Halve import time by removing torch dependency (#147)

* Halve import time by removing torch dependency

Author: aymeric-roucher

.github/workflows/quality.yml

@@ -13,7 +13,7 @@ jobs:
       - name: Set up Python
         uses: actions/setup-python@v2
         with:
-          python-version: "3.10"
+          python-version: "3.12"
 
       # Setup venv
       # TODO: revisit when https://github.com/astral-sh/uv/issues/1526 is addressed.

docs/source/en/examples/multiagents.md

@@ -48,10 +48,10 @@ Run the line below to install the required dependencies:
 
 Let's login in order to call the HF Inference API:
 
-```py
-from huggingface_hub import notebook_login
+```
+from huggingface_hub import login
 
-notebook_login()
+login()
 ```
 
 ⚡️ Our agent will be powered by [Qwen/Qwen2.5-Coder-32B-Instruct](https://huggingface.co/Qwen/Qwen2.5-Coder-32B-Instruct) using `HfApiModel` class that uses HF's Inference API: the Inference API allows to quickly and easily run any OS model.

docs/source/en/tutorials/tools.md

@@ -177,7 +177,7 @@ agent.run("How many more blocks (also denoted as layers) are in BERT base encode
 
 ### Manage your agent's toolbox
 
-You can manage an agent's toolbox by adding or replacing a tool.
+You can manage an agent's toolbox by adding or replacing a tool in attribute `agent.tools`, since it is a standard dictionary.
 
 Let's add the `model_download_tool` to an existing agent initialized with only the default toolbox.
 
@@ -187,7 +187,7 @@ from smolagents import HfApiModel
 model = HfApiModel("Qwen/Qwen2.5-Coder-32B-Instruct")
 
 agent = CodeAgent(tools=[], model=model, add_base_tools=True)
-agent.tools.append(model_download_tool)
+agent.tools[model_download_tool.name] = model_download_tool
 ```
 Now we can leverage the new tool:
 

pyproject.toml

@@ -12,9 +12,6 @@ authors = [
 readme = "README.md"
 requires-python = ">=3.10"
 dependencies = [
-  "torch",
-  "torchaudio",
-  "torchvision",
   "transformers>=4.0.0",
   "requests>=2.32.3",
   "rich>=13.9.4",
@@ -30,10 +27,22 @@ dependencies = [
 ]
 
 [tool.ruff]
-ignore = ["F403"]
+lint.ignore = ["F403"]
 
 [project.optional-dependencies]
+dev = [
+  "torch",
+  "torchaudio",
+  "torchvision",
+  "sqlalchemy",
+  "accelerate",
+  "soundfile",
+  "litellm>=1.55.10",
+]
 test = [
+  "torch",
+  "torchaudio",
+  "torchvision",
   "pytest>=8.1.0",
   "sqlalchemy",
   "ruff>=0.5.0",

src/smolagents/default_tools.py

@@ -20,11 +20,9 @@
 from typing import Dict, Optional
 
 from huggingface_hub import hf_hub_download, list_spaces
-from transformers.models.whisper import (
-    WhisperForConditionalGeneration,
-    WhisperProcessor,
-)
-from transformers.utils import is_offline_mode
+
+
+from transformers.utils import is_offline_mode, is_torch_available
 
 from .local_python_executor import (
     BASE_BUILTIN_MODULES,
@@ -34,6 +32,15 @@
 from .tools import TOOL_CONFIG_FILE, PipelineTool, Tool
 from .types import AgentAudio
 
+if is_torch_available():
+    from transformers.models.whisper import (
+        WhisperForConditionalGeneration,
+        WhisperProcessor,
+    )
+else:
+    WhisperForConditionalGeneration = object
+    WhisperProcessor = object
+
 
 @dataclass
 class PreTool:

src/smolagents/models.py

@@ -22,7 +22,6 @@
 from enum import Enum
 from typing import Dict, List, Optional
 
-import torch
 from huggingface_hub import (
     InferenceClient,
     ChatCompletionOutputMessage,
@@ -35,6 +34,7 @@
     AutoTokenizer,
     StoppingCriteria,
     StoppingCriteriaList,
+    is_torch_available,
 )
 import openai
 
@@ -147,29 +147,12 @@ def __init__(self):
         self.last_input_token_count = None
         self.last_output_token_count = None
 
-    def get_token_counts(self):
+    def get_token_counts(self) -> Dict[str, int]:
         return {
             "input_token_count": self.last_input_token_count,
             "output_token_count": self.last_output_token_count,
         }
 
-    def generate(
-        self,
-        messages: List[Dict[str, str]],
-        stop_sequences: Optional[List[str]] = None,
-        grammar: Optional[str] = None,
-        max_tokens: int = 1500,
-    ):
-        raise NotImplementedError
-
-    def get_tool_call(
-        self,
-        messages: List[Dict[str, str]],
-        available_tools: List[Tool],
-        stop_sequences,
-    ):
-        raise NotImplementedError
-
     def __call__(
         self,
         messages: List[Dict[str, str]],
@@ -256,6 +239,10 @@ def __call__(
         max_tokens: int = 1500,
         tools_to_call_from: Optional[List[Tool]] = None,
     ) -> str:
+        """
+        Gets an LLM output message for the given list of input messages.
+        If argument `tools_to_call_from` is passed, the model's tool calling options will be used to return a tool call.
+        """
         messages = get_clean_message_list(
             messages, role_conversions=tool_role_conversions
         )
@@ -293,6 +280,10 @@ class TransformersModel(Model):
 
     def __init__(self, model_id: Optional[str] = None, device: Optional[str] = None):
         super().__init__()
+        if not is_torch_available():
+            raise ImportError("Please install torch in order to use TransformersModel.")
+        import torch
+
         default_model_id = "HuggingFaceTB/SmolLM2-1.7B-Instruct"
         if model_id is None:
             model_id = default_model_id

src/smolagents/tools.py

@@ -27,7 +27,6 @@
 from pathlib import Path
 from typing import Callable, Dict, Optional, Union, get_type_hints
 
