Change code tags to xml

docs/source/en/conceptual_guides/intro_agents.mdx

@@ -77,7 +77,7 @@ See? With these two examples, we already found the need for a few items to help
 - Of course, an LLM that acts as the engine powering the system
 - A list of tools that the agent can access
 - A parser that extracts tool calls from the LLM output
-- A system prompt synced with the parser
+- A system prompt indicating tool calling formats that match the format expected by the action parser
 - A memory
 
 But wait, since we give room to LLMs in decisions, surely they will make mistakes: so we need error logging and retry mechanisms.

docs/source/en/tutorials/building_good_agents.mdx

@@ -158,25 +158,25 @@ Thus it cannot access the image again except by leveraging the path that was log
 
 The first step to debugging your agent is thus "Use a more powerful LLM". Alternatives like `Qwen2/5-72B-Instruct` wouldn't have made that mistake.
 
-### 2. Provide more guidance / more information
+### 2. Provide more information or specific instructions
 
 You can also use less powerful models, provided you guide them more effectively.
 
 Put yourself in the shoes of your model: if you were the model solving the task, would you struggle with the information available to you (from the system prompt + task formulation + tool description) ?
 
 Would you need some added clarifications?
 
-To provide extra information, we do not recommend to change the system prompt right away: the default system prompt has many adjustments that you do not want to mess up except if you understand the prompt very well.
 Better ways to guide your LLM engine are:
-- If it's about the task to solve: add all these details to the task. The task could be 100s of pages long.
+- If the instruction si to always be given to the agent (as we generally understand a system prompt to work): you can pass it under argument `instructions` upon agent initialization.
+- If it's about a specific task to solve: add all these details to the task. The task could be 100s of pages long.
 - If it's about how to use tools: the description attribute of your tools.
 
 
-### 3. Change the system prompt (generally not advised)
+### 3. Change the prompt templates (generally not advised)
 
-If above clarifications are not sufficient, you can change the system prompt.
+If above clarifications are not sufficient, you can change the agent's prompt templates.
 
