clarifai 11.8.4__tar.gz → 11.9.0__tar.gz

{clarifai-11.8.4/clarifai.egg-info → clarifai-11.9.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: clarifai
-Version: 11.8.4
+Version: 11.9.0
 Home-page: https://github.com/Clarifai/clarifai-python
 Author: Clarifai
 Author-email: support@clarifai.com
@@ -19,8 +19,8 @@ Classifier: Operating System :: OS Independent
 Requires-Python: >=3.8
 Description-Content-Type: text/markdown
 License-File: LICENSE
-Requires-Dist: clarifai-grpc>=11.8.2
-Requires-Dist: clarifai-protocol>=0.0.32
+Requires-Dist: clarifai-grpc>=11.9.8
+Requires-Dist: clarifai-protocol>=0.0.33
 Requires-Dist: numpy>=1.22.0
 Requires-Dist: tqdm>=4.65.0
 Requires-Dist: PyYAML>=6.0.1

clarifai-11.9.0/clarifai/__init__.py

@@ -0,0 +1 @@
+__version__ = "11.9.0"

{clarifai-11.8.4 → clarifai-11.9.0}/clarifai/cli/model.py

@@ -29,6 +29,7 @@ from clarifai.utils.constants import (
     DEFAULT_LOCAL_RUNNER_NODEPOOL_ID,
     DEFAULT_OLLAMA_MODEL_REPO_BRANCH,
     DEFAULT_PYTHON_MODEL_REPO_BRANCH,
+    DEFAULT_SGLANG_MODEL_REPO_BRANCH,
     DEFAULT_TOOLKIT_MODEL_REPO,
     DEFAULT_VLLM_MODEL_REPO_BRANCH,
 )
@@ -76,15 +77,15 @@ def model():
 @click.option(
     '--toolkit',
     type=click.Choice(
-        ['ollama', 'huggingface', 'lmstudio', 'vllm', 'python'], case_sensitive=False
+        ['ollama', 'huggingface', 'lmstudio', 'vllm', 'sglang', 'python'], case_sensitive=False
     ),
     required=False,
-    help='Toolkit to use for model initialization. Currently supports "ollama", "huggingface", "lmstudio", "vllm" and "python".',
+    help='Toolkit to use for model initialization. Currently supports "ollama", "huggingface", "lmstudio", "vllm", "sglang" and "python".',
 )
 @click.option(
     '--model-name',
     required=False,
-    help='Model name to configure when using --toolkit. For ollama toolkit, this sets the Ollama model to use (e.g., "llama3.1", "mistral", etc.). For vllm & huggingface toolkit, this sets the Hugging Face model repo_id (e.g., "unsloth/Llama-3.2-1B-Instruct").\n For lmstudio toolkit, this sets the LM Studio model name (e.g., "qwen/qwen3-4b-thinking-2507").\n',
+    help='Model name to configure when using --toolkit. For ollama toolkit, this sets the Ollama model to use (e.g., "llama3.1", "mistral", etc.). For vllm, sglang & huggingface toolkit, this sets the Hugging Face model repo_id (e.g., "unsloth/Llama-3.2-1B-Instruct").\n For lmstudio toolkit, this sets the LM Studio model name (e.g., "qwen/qwen3-4b-thinking-2507").\n',
 )
 @click.option(
     '--port',
@@ -129,8 +130,8 @@ def init(
     MODEL_TYPE_ID: Type of model to create. If not specified, defaults to "text-to-text" for text models.\n
     GITHUB_PAT: GitHub Personal Access Token for authentication when cloning private repositories.\n
     GITHUB_URL: GitHub repository URL or "repo" format to clone a repository from. If provided, the entire repository contents will be copied to the target directory instead of using default templates.\n
-    TOOLKIT: Toolkit to use for model initialization. Currently supports "ollama", "huggingface", "lmstudio", "vllm" and "python".\n
-    MODEL_NAME: Model name to configure when using --toolkit. For ollama toolkit, this sets the Ollama model to use (e.g., "llama3.1", "mistral", etc.). For vllm & huggingface toolkit, this sets the Hugging Face model repo_id (e.g., "Qwen/Qwen3-4B-Instruct-2507"). For lmstudio toolkit, this sets the LM Studio model name (e.g., "qwen/qwen3-4b-thinking-2507").\n
+    TOOLKIT: Toolkit to use for model initialization. Currently supports "ollama", "huggingface", "lmstudio", "vllm", "sglang" and "python".\n
+    MODEL_NAME: Model name to configure when using --toolkit. For ollama toolkit, this sets the Ollama model to use (e.g., "llama3.1", "mistral", etc.). For vllm, sglang & huggingface toolkit, this sets the Hugging Face model repo_id (e.g., "Qwen/Qwen3-4B-Instruct-2507"). For lmstudio toolkit, this sets the LM Studio model name (e.g., "qwen/qwen3-4b-thinking-2507").\n
     PORT: Port to run the (Ollama/lmstudio) server on. Defaults to 23333.\n
     CONTEXT_LENGTH: Context length for the (Ollama/lmstudio) model. Defaults to 8192.\n
     """
@@ -183,6 +184,9 @@ def init(
     elif toolkit == 'vllm':
         github_url = DEFAULT_TOOLKIT_MODEL_REPO
         branch = DEFAULT_VLLM_MODEL_REPO_BRANCH
+    elif toolkit == 'sglang':
+        github_url = DEFAULT_TOOLKIT_MODEL_REPO
+        branch = DEFAULT_SGLANG_MODEL_REPO_BRANCH
     elif toolkit == 'python':
         github_url = DEFAULT_TOOLKIT_MODEL_REPO
         branch = DEFAULT_PYTHON_MODEL_REPO_BRANCH
@@ -320,7 +324,9 @@ def init(
     if (user_id or model_name or port or context_length) and (toolkit == 'lmstudio'):
         customize_lmstudio_model(model_path, user_id, model_name, port, context_length)
 