-import torch
 from huggingface_hub import (
     create_repo,
     get_collection,
@@ -37,7 +36,6 @@
 )
 from huggingface_hub.utils import RepositoryNotFoundError
 from packaging import version
-from transformers import AutoProcessor
 from transformers.dynamic_module_utils import get_imports
 from transformers.utils import (
     TypeHintParsingException,
@@ -54,13 +52,14 @@
 
 logger = logging.getLogger(__name__)
 
-
-if is_torch_available():
-    pass
-
 if is_accelerate_available():
-    pass
+    from accelerate import PartialState
+    from accelerate.utils import send_to_device
 
+if is_torch_available():
+    from transformers import AutoProcessor
+else:
+    AutoProcessor = object
 
 TOOL_CONFIG_FILE = "tool_config.json"
 
@@ -1026,8 +1025,6 @@ def setup(self):
         """
         Instantiates the `pre_processor`, `model` and `post_processor` if necessary.
         """
-        from accelerate import PartialState
-
         if isinstance(self.pre_processor, str):
             self.pre_processor = self.pre_processor_class.from_pretrained(
                 self.pre_processor, **self.hub_kwargs
@@ -1066,6 +1063,8 @@ def forward(self, inputs):
         """
         Sends the inputs through the `model`.
         """
+        import torch
+
         with torch.no_grad():
             return self.model(**inputs)
 
@@ -1076,16 +1075,15 @@ def decode(self, outputs):
         return self.post_processor(outputs)
 
     def __call__(self, *args, **kwargs):
+        import torch
+
         args, kwargs = handle_agent_input_types(*args, **kwargs)
 
         if not self.is_initialized:
             self.setup()
 
         encoded_inputs = self.encode(*args, **kwargs)
 
-        import torch
-        from accelerate.utils import send_to_device
-
         tensor_inputs = {
             k: v for k, v in encoded_inputs.items() if isinstance(v, torch.Tensor)
         }

src/smolagents/types.py

@@ -22,10 +22,10 @@
 import numpy as np
 import requests
 from transformers.utils import (
-    is_soundfile_availble,
     is_torch_available,
     is_vision_available,
 )
+from transformers.utils.import_utils import _is_package_available
 
 logger = logging.getLogger(__name__)
 
@@ -41,7 +41,7 @@
 else:
     Tensor = object
 
-if is_soundfile_availble():
+if _is_package_available("soundfile"):
     import soundfile as sf
 
 
@@ -189,7 +189,7 @@ class AgentAudio(AgentType, str):
     def __init__(self, value, samplerate=16_000):
         super().__init__(value)
 
-        if not is_soundfile_availble():
+        if not _is_package_available("soundfile"):
             raise ImportError("soundfile must be installed in order to handle audio.")
 
         self._path = None
@@ -253,7 +253,7 @@ def to_string(self):
 INSTANCE_TYPE_MAPPING = {
     str: AgentText,
     ImageType: AgentImage,
-    torch.Tensor: AgentAudio,
+    Tensor: AgentAudio,
 }
 
 if is_torch_available():

tests/test_types.py

@@ -18,20 +18,19 @@
 import uuid
 from pathlib import Path
 
-import torch
 from PIL import Image
 from transformers.testing_utils import (
     require_soundfile,
     require_torch,
     require_vision,
 )
-from transformers.utils import (
-    is_soundfile_availble,
+from transformers.utils.import_utils import (
+    _is_package_available,
 )
 
 from smolagents.types import AgentAudio, AgentImage, AgentText
 
-if is_soundfile_availble():
+if _is_package_available("soundfile"):
     import soundfile as sf
 
 
@@ -44,6 +43,8 @@ def get_new_path(suffix="") -> str:
 @require_torch
 class AgentAudioTests(unittest.TestCase):
     def test_from_tensor(self):
+        import torch
+
         tensor = torch.rand(12, dtype=torch.float64) - 0.5
         agent_type = AgentAudio(tensor)
         path = str(agent_type.to_string())
@@ -61,6 +62,8 @@ def test_from_tensor(self):
         self.assertTrue(torch.allclose(tensor, torch.tensor(new_tensor), atol=1e-4))
 
     def test_from_string(self):
+        import torch
+
         tensor = torch.rand(12, dtype=torch.float64) - 0.5
         path = get_new_path(suffix=".wav")
         sf.write(path, tensor, 16000)
@@ -75,6 +78,8 @@ def test_from_string(self):
 @require_torch
 class AgentImageTests(unittest.TestCase):
     def test_from_tensor(self):
+        import torch
+
         tensor = torch.randint(0, 256, (64, 64, 3))
         agent_type = AgentImage(tensor)
         path = str(agent_type.to_string())