join(self.model_path, "requirements.txt")
+    self.venv_dir, self.temp_dir = self.create_temp_venv()
+    self.python_executable = os.path.join(self.venv_dir, "bin", "python")
+
+  def create_temp_venv(self):
+    """Create a temporary virtual environment."""
+    logger.info("Creating temporary virtual environment...")
+    temp_dir = tempfile.mkdtemp()
+    venv_dir = os.path.join(temp_dir, "venv")
+    venv.create(venv_dir, with_pip=True)
+
+    logger.info(f"Created temporary virtual environment at {venv_dir}")
+    return venv_dir, temp_dir
+
+  def install_requirements(self):
+    """Install the dependencies from requirements.txt and Clarifai."""
+    pip_executable = os.path.join(self.venv_dir, "bin", "pip")
+    try:
+      logger.info(
+          f"Installing requirements from {self.requirements_file}... in the virtual environment {self.venv_dir}"
+      )
+      subprocess.check_call([pip_executable, "install", "-r", self.requirements_file])
+      logger.info("Installing Clarifai package...")
+      subprocess.check_call([pip_executable, "install", "clarifai"])
+      logger.info("Requirements installed successfully!")
+    except subprocess.CalledProcessError as e:
+      logger.error(f"Error installing requirements: {e}")
+      self.clean_up()
+      sys.exit(1)
+
+  def _get_model_runner(self):
+    """Dynamically import the runner class from the model file."""
+
+    # import the runner class that to be implement by the user
+    runner_path = os.path.join(self.model_path, "1", "model.py")
+
+    # arbitrary name given to the module to be imported
+    module = "runner_module"
+
+    spec = importlib.util.spec_from_file_location(module, runner_path)
+    runner_module = importlib.util.module_from_spec(spec)
+    sys.modules[module] = runner_module
+    spec.loader.exec_module(runner_module)
+
+    # Find all classes in the model.py file that are subclasses of BaseRunner
+    classes = [
+        cls for _, cls in inspect.getmembers(runner_module, inspect.isclass)
+        if issubclass(cls, BaseRunner) and cls.__module__ == runner_module.__name__
+    ]
+
+    #  Ensure there is exactly one subclass of BaseRunner in the model.py file
+    if len(classes) != 1:
+      raise Exception("Expected exactly one subclass of BaseRunner, found: {}".format(
+          len(classes)))
+
+    MyRunner = classes[0]
+    return MyRunner
+
+  def _build_request(self):
+    """Create a mock inference request for testing the model."""
+
+    uploader = ModelUploader(self.model_path)
+    model_version_proto = uploader.get_model_version_proto()
+    model_version_proto.id = "model_version"
+
+    return service_pb2.PostModelOutputsRequest(
+        model=resources_pb2.Model(model_version=model_version_proto),
+        inputs=[
+            resources_pb2.Input(data=resources_pb2.Data(
+                text=resources_pb2.Text(raw="How many people live in new york?"),
+                image=resources_pb2.Image(url="https://samples.clarifai.com/metro-north.jpg"),
+                audio=resources_pb2.Audio(url="https://samples.clarifai.com/GoodMorning.wav"),
+            ))
+        ],
+    )
+
+  def _run_model_inference(self, runner):
+    """Perform inference using the runner."""
+    request = self._build_request()
+
+    try:
+      return runner.predict(request)
+    except Exception as e:
+      logger.error(f"Model Prediction failed: {e}")
+      traceback.print_exc()
+      return service_pb2.MultiOutputResponse(status=status_pb2.Status(
+          code=status_code_pb2.MODEL_PREDICTION_FAILED,
+          description="Prediction failed",
+          details="",
+          internal_details=str(e),
+      ))
+
+  def _run_test(self):
+    """Test the model locally by making a prediction."""
+    # validate that we have checkpoints downloaded before constructing MyRunner
+    uploader = ModelUploader(self.model_path)
+    uploader.download_checkpoints()
+    # construct MyRunner which will call load_model()
+    MyRunner = self._get_model_runner()
+    runner = MyRunner(
+        runner_id="n/a",
+        nodepool_id="n/a",
+        compute_cluster_id="n/a",
+    )
+    runner.load_model()
+
+    # send an inference.
+    response = self._run_model_inference(runner)
+    if response.outputs[0].status.code != status_code_pb2.SUCCESS:
+      logger.error(f"Moddel Prediction failed: {response}")
+    else:
+      logger.info(f"Model Prediction succeeded: {response}")
+
+  def test_model(self):
+    """Test the model by running it locally in the virtual environment."""
+    command = [
+        self.python_executable,
+        "-c",
+        f"import sys; sys.path.append('{os.path.dirname(os.path.abspath(__file__))}'); "
+        f"from model_run_locally import ModelRunLocally; ModelRunLocally('{self.model_path}')._run_test()",
+    ]
+    try:
+      logger.info("Testing the model locally...")
+      subprocess.check_call(command)
+      logger.info("Model tested successfully!")
+    except subprocess.CalledProcessError as e:
+      logger.error(f"Error testing the model: {e}")
+      sys.exit(1)
+
+  # run the model server
+  def run_model_server(self):
+    """Run the Clarifai Runners's model server."""
+
+    command = [
+        self.python_executable, "-m", "clarifai.runners.server", "--model_path", self.model_path,
+        "--start_dev_server"
+    ]
+    try:
+      logger.info(f"Starting model server with model at {self.model_path}...")
+      subprocess.check_call(command)
+      logger.info("Model server started successfully!")
+    except subprocess.CalledProcessError as e:
+      logger.error(f"Error running model server: {e}")
+      self.clean_up()
+      sys.exit(1)
+
+  def clean_up(self):
+    """Clean up the temporary virtual environment."""
+    if os.path.exists(self.temp_dir):
+      logger.info("Cleaning up temporary virtual environment...")
+      shutil.rmtree(self.temp_dir)
+
+
+def main():
+  parser = argparse.ArgumentParser()
+  parser.add_argument(
+      '--model_path', type=str, required=True, help='Path of the model folder to upload')
+  args = parser.parse_args()
+
+  model_path = args.model_path
+  manager = ModelRunLocally(model_path)
+
+  try:
+    manager.install_requirements()
+    manager.test_model()
+  finally:
+    manager.clean_up()
+
+
+if __name__ == "__main__":
+  main()