-    if (user_id or model_name) and (toolkit == 'huggingface' or toolkit == 'vllm'):
+    if (user_id or model_name) and (
+        toolkit == 'huggingface' or toolkit == 'vllm' or toolkit == 'sglang'
+    ):
         # Update the config.yaml file with the provided model name
         customize_huggingface_model(model_path, user_id, model_name)
 
@@ -1010,6 +1016,7 @@ def local_runner(ctx, model_path, pool_size, verbose):
             logger.info("Customizing Ollama model with provided parameters...")
             customize_ollama_model(
                 model_path=model_path,
+                user_id=user_id,
                 verbose=True if verbose else False,
             )
         except Exception as e:

{clarifai-11.8.4 → clarifai-11.9.0}/clarifai/client/__init__.py

@@ -2,11 +2,14 @@ from clarifai.client.app import App
 from clarifai.client.auth.register import V2Stub
 from clarifai.client.auth.stub import create_stub
 from clarifai.client.base import BaseClient
+from clarifai.client.compute_cluster import ComputeCluster
 from clarifai.client.dataset import Dataset
+from clarifai.client.deployment import Deployment
 from clarifai.client.input import Inputs
 from clarifai.client.lister import Lister
 from clarifai.client.model import Model
 from clarifai.client.module import Module
+from clarifai.client.nodepool import Nodepool
 from clarifai.client.pipeline import Pipeline
 from clarifai.client.pipeline_step import PipelineStep
 from clarifai.client.search import Search
@@ -28,4 +31,7 @@ __all__ = [
     'Inputs',
     'BaseClient',
     'Search',
+    'ComputeCluster',
+    'Nodepool',
+    'Deployment',
 ]

