Do not record tool call id if no tool has been called

Makefile

@@ -1,18 +1,20 @@
 .PHONY: quality style test docs utils
 
+PYTHON ?= python
+
 check_dirs := examples src tests utils
 
 # Check code quality of the source code
 quality:
-	ruff check $(check_dirs)
-	ruff format --check $(check_dirs)
-	python utils/check_tests_in_ci.py
+	$(PYTHON) -m ruff check $(check_dirs)
+	$(PYTHON) -m ruff format --check $(check_dirs)
+	$(PYTHON) utils/check_tests_in_ci.py
 
 # Format source code automatically
 style:
-	ruff check $(check_dirs) --fix
-	ruff format $(check_dirs)
+	$(PYTHON) -m ruff check $(check_dirs) --fix
+	$(PYTHON) -m ruff format $(check_dirs)
 
 # Run smolagents tests
 test:
-	pytest ./tests/
+	$(PYTHON) -m pytest ./tests/

pyproject.toml

@@ -83,6 +83,7 @@ test = [
 dev = [
   "smolagents[quality,test]",
   "sqlalchemy", # for ./examples
+  "mlx-lm",
 ]
 
 [tool.pytest.ini_options]
@@ -111,4 +112,4 @@ lines-after-imports = 2
 
 [project.scripts]
 smolagent = "smolagents.cli:main"
-webagent = "smolagents.vision_web_browser:main"
+webagent = "smolagents.vision_web_browser:main"

src/smolagents/memory.py

@@ -1,17 +1,12 @@
 from dataclasses import asdict, dataclass
 from logging import getLogger
-from typing import TYPE_CHECKING, Any, Dict, List, TypedDict, Union
+from typing import Any, List, TypedDict, Union
 
 from smolagents.models import ChatMessage, MessageRole
 from smolagents.monitoring import AgentLogger, LogLevel
 from smolagents.utils import AgentError, make_json_serializable
 
 
-if TYPE_CHECKING:
-    from smolagents.models import ChatMessage
-    from smolagents.monitoring import AgentLogger
-
-
 logger = getLogger(__name__)
 
 
@@ -42,7 +37,7 @@ class MemoryStep:
     def dict(self):
         return asdict(self)
 
-    def to_messages(self, **kwargs) -> List[Dict[str, Any]]:
+    def to_messages(self, **kwargs) -> List[Message]:
         raise NotImplementedError
 
 
@@ -55,7 +50,7 @@ class ActionStep(MemoryStep):
     step_number: int | None = None
     error: AgentError | None = None
     duration: float | None = None
-    model_output_message: ChatMessage = None
+    model_output_message: ChatMessage | None = None
     model_output: str | None = None
     observations: str | None = None
     observations_images: List[str] | None = None
@@ -78,7 +73,7 @@ def dict(self):
         }
 
     def to_messages(self, summary_mode: bool = False, show_model_input_messages: bool = False) -> List[Message]:
-        messages = []
+        messages: List[Message] = []
         if self.model_input_messages is not None and show_model_input_messages:
             messages.append(Message(role=MessageRole.SYSTEM, content=self.model_input_messages))
         if self.model_output is not None and not summary_mode:
@@ -106,7 +101,8 @@ def to_messages(self, summary_mode: bool = False, show_model_input_messages: boo
                     content=[
                         {
                             "type": "text",
-                            "text": f"Call id: {self.tool_calls[0].id}\nObservation:\n{self.observations}",
+                            "text": (f"Call id: {self.tool_calls[0].id}\n" if self.tool_calls else "")
+                            + f"Observation:\n{self.observations}",
                         }
                     ],
                 )

src/smolagents/models.py

@@ -105,7 +105,7 @@ def from_hf_api(cls, message, raw) -> "ChatMessage":
         return cls(role=message.role, content=message.content, tool_calls=tool_calls, raw=raw)
 
     @classmethod
-    def from_dict(cls, data: dict) -> "ChatMessage":
+    def from_dict(cls, data: dict[str, Any]) -> "ChatMessage":
         if data.get("tool_calls"):
             tool_calls = [
                 ChatMessageToolCall(
@@ -123,11 +123,11 @@ def dict(self):
 def parse_json_if_needed(arguments: Union[str, dict]) -> Union[str, dict]:
     if isinstance(arguments, dict):
         return arguments
-    else:
-        try:
-            return json.loads(arguments)
-        except Exception:
-            return arguments
+
+    try:
+        return json.loads(arguments)
+    except Exception:
+        return arguments
 
 
 def parse_tool_args_if_needed(message: ChatMessage) -> ChatMessage:
@@ -155,7 +155,7 @@ def roles(cls):
 }
 
 
-def get_tool_json_schema(tool: Tool) -> Dict:
+def get_tool_json_schema(tool: Tool) -> dict[str, Any]:
     properties = deepcopy(tool.inputs)
     required = []
     for key, value in properties.items():
@@ -298,7 +298,7 @@ def _prepare_completion_kwargs(
 
         return completion_kwargs
 
-    def get_token_counts(self) -> Dict[str, int]:
+    def get_token_counts(self) -> dict[str, int | None]:
         return {
             "input_token_count": self.last_input_token_count,
             "output_token_count": self.last_output_token_count,
@@ -331,11 +331,11 @@ def __call__(
         """
         pass  # To be implemented in child classes!
 