{clarifai-10.9.0 → clarifai-10.9.1}/clarifai/runners/models/model_upload.py

@@ -10,7 +10,7 @@ from google.protobuf import json_format
 from rich import print
 
 from clarifai.client import BaseClient
-from clarifai.runners.utils.loader import HuggingFaceLoarder
+from clarifai.runners.utils.loader import HuggingFaceLoader
 from clarifai.urls.helper import ClarifaiUrlHelper
 from clarifai.utils.logging import logger
 
@@ -59,6 +59,27 @@ class ModelUploader:
       config = yaml.safe_load(file)
     return config
 
+  def _validate_config_checkpoints(self):
+    if not self.config.get("checkpoints"):
+      logger.info("No checkpoints specified in the config file")
+      return None
+
+    assert "type" in self.config.get("checkpoints"), "No loader type specified in the config file"
+    loader_type = self.config.get("checkpoints").get("type")
+    if not loader_type:
+      logger.info("No loader type specified in the config file for checkpoints")
+    assert loader_type == "huggingface", "Only huggingface loader supported for now"
+    if loader_type == "huggingface":
+      assert "repo_id" in self.config.get("checkpoints"), "No repo_id specified in the config file"
+      repo_id = self.config.get("checkpoints").get("repo_id")
+
+      # prefer env var for HF_TOKEN but if not provided then use the one from config.yaml if any.
+      if 'HF_TOKEN' in os.environ:
+        hf_token = os.environ['HF_TOKEN']
+      else:
+        hf_token = self.config.get("checkpoints").get("hf_token", None)
+      return repo_id, hf_token
+
   @property
   def client(self):
     if self._client is None:
@@ -180,27 +201,9 @@ class ModelUploader:
     return f"{self.folder}.tar.gz"
 
