KeyArgo
/
smolagents
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Replace max_iteration with max_steps for consistency

This commit is contained in:
Aymeric 2025-01-06 17:24:02 +01:00
parent 07015d12fe
commit e9119c9df5
12 changed files with 71 additions and 68 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
docs/source/en/examples/multiagents.md
View File

@ -120,7 +120,7 @@ Now that we have all the tools `search` and `visit_webpage`, we can use them to
Which configuration to choose for this agent?
- Web browsing is a single-timeline task that does not require parallel tool calls, so JSON tool calling works well for that. We thus choose a `JsonAgent`.
- Also, since sometimes web search requires exploring many pages before finding the correct answer, we prefer to increase the number of `max_iterations` to 10.
- Also, since sometimes web search requires exploring many pages before finding the correct answer, we prefer to increase the number of `max_steps` to 10.
```py
from smolagents import (
@ -137,7 +137,7 @@ model = HfApiModel(model_id)
web_agent = ToolCallingAgent(
tools=[DuckDuckGoSearchTool(), visit_webpage],
model=model,
max_iterations=10,
max_steps=10,
)
```

2
docs/source/en/examples/rag.md
View File

@ -137,7 +137,7 @@ _Note:_ The Inference API hosts models based on various criteria, and deployed m
from smolagents import HfApiModel, CodeAgent
agent = CodeAgent(
tools=[retriever_tool], model=HfApiModel("meta-llama/Llama-3.3-70B-Instruct"), max_iterations=4, verbose=True
tools=[retriever_tool], model=HfApiModel("meta-llama/Llama-3.3-70B-Instruct"), max_steps=4, verbose=True
)
```

5
docs/source/en/reference/agents.md
View File

@ -125,6 +125,7 @@ print(model(messages))
### LiteLLMModel
The `LiteLLMModel` leverages [LiteLLM](https://www.litellm.ai/) to support 100+ LLMs from various providers.
You can pass kwargs upon model initialization that will then be used whenever using the model, for instance below we pass `temperature`.
```python
from smolagents import LiteLLMModel
@ -135,8 +136,8 @@ messages = [
{"role": "user", "content": "No need to help, take it easy."},
]
model = LiteLLMModel("anthropic/claude-3-5-sonnet-latest")
print(model(messages))
model = LiteLLMModel("anthropic/claude-3-5-sonnet-latest", temperature=0.2)
print(model(messages, max_tokens=10))
```
[[autodoc]] LiteLLMModel

2
docs/source/en/tutorials/building_good_agents.md
View File

@ -194,7 +194,7 @@ If after trying the above, you still want to change the system prompt, your new
Then you can change the system prompt as follows:
```py
from smolagents.prompts import CODE_SYSTEM_PROMPT, HfApiModel
from smolagents.prompts import CODE_SYSTEM_PROMPT
modified_system_prompt = CODE_SYSTEM_PROMPT + "\nHere you go!" # Change the system prompt here

8
examples/benchmark.ipynb
View File

@ -233,7 +233,7 @@
" agent = ToolCallingAgent(\n",
" tools=[GoogleSearchTool(), VisitWebpageTool(), PythonInterpreterTool()],\n",
" model=HfApiModel(model_id),\n",
" max_iterations=10,\n",
" max_steps=10,\n",
" )\n",
" file_name = f\"output/{model_id.replace('/', '_')}-{action_type}-26-dec-2024.jsonl\"\n",
" answer_questions(eval_ds, file_name, agent, model_id, action_type)\n",
@ -243,7 +243,7 @@
" tools=[GoogleSearchTool(), VisitWebpageTool()],\n",
" model=HfApiModel(model_id),\n",
" additional_authorized_imports=[\"numpy\"],\n",
" max_iterations=10,\n",
" max_steps=10,\n",
" )\n",
" file_name = f\"output/{model_id.replace('/', '_')}-{action_type}-26-dec-2024.jsonl\"\n",
" answer_questions(eval_ds, file_name, agent, model_id, action_type)"
@ -272,7 +272,7 @@
" agent = ToolCallingAgent(\n",
" tools=[GoogleSearchTool(), VisitWebpageTool(), PythonInterpreterTool()],\n",
" model=LiteLLMModel(model_id),\n",
" max_iterations=10,\n",
" max_steps=10,\n",
" )\n",
" file_name = f\"output/{model_id.replace('/', '_')}-{action_type}-26-dec-2024.jsonl\"\n",
" answer_questions(eval_ds, file_name, agent, model_id, action_type)\n",
@ -282,7 +282,7 @@
" tools=[GoogleSearchTool(), VisitWebpageTool()],\n",
" model=LiteLLMModel(model_id),\n",
" additional_authorized_imports=[\"numpy\"],\n",
" max_iterations=10,\n",
" max_steps=10,\n",
" )\n",
" file_name = f\"output/{model_id.replace('/', '_')}-{action_type}-26-dec-2024.jsonl\"\n",
" answer_questions(eval_ds, file_name, agent, model_id, action_type)"