{clarifai-11.8.4 → clarifai-11.9.0}/clarifai/client/model_client.py

@@ -11,6 +11,7 @@ from clarifai.constants.model import MAX_MODEL_PREDICT_INPUTS
 from clarifai.errors import UserError
 from clarifai.runners.utils import code_script, method_signatures
 from clarifai.runners.utils.method_signatures import (
+    RESERVED_PARAM_WITH_PROTO,
     CompatibilitySerializer,
     deserialize,
     get_stream_from_signature,
@@ -204,6 +205,9 @@ class ModelClient:
 
             def bind_f(method_name, method_argnames, call_func, async_call_func):
                 def sync_f(*args, **kwargs):
+                    # Extract with_proto parameter if present
+                    with_proto = kwargs.pop(RESERVED_PARAM_WITH_PROTO, False)
+
                     if len(args) > len(method_argnames):
                         raise TypeError(
                             f"{method_name}() takes {len(method_argnames)} positional arguments but {len(args)} were given"
@@ -221,7 +225,7 @@ class ModelClient:
                         )
                         if is_batch_input_valid and (not is_openai_chat_format(batch_inputs)):
                             # If the batch input is valid, call the function with the batch inputs and the method name
-                            return call_func(batch_inputs, method_name)
+                            return call_func(batch_inputs, method_name, with_proto=with_proto)
 
                     for name, arg in zip(
                         method_argnames, args
@@ -229,10 +233,13 @@ class ModelClient:
                         if name in kwargs:
                             raise TypeError(f"Multiple values for argument {name}")
                         kwargs[name] = arg
-                    return call_func(kwargs, method_name)
+                    return call_func(kwargs, method_name, with_proto=with_proto)
 
                 async def async_f(*args, **kwargs):
                     # Async version to call the async function
+                    # Extract with_proto parameter if present
+                    with_proto = kwargs.pop(RESERVED_PARAM_WITH_PROTO, False)
+
                     if len(args) > len(method_argnames):
                         raise TypeError(
                             f"{method_name}() takes {len(method_argnames)} positional arguments but {len(args)} were given"
@@ -249,7 +256,9 @@ class ModelClient:
                         )
                         if is_batch_input_valid and (not is_openai_chat_format(batch_inputs)):
                             # If the batch input is valid, call the function with the batch inputs and the method name
-                            return async_call_func(batch_inputs, method_name)
+                            return async_call_func(
+                                batch_inputs, method_name, with_proto=with_proto
+                            )
 
                     for name, arg in zip(
                         method_argnames, args
@@ -258,7 +267,7 @@ class ModelClient:
                             raise TypeError(f"Multiple values for argument {name}")
                         kwargs[name] = arg
 
-                    return async_call_func(kwargs, method_name)
+                    return async_call_func(kwargs, method_name, with_proto=with_proto)
 
                 class MethodWrapper:
                     def __call__(self, *args, **kwargs):
@@ -364,6 +373,7 @@ class ModelClient:
         self,
         inputs,  # TODO set up functions according to fetched signatures?
         method_name: str = 'predict',
+        with_proto: bool = False,
     ) -> Any:
         input_signature = self._method_signatures[method_name].input_fields
         output_signature = self._method_signatures[method_name].output_fields
@@ -385,9 +395,12 @@ class ModelClient:
         outputs = []
         for output in response.outputs:
             outputs.append(deserialize(output.data, output_signature, is_output=True))
-        if batch_input:
-            return outputs
-        return outputs[0]
+
+        result = outputs if batch_input else outputs[0]
+
+        if with_proto:
+            return result, response
+        return result
 
     def _predict_by_proto(
         self,
@@ -448,15 +461,17 @@ class ModelClient:
         self,
         inputs,
         method_name: str = 'predict',
+        with_proto: bool = False,
     ) -> Any:
         """Asynchronously process inputs and make predictions.
 