   def download_checkpoints(self):
-    if not self.config.get("checkpoints"):
-      logger.info("No checkpoints specified in the config file")
-      return
-
-    assert "type" in self.config.get("checkpoints"), "No loader type specified in the config file"
-    loader_type = self.config.get("checkpoints").get("type")
-    if not loader_type:
-      logger.info("No loader type specified in the config file for checkpoints")
-    assert loader_type == "huggingface", "Only huggingface loader supported for now"
-    if loader_type == "huggingface":
-      assert "repo_id" in self.config.get("checkpoints"), "No repo_id specified in the config file"
-      repo_id = self.config.get("checkpoints").get("repo_id")
-
-      # prefer env var for HF_TOKEN but if not provided then use the one from config.yaml if any.
-      if 'HF_TOKEN' in os.environ:
-        hf_token = os.environ['HF_TOKEN']
-      else:
-        hf_token = self.config.get("checkpoints").get("hf_token", None)
-        assert hf_token != 'hf_token', "The default 'hf_token' is not valid. Please provide a valid token or leave that field out of config.yaml if not needed."
-      loader = HuggingFaceLoarder(repo_id=repo_id, token=hf_token)
-
+    repo_id, hf_token = self._validate_config_checkpoints()
+    if repo_id and hf_token:
+      loader = HuggingFaceLoader(repo_id=repo_id, token=hf_token)
       success = loader.download_checkpoints(self.checkpoint_path)
 
       if not success:
@@ -232,7 +235,7 @@ class ModelUploader:
     concepts = config.get('concepts')
     logger.info(f"Updated config.yaml with {len(concepts)} concepts.")
 
-  def _get_model_version_proto(self):
+  def get_model_version_proto(self):
 
     model_version_proto = resources_pb2.ModelVersion(
         pretrained_model_config=resources_pb2.PretrainedModelConfig(),
@@ -242,8 +245,7 @@ class ModelUploader:
     model_type_id = self.config.get('model').get('model_type_id')
     if model_type_id in self.CONCEPTS_REQUIRED_MODEL_TYPE:
 
-      loader = HuggingFaceLoarder()
-      labels = loader.fetch_labels(self.checkpoint_path)
+      labels = HuggingFaceLoader.fetch_labels(self.checkpoint_path)
       # sort the concepts by id and then update the config file
       labels = sorted(labels.items(), key=lambda x: int(x[0]))
 
@@ -258,6 +260,21 @@ class ModelUploader:
     file_path = f"{self.folder}.tar.gz"
     logger.info(f"Will tar it into file: {file_path}")
 
+    model_type_id = self.config.get('model').get('model_type_id')
+    repo_id, hf_token = self._validate_config_checkpoints()
+
+    loader = HuggingFaceLoader(repo_id=repo_id, token=hf_token)
+
+    if not download_checkpoints and not loader.validate_download(self.checkpoint_path) and (
+        model_type_id in self.CONCEPTS_REQUIRED_MODEL_TYPE) and 'concepts' not in self.config:
+      logger.error(
+          f"Model type {model_type_id} requires concepts to be specified in the config file or download the model checkpoints to infer the concepts."
+      )
+      input("Press Enter to download the checkpoints to infer the concepts and continue...")
+      self.download_checkpoints()
+
+    model_version_proto = self.get_model_version_proto()
+
     if download_checkpoints:
       tar_cmd = f"tar --exclude=*~ -czvf {self.tar_file} -C {self.folder} ."
     else:  # we don't want to send the checkpoints up even if they are in the folder.
@@ -268,8 +285,6 @@ class ModelUploader:
     os.system(tar_cmd)
     logger.info("Tarring complete, about to start upload.")
 
-    model_version_proto = self._get_model_version_proto()
-
     file_size = os.path.getsize(self.tar_file)
     logger.info(f"Size of the tar is: {file_size} bytes")
 

{clarifai-10.9.0 → clarifai-10.9.1}/clarifai/runners/utils/loader.py

@@ -6,7 +6,9 @@ import subprocess
 from clarifai.utils.logging import logger
 
 
-class HuggingFaceLoarder:
+class HuggingFaceLoader:
+
+  HF_DOWNLOAD_TEXT = "The 'huggingface_hub' package is not installed. Please install it using 'pip install huggingface_hub'."
 