2
examples/rag.py
View File

@ -60,7 +60,7 @@ from smolagents import HfApiModel, CodeAgent
retriever_tool = RetrieverTool(docs_processed)
agent = CodeAgent(
tools=[retriever_tool], model=HfApiModel("meta-llama/Llama-3.3-70B-Instruct"), max_iterations=4, verbose=True
tools=[retriever_tool], model=HfApiModel("meta-llama/Llama-3.3-70B-Instruct"), max_steps=4, verbose=True
)
agent_output = agent.run("For a transformers model training, which is slower, the forward or the backward pass?")

58
src/smolagents/agents.py
View File

@ -32,7 +32,7 @@ from .utils import (
AgentParsingError,
AgentExecutionError,
AgentGenerationError,
AgentMaxIterationsError,
AgentMaxStepsError,
)
from .types import AgentAudio, AgentImage, handle_agent_output_types
from .default_tools import FinalAnswerTool
@ -78,7 +78,7 @@ class ActionStep(AgentStep):
tool_call: ToolCall | None = None
start_time: float | None = None
end_time: float | None = None
iteration: int | None = None
step: int | None = None
error: AgentError | None = None
duration: float | None = None
llm_output: str | None = None
@ -163,7 +163,7 @@ class MultiStepAgent:
model: Callable[[List[Dict[str, str]]], str],
system_prompt: Optional[str] = None,
tool_description_template: Optional[str] = None,
max_iterations: int = 6,
max_steps: int = 6,
tool_parser: Optional[Callable] = None,
add_base_tools: bool = False,
verbose: bool = False,
@ -184,7 +184,7 @@ class MultiStepAgent:
if tool_description_template
else DEFAULT_TOOL_DESCRIPTION_TEMPLATE
)
self.max_iterations = max_iterations
self.max_steps = max_steps
self.tool_parser = tool_parser
self.grammar = grammar
self.planning_interval = planning_interval
@ -500,20 +500,18 @@ You have been provided with these additional arguments, that you can access usin
Runs the agent in streaming mode, yielding steps as they are executed: should be launched only in the `run` method.
"""
final_answer = None
iteration = 0
while final_answer is None and iteration < self.max_iterations:
step = 0
while final_answer is None and step < self.max_steps:
step_start_time = time.time()
step_log = ActionStep(iteration=iteration, start_time=step_start_time)
step_log = ActionStep(step=step, start_time=step_start_time)
try:
if (
self.planning_interval is not None
and iteration % self.planning_interval == 0
and step % self.planning_interval == 0
):
self.planning_step(
task, is_first_step=(iteration == 0), iteration=iteration
)
self.planning_step(task, is_first_step=(step == 0), step=step)
console.print(
Rule(f"[bold]Step {iteration}", characters="━", style=YELLOW_HEX)
Rule(f"[bold]Step {step}", characters="━", style=YELLOW_HEX)
)
# Run one step!
@ -526,12 +524,12 @@ You have been provided with these additional arguments, that you can access usin
self.logs.append(step_log)
for callback in self.step_callbacks:
callback(step_log)
iteration += 1
step += 1
yield step_log
if final_answer is None and iteration == self.max_iterations:
error_message = "Reached max iterations."
final_step_log = ActionStep(error=AgentMaxIterationsError(error_message))
if final_answer is None and step == self.max_steps:
error_message = "Reached max steps."
final_step_log = ActionStep(error=AgentMaxStepsError(error_message))
self.logs.append(final_step_log)
final_answer = self.provide_final_answer(task)
console.print(Text(f"Final answer: {final_answer}"))
@ -549,20 +547,18 @@ You have been provided with these additional arguments, that you can access usin
Runs the agent in direct mode, returning outputs only at the end: should be launched only in the `run` method.
"""
final_answer = None
iteration = 0
while final_answer is None and iteration < self.max_iterations:
step = 0
while final_answer is None and step < self.max_steps:
step_start_time = time.time()
step_log = ActionStep(iteration=iteration, start_time=step_start_time)
step_log = ActionStep(step=step, start_time=step_start_time)
try:
if (
self.planning_interval is not None
and iteration % self.planning_interval == 0
and step % self.planning_interval == 0
):
self.planning_step(
task, is_first_step=(iteration == 0), iteration=iteration
)
self.planning_step(task, is_first_step=(step == 0), step=step)
console.print(
Rule(f"[bold]Step {iteration}", characters="━", style=YELLOW_HEX)
Rule(f"[bold]Step {step}", characters="━", style=YELLOW_HEX)
)
# Run one step!
@ -577,11 +573,11 @@ You have been provided with these additional arguments, that you can access usin
self.logs.append(step_log)
for callback in self.step_callbacks:
callback(step_log)
iteration += 1
step += 1
if final_answer is None and iteration == self.max_iterations:
error_message = "Reached max iterations."
final_step_log = ActionStep(error=AgentMaxIterationsError(error_message))
if final_answer is None and step == self.max_steps:
error_message = "Reached max steps."
final_step_log = ActionStep(error=AgentMaxStepsError(error_message))
self.logs.append(final_step_log)
final_answer = self.provide_final_answer(task)
console.print(Text(f"Final answer: {final_answer}"))
@ -592,14 +588,14 @@ You have been provided with these additional arguments, that you can access usin
return handle_agent_output_types(final_answer)
def planning_step(self, task, is_first_step: bool, iteration: int):
def planning_step(self, task, is_first_step: bool, step: int):
"""
Used periodically by the agent to plan the next steps to reach the objective.
Args:
task (`str`): The task to perform
is_first_step (`bool`): If this step is not the first one, the plan should be an update over a previous plan.
iteration (`int`): The number of the current step, used as an indication for the LLM.
step (`int`): The number of the current step, used as an indication for the LLM.
"""
if is_first_step:
message_prompt_facts = {
@ -687,7 +683,7 @@ Now begin!""",
show_agents_descriptions(self.managed_agents)
),
facts_update=facts_update,
remaining_steps=(self.max_iterations - iteration),
remaining_steps=(self.max_steps - step),
),
}
plan_update = self.model(