-    def to_dict(self) -> Dict:
+    def to_dict(self) -> dict[str, Any]:
         """
         Converts the model into a JSON-compatible dictionary.
         """
-        model_dictionary = {
+        model_dictionary: dict[str, Any] = {
             **self.kwargs,
             "last_input_token_count": self.last_input_token_count,
             "last_output_token_count": self.last_output_token_count,

tests/test_agents.py

@@ -33,7 +33,7 @@
     populate_template,
 )
 from smolagents.default_tools import DuckDuckGoSearchTool, FinalAnswerTool, PythonInterpreterTool, VisitWebpageTool
-from smolagents.memory import PlanningStep
+from smolagents.memory import ActionStep, PlanningStep
 from smolagents.models import (
     ChatMessage,
     ChatMessageToolCall,
@@ -146,7 +146,7 @@ def __call__(self, messages, tools_to_call_from=None, stop_sequences=None, gramm
             )
 
 
-def fake_code_model(messages, stop_sequences=None, grammar=None) -> str:
+def fake_code_model(messages, stop_sequences=None, grammar=None) -> ChatMessage:
     prompt = str(messages)
     if "special_marker" not in prompt:
         return ChatMessage(
@@ -172,7 +172,7 @@ def fake_code_model(messages, stop_sequences=None, grammar=None) -> str:
         )
 
 
-def fake_code_model_error(messages, stop_sequences=None) -> str:
+def fake_code_model_error(messages, stop_sequences=None) -> ChatMessage:
     prompt = str(messages)
     if "special_marker" not in prompt:
         return ChatMessage(
@@ -202,7 +202,7 @@ def error_function():
         )
 
 
-def fake_code_model_syntax_error(messages, stop_sequences=None) -> str:
+def fake_code_model_syntax_error(messages, stop_sequences=None) -> ChatMessage:
     prompt = str(messages)
     if "special_marker" not in prompt:
         return ChatMessage(
@@ -231,7 +231,7 @@ def fake_code_model_syntax_error(messages, stop_sequences=None) -> str:
         )
 
 
-def fake_code_model_import(messages, stop_sequences=None) -> str:
+def fake_code_model_import(messages, stop_sequences=None) -> ChatMessage:
     return ChatMessage(
         role="assistant",
         content="""
@@ -245,7 +245,7 @@ def fake_code_model_import(messages, stop_sequences=None) -> str:
     )
 
 
-def fake_code_functiondef(messages, stop_sequences=None) -> str:
+def fake_code_functiondef(messages, stop_sequences=None) -> ChatMessage:
     prompt = str(messages)
     if "special_marker" not in prompt:
         return ChatMessage(
@@ -276,7 +276,7 @@ def moving_average(x, w):
         )
 
 
-def fake_code_model_single_step(messages, stop_sequences=None, grammar=None) -> str:
+def fake_code_model_single_step(messages, stop_sequences=None, grammar=None) -> ChatMessage:
     return ChatMessage(
         role="assistant",
         content="""
@@ -290,7 +290,7 @@ def fake_code_model_single_step(messages, stop_sequences=None, grammar=None) ->
     )
 
 
-def fake_code_model_no_return(messages, stop_sequences=None, grammar=None) -> str:
+def fake_code_model_no_return(messages, stop_sequences=None, grammar=None) -> ChatMessage:
     return ChatMessage(
         role="assistant",
         content="""
@@ -304,6 +304,18 @@ def fake_code_model_no_return(messages, stop_sequences=None, grammar=None) -> st
     )
 
 
+def fake_no_valid_code_block(messages, stop_sequences=None, grammar=None) -> ChatMessage:
+    return ChatMessage(
+        role="assistant",
+        content="""
+Thought: I should multiply 2 by 3.6452. special_marker
+Code:
+ I am Incorrect Python Code !rint(result)
+```
+""",
+    )
+
+
 class AgentTests(unittest.TestCase):
     def test_fake_toolcalling_agent(self):
         agent = ToolCallingAgent(tools=[PythonInterpreterTool()], model=FakeToolCallModel())
@@ -735,6 +747,26 @@ def test_planning_step(self, step, expected_messages_list):
                 for content, expected_content in zip(message["content"], expected_message["content"]):
                     assert content == expected_content
 
+    def test_agent_memory_to_messages_suceeds_when_tool_fails_but_obeservation_is_set(self):
+        tool = PythonInterpreterTool()
+
+        def _fake_callback(memory_step: ActionStep, agent: CodeAgent) -> None:
+            memory_step.observations = "observed something"
+
+        agent = CodeAgent(
+            tools=[tool],
+            model=fake_no_valid_code_block,
+            step_callbacks=[_fake_callback],
+        )
+
+        # Perform a task. A tool call will fail since the fake model returns invalid response, but a _fake_callback
+        # sets an observation
+        agent.run("some task")
+
+        # This should not fail, even though no tool call have been recorded, and observation came from a callback
+        for s in agent.memory.steps:
+            s.to_messages()
+
     @pytest.mark.parametrize(
         "images, expected_messages_list",
         [
@@ -904,6 +936,10 @@ def fake_code_model(messages, stop_sequences=None, grammar=None) -> str:
 
 
 class MultiAgentsTests(unittest.TestCase):
+    @pytest.fixture(autouse=True)
+    def initdir(self, tmp_path, monkeypatch):
+        monkeypatch.chdir(tmp_path)
+
     def test_multiagents_save(self):
         model = HfApiModel("Qwen/Qwen2.5-Coder-32B-Instruct", max_tokens=2096, temperature=0.5)