   def __init__(self, repo_id=None, token=None):
     self.repo_id = repo_id
@@ -14,9 +16,7 @@ class HuggingFaceLoarder:
     if token:
       try:
         if importlib.util.find_spec("huggingface_hub") is None:
-          raise ImportError(
-              "The 'huggingface_hub' package is not installed. Please install it using 'pip install huggingface_hub'."
-          )
+          raise ImportError(self.HF_DOWNLOAD_TEXT)
         os.environ['HF_TOKEN'] = token
         subprocess.run(f'huggingface-cli login --token={os.environ["HF_TOKEN"]}', shell=True)
       except Exception as e:
@@ -27,9 +27,7 @@ class HuggingFaceLoarder:
     try:
       from huggingface_hub import snapshot_download
     except ImportError:
-      raise ImportError(
-          "The 'huggingface_hub' package is not installed. Please install it using 'pip install huggingface_hub'."
-      )
+      raise ImportError(self.HF_DOWNLOAD_TEXT)
     if os.path.exists(checkpoint_path) and self.validate_download(checkpoint_path):
       logger.info("Checkpoints already exist")
       return True
@@ -54,20 +52,26 @@ class HuggingFaceLoarder:
 
   def validate_hf_model(self,):
     # check if model exists on HF
-
-    from huggingface_hub import file_exists, repo_exists
+    try:
+      from huggingface_hub import file_exists, repo_exists
+    except ImportError:
+      raise ImportError(self.HF_DOWNLOAD_TEXT)
     return repo_exists(self.repo_id) and file_exists(self.repo_id, 'config.json')
 
   def validate_download(self, checkpoint_path: str):
     # check if model exists on HF
-    from huggingface_hub import list_repo_files
+    try:
+      from huggingface_hub import list_repo_files
+    except ImportError:
+      raise ImportError(self.HF_DOWNLOAD_TEXT)
     checkpoint_dir_files = [
         f for dp, dn, fn in os.walk(os.path.expanduser(checkpoint_path)) for f in fn
     ]
     return (len(checkpoint_dir_files) >= len(list_repo_files(self.repo_id))) and len(
         list_repo_files(self.repo_id)) > 0
 
-  def fetch_labels(self, checkpoint_path: str):
+  @staticmethod
+  def fetch_labels(checkpoint_path: str):
     # Fetch labels for classification, detection and segmentation models
     config_path = os.path.join(checkpoint_path, 'config.json')
     with open(config_path, 'r') as f:

{clarifai-10.9.0 → clarifai-10.9.1/clarifai.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: clarifai
-Version: 10.9.0
+Version: 10.9.1
 Summary: Clarifai Python SDK
 Home-page: https://github.com/Clarifai/clarifai-python
 Author: Clarifai
@@ -20,7 +20,7 @@ Classifier: Operating System :: OS Independent
 Requires-Python: >=3.8
 Description-Content-Type: text/markdown
 License-File: LICENSE
-Requires-Dist: clarifai-grpc>=10.8.7
+Requires-Dist: clarifai-grpc>=10.8.8
 Requires-Dist: clarifai-protocol>=0.0.6
 Requires-Dist: numpy>=1.22.0
 Requires-Dist: tqdm>=4.65.0

{clarifai-10.9.0 → clarifai-10.9.1}/clarifai.egg-info/SOURCES.txt

@@ -72,6 +72,7 @@ clarifai/runners/dockerfile_template/Dockerfile.cuda.template
 clarifai/runners/models/__init__.py
 clarifai/runners/models/base_typed_model.py
 clarifai/runners/models/model_class.py
+clarifai/runners/models/model_run_locally.py
 clarifai/runners/models/model_runner.py
 clarifai/runners/models/model_servicer.py
 clarifai/runners/models/model_upload.py

{clarifai-10.9.0 → clarifai-10.9.1}/clarifai.egg-info/requires.txt

@@ -1,4 +1,4 @@
-clarifai-grpc>=10.8.7
+clarifai-grpc>=10.8.8
 clarifai-protocol>=0.0.6
 numpy>=1.22.0
 tqdm>=4.65.0

{clarifai-10.9.0 → clarifai-10.9.1}/requirements.txt

@@ -1,4 +1,4 @@
-clarifai-grpc>=10.8.7
+clarifai-grpc>=10.8.8
 clarifai-protocol>=0.0.6
 numpy>=1.22.0
 tqdm>=4.65.0

clarifai-10.9.0/clarifai/__init__.py

@@ -1 +0,0 @@
-__version__ = "10.9.0"