18
src/smolagents/models.py
View File

@ -281,12 +281,12 @@ class HfApiModel(Model):
class TransformersModel(Model):
"""This engine initializes a model and tokenizer from the given `model_id`.
Parameters:
model_id (`str`, *optional*, defaults to `"HuggingFaceTB/SmolLM2-1.7B-Instruct"`):
The Hugging Face model ID to be used for inference. This can be a path or model identifier from the Hugging Face model hub.
device (`str`, optional, defaults to `"cuda"` if available, else `"cpu"`.):
The device to load the model on (`"cpu"` or `"cuda"`).
Parameters:
model_id (`str`, *optional*, defaults to `"HuggingFaceTB/SmolLM2-1.7B-Instruct"`):
The Hugging Face model ID to be used for inference. This can be a path or model identifier from the Hugging Face model hub.
device (`str`, optional, defaults to `"cuda"` if available, else `"cpu"`.):
The device to load the model on (`"cpu"` or `"cuda"`).
"""
def __init__(self, model_id: Optional[str] = None, device: Optional[str] = None):
@ -310,7 +310,9 @@ class TransformersModel(Model):
f"Failed to load tokenizer and model for {model_id=}: {e}. Loading default tokenizer and model instead from {model_id=}."
)
self.tokenizer = AutoTokenizer.from_pretrained(default_model_id)
self.model = AutoModelForCausalLM.from_pretrained(default_model_id).to(self.device)
self.model = AutoModelForCausalLM.from_pretrained(default_model_id).to(
self.device
)
def make_stopping_criteria(self, stop_sequences: List[str]) -> StoppingCriteriaList:
class StopOnStrings(StoppingCriteria):
@ -424,7 +426,7 @@ class LiteLLMModel(Model):
model_id="anthropic/claude-3-5-sonnet-20240620",
api_base=None,
api_key=None,
**kwargs
**kwargs,
):
super().__init__()
self.model_id = model_id