         Args:
             inputs: Input data to process
             method_name (str): Name of the method to call
+            with_proto (bool): If True, return both the processed result and the raw protobuf response
 
         Returns:
-            Processed prediction results
+            Processed prediction results, optionally with protobuf response
         """
         # method_name is set to 'predict' by default, this is because to replicate the input and output signature behaviour of sync to async predict.
         input_signature = self._method_signatures[method_name].input_fields
@@ -477,7 +492,11 @@ class ModelClient:
         for output in response.outputs:
             outputs.append(deserialize(output.data, output_signature, is_output=True))
 
-        return outputs if batch_input else outputs[0]
+        result = outputs if batch_input else outputs[0]
+
+        if with_proto:
+            return result, response
+        return result
 
     async def _async_predict_by_proto(
         self,
@@ -551,6 +570,7 @@ class ModelClient:
         self,
         inputs,  # TODO set up functions according to fetched signatures?
         method_name: str = 'generate',
+        with_proto: bool = False,
     ) -> Any:
         input_signature = self._method_signatures[method_name].input_fields
         output_signature = self._method_signatures[method_name].output_fields
@@ -572,10 +592,13 @@ class ModelClient:
             outputs = []
             for output in response.outputs:
                 outputs.append(deserialize(output.data, output_signature, is_output=True))
-            if batch_input:
-                yield outputs
+
+            result = outputs if batch_input else outputs[0]
+
+            if with_proto:
+                yield result, response
             else:
-                yield outputs[0]
+                yield result
 
     def _generate_by_proto(
         self,
@@ -641,6 +664,7 @@ class ModelClient:
         self,
         inputs,
         method_name: str = 'generate',
+        with_proto: bool = False,
     ) -> Any:
         # method_name is set to 'generate' by default, this is because to replicate the input and output signature behaviour of sync to async generate.
         input_signature = self._method_signatures[method_name].input_fields
@@ -654,18 +678,21 @@ class ModelClient:
         proto_inputs = []
         for input in inputs:
             proto = resources_pb2.Input()
-        serialize(input, input_signature, proto.data)
-        proto_inputs.append(proto)
+            serialize(input, input_signature, proto.data)
+            proto_inputs.append(proto)
         response_stream = self._async_generate_by_proto(proto_inputs, method_name)
 
         async for response in response_stream:
             outputs = []
             for output in response.outputs:
                 outputs.append(deserialize(output.data, output_signature, is_output=True))
-            if batch_input:
-                yield outputs
+
+            result = outputs if batch_input else outputs[0]
+
+            if with_proto:
+                yield result, response
             else:
-                yield outputs[0]
+                yield result
 
     async def _async_generate_by_proto(
         self,
@@ -734,6 +761,7 @@ class ModelClient:
         self,
         inputs,
         method_name: str = 'stream',
+        with_proto: bool = False,
     ) -> Any:
         input_signature = self._method_signatures[method_name].input_fields
         output_signature = self._method_signatures[method_name].output_fields
@@ -775,7 +803,12 @@ class ModelClient:
 
         for response in response_stream:
             assert len(response.outputs) == 1, 'streaming methods must have exactly one output'
-            yield deserialize(response.outputs[0].data, output_signature, is_output=True)
+            result = deserialize(response.outputs[0].data, output_signature, is_output=True)
+
+            if with_proto:
+                yield result, response
+            else:
+                yield result
 
     def _req_iterator(
         self,
@@ -843,6 +876,7 @@ class ModelClient:
         self,
         inputs,
         method_name: str = 'stream',
+        with_proto: bool = False,
     ) -> Any:
         # method_name is set to 'stream' by default, this is because to replicate the input and output signature behaviour of sync to async stream.
         input_signature = self._method_signatures[method_name].input_fields
@@ -885,7 +919,12 @@ class ModelClient:
 
         async for response in response_stream:
             assert len(response.outputs) == 1, 'streaming methods must have exactly one output'
-            yield deserialize(response.outputs[0].data, output_signature, is_output=True)
+            result = deserialize(response.outputs[0].data, output_signature, is_output=True)
+
+            if with_proto:
+                yield result, response
+            else:
+                yield result
 
     async def _async_stream_by_proto(
         self,