10
src/smolagents/tools.py
View File

@ -575,7 +575,6 @@ class Tool:
from gradio_client import Client, handle_file
class SpaceToolWrapper(Tool):
skip_forward_signature_validation = True
def __init__(
@ -907,12 +906,17 @@ class ToolCollection:
```
"""
def __init__(self, collection_slug: str, token: Optional[str] = None, trust_remote_code=False):
def __init__(
self, collection_slug: str, token: Optional[str] = None, trust_remote_code=False
):
self._collection = get_collection(collection_slug, token=token)
self._hub_repo_ids = {
item.item_id for item in self._collection.items if item.item_type == "space"
}
self.tools = {Tool.from_hub(repo_id,token,trust_remote_code) for repo_id in self._hub_repo_ids}
self.tools = {
Tool.from_hub(repo_id, token, trust_remote_code)
for repo_id in self._hub_repo_ids
}
def tool(tool_function: Callable) -> Tool:

4
src/smolagents/utils.py
View File

@ -31,7 +31,7 @@ def is_pygments_available():
return _pygments_available
console = Console()
console = Console(width=200)
BASE_BUILTIN_MODULES = [
"collections",
@ -69,7 +69,7 @@ class AgentExecutionError(AgentError):
pass
class AgentMaxIterationsError(AgentError):
class AgentMaxStepsError(AgentError):
"""Exception raised for errors in execution in the agent"""
pass

10
tests/test_agents.py
View File

@ -22,7 +22,7 @@ from pathlib import Path
from smolagents.types import AgentText, AgentImage
from smolagents.agents import (
AgentMaxIterationsError,
AgentMaxStepsError,
ManagedAgent,
CodeAgent,
ToolCallingAgent,
@ -279,15 +279,15 @@ class AgentTests(unittest.TestCase):
def test_setup_agent_with_empty_toolbox(self):
ToolCallingAgent(model=FakeToolCallModel(), tools=[])
def test_fails_max_iterations(self):
def test_fails_max_steps(self):
agent = CodeAgent(
tools=[PythonInterpreterTool()],
model=fake_code_model_no_return, # use this callable because it never ends
max_iterations=5,
max_steps=5,
)
agent.run("What is 2 multiplied by 3.6452?")
assert len(agent.logs) == 8
assert type(agent.logs[-1].error) is AgentMaxIterationsError
assert type(agent.logs[-1].error) is AgentMaxStepsError
def test_init_agent_with_different_toolsets(self):
toolset_1 = []
@ -325,7 +325,7 @@ class AgentTests(unittest.TestCase):
agent = CodeAgent(
tools=[],
model=fake_code_functiondef,
max_iterations=2,
max_steps=2,
additional_authorized_imports=["numpy"],
)
res = agent.run("ok")

16
tests/test_monitoring.py
View File

@ -41,7 +41,7 @@ final_answer('This is the final answer.')
agent = CodeAgent(
tools=[],
model=FakeLLMModel(),
max_iterations=1,
max_steps=1,
)
agent.run("Fake task")
@ -61,7 +61,7 @@ final_answer('This is the final answer.')
agent = ToolCallingAgent(
tools=[],
model=FakeLLMModel(),
max_iterations=1,
max_steps=1,
)
agent.run("Fake task")
@ -69,7 +69,7 @@ final_answer('This is the final answer.')
self.assertEqual(agent.monitor.total_input_token_count, 10)
self.assertEqual(agent.monitor.total_output_token_count, 20)
def test_code_agent_metrics_max_iterations(self):
def test_code_agent_metrics_max_steps(self):
class FakeLLMModel:
def __init__(self):
self.last_input_token_count = 10
@ -81,7 +81,7 @@ final_answer('This is the final answer.')
agent = CodeAgent(
tools=[],
model=FakeLLMModel(),
max_iterations=1,
max_steps=1,
)
agent.run("Fake task")
@ -103,7 +103,7 @@ final_answer('This is the final answer.')
agent = CodeAgent(
tools=[],
model=FakeLLMModel(),
max_iterations=1,
max_steps=1,
)
agent.run("Fake task")
@ -123,7 +123,7 @@ final_answer('This is the final answer.')
agent = CodeAgent(
tools=[],
model=dummy_model,
max_iterations=1,
max_steps=1,
)
# Use stream_to_gradio to capture the output
@ -145,7 +145,7 @@ final_answer('This is the final answer.')
agent = ToolCallingAgent(
tools=[],
model=FakeLLM(),
max_iterations=1,
max_steps=1,
)
# Use stream_to_gradio to capture the output
@ -172,7 +172,7 @@ final_answer('This is the final answer.')
agent = CodeAgent(
tools=[],
model=dummy_model,
max_iterations=1,
max_steps=1,
)
# Use stream_to_gradio to capture the output