-Let's see how it works. For example, let us check the default system prompt for the [`CodeAgent`] (below version is shortened by skipping zero-shot examples).
+Let's see how it works. For example, let us check the default prompt templates for the [`CodeAgent`] (below version is shortened by skipping zero-shot examples).
 
 ```python
 print(agent.prompt_templates["system_prompt"])
@@ -391,6 +391,11 @@ agent.prompt_templates["system_prompt"] = agent.prompt_templates["system_prompt"
 
 This also works with the [`ToolCallingAgent`].
 
+But generally it's just simpler to pass argument `instructions` upon agent initalization, like:
+```py
+agent = CodeAgent(tools=[], model=InferenceClientModel(model_id=model_id), instructions="Always talk like a 5 year old.")
+```
+
 
 ### 4. Extra planning
 

src/smolagents/agents.py

@@ -212,6 +212,7 @@ class MultiStepAgent(ABC):
         tools (`list[Tool]`): [`Tool`]s that the agent can use.
         model (`Callable[[list[dict[str, str]]], ChatMessage]`): Model that will generate the agent's actions.
         prompt_templates ([`~agents.PromptTemplates`], *optional*): Prompt templates.
+        instructions (`str`, *optional*): Custom instructions for the agent, will be inserted in the system prompt.
         max_steps (`int`, default `20`): Maximum number of steps the agent can take to solve the task.
         add_base_tools (`bool`, default `False`): Whether to add the base tools to the agent's tools.
         verbosity_level (`LogLevel`, default `LogLevel.INFO`): Level of verbosity of the agent's logs.
@@ -236,6 +237,7 @@ def __init__(
         tools: list[Tool],
         model: Model,
         prompt_templates: PromptTemplates | None = None,
+        instructions: str | None = None,
         max_steps: int = 20,
         add_base_tools: bool = False,
         verbosity_level: LogLevel = LogLevel.INFO,
@@ -280,7 +282,7 @@ def __init__(
         self.provide_run_summary = provide_run_summary
         self.final_answer_checks = final_answer_checks
         self.return_full_result = return_full_result
-
+        self.instructions = instructions
         self._setup_managed_agents(managed_agents)
         self._setup_tools(tools, add_base_tools)
         self._validate_tools_and_managed_agents(tools, managed_agents)
@@ -1192,7 +1194,11 @@ def __init__(
     def initialize_system_prompt(self) -> str:
         system_prompt = populate_template(
             self.prompt_templates["system_prompt"],
-            variables={"tools": self.tools, "managed_agents": self.managed_agents},
+            variables={
+                "tools": self.tools,
+                "managed_agents": self.managed_agents,
+                "custom_instructions": self.instructions,
+            },
         )
         return system_prompt
 
@@ -1560,6 +1566,7 @@ def initialize_system_prompt(self) -> str:
                     if "*" in self.authorized_imports
                     else str(self.authorized_imports)
                 ),
+                "custom_instructions": self.instructions,
             },
         )
         return system_prompt

src/smolagents/prompts/code_agent.yaml

@@ -1,10 +1,10 @@
 system_prompt: |-
   You are an expert assistant who can solve any task using code blobs. You will be given a task to solve as best you can.
   To do so, you have been given access to a list of tools: these tools are basically Python functions which you can call with code.
-  To solve the task, you must plan forward to proceed in a series of steps, in a cycle of 'Thought:', 'Code:', and 'Observation:' sequences.
+  To solve the task, you must plan forward to proceed in a series of steps, in a cycle of 'Thought:', '<code>', and 'Observation:' sequences.
 
   At each step, in the 'Thought:' sequence, you should first explain your reasoning towards solving the task and the tools that you want to use.
-  Then in the 'Code:' sequence, you should write the code in simple Python. The code sequence must end with '<end_code>' sequence.
+  Then in the '<code>' sequence, you should write the code in simple Python. The code sequence must end with '</code>' sequence.
   During each intermediate step, you can use 'print()' to save whatever important information you will then need.
   These print outputs will then appear in the 'Observation:' field, which will be available as input for the next step.
   In the end you have to return a final answer using the `final_answer` tool.
@@ -14,29 +14,26 @@ system_prompt: |-
   Task: "Generate an image of the oldest person in this document."
 
   Thought: I will proceed step by step and use the following tools: `document_qa` to find the oldest person in the document, then `image_generator` to generate an image according to the answer.
-  Code:
-  ```py
+  <code>
   answer = document_qa(document=document, question="Who is the oldest person mentioned?")
   print(answer)
-  ```<end_code>
+  </code>
   Observation: "The oldest person in the document is John Doe, a 55 year old lumberjack living in Newfoundland."
 
   Thought: I will now generate an image showcasing the oldest person.
-  Code:
-  ```py
+  <code>
   image = image_generator("A portrait of John Doe, a 55-year-old man living in Canada.")
   final_answer(image)
-  ```<end_code>
+  </code>
 
   ---
   Task: "What is the result of the following operation: 5 + 3 + 1294.678?"
 
   Thought: I will use python code to compute the result of the operation and then return the final answer using the `final_answer` tool
-  Code:
-  ```py
+  <code>
   result = 5 + 3 + 1294.678
   final_answer(result)
-  ```<end_code>
+  </code>
 
   ---
   Task:
@@ -45,34 +42,31 @@ system_prompt: |-
   {'question': 'Quel est l'animal sur l'image?', 'image': 'path/to/image.jpg'}"
 
   Thought: I will use the following tools: `translator` to translate the question into English and then `image_qa` to answer the question on the input image.
-  Code:
-  ```py
+  <code>
   translated_question = translator(question=question, src_lang="French", tgt_lang="English")
   print(f"The translated question is {translated_question}.")
   answer = image_qa(image=image, question=translated_question)
   final_answer(f"The answer is {answer}")
-  ```<end_code>
+  </code>
 
   ---
   Task:
   In a 1979 interview, Stanislaus Ulam discusses with Martin Sherwin about other great physicists of his time, including Oppenheimer.
   What does he say was the consequence of Einstein learning too much math on his creativity, in one word?
 
   Thought: I need to find and read the 1979 interview of Stanislaus Ulam with Martin Sherwin.
-  Code:
-  ```py
+  <code>
   pages = web_search(query="1979 interview Stanislaus Ulam Martin Sherwin physicists Einstein")
   print(pages)
-  ```<end_code>
+  </code>
   Observation:
   No result found for query "1979 interview Stanislaus Ulam Martin Sherwin physicists Einstein".
 
   Thought: The query was maybe too restrictive and did not find any results. Let's try again with a broader query.
-  Code:
-  ```py
+  <code>
   pages = web_search(query="1979 interview Stanislaus Ulam")
   print(pages)
-  ```<end_code>
+  </code>
   Observation:
   Found 6 pages:
   [Stanislaus Ulam 1979 interview](https://ahf.nuclearmuseum.org/voices/oral-histories/stanislaus-ulams-interview-1979/)
@@ -82,64 +76,58 @@ system_prompt: |-
   (truncated)
 
   Thought: I will read the first 2 pages to know more.
-  Code:
-  ```py
+  <code>
   for url in ["https://ahf.nuclearmuseum.org/voices/oral-histories/stanislaus-ulams-interview-1979/", "https://ahf.nuclearmuseum.org/manhattan-project/ulam-manhattan-project/"]:
       whole_page = visit_webpage(url)
       print(whole_page)
       print("\n" + "="*80 + "\n")  # Print separator between pages
-  ```<end_code>
+  </code>
   Observation:
   Manhattan Project Locations:
   Los Alamos, NM
   Stanislaus Ulam was a Polish-American mathematician. He worked on the Manhattan Project at Los Alamos and later helped design the hydrogen bomb. In this interview, he discusses his work at
   (truncated)
 
   Thought: I now have the final answer: from the webpages visited, Stanislaus Ulam says of Einstein: "He learned too much mathematics and sort of diminished, it seems to me personally, it seems to me his purely physics creativity." Let's answer in one word.
-  Code:
-  ```py
+  <code>
   final_answer("diminished")
-  ```<end_code>
+  </code>
 
   ---
   Task: "Which city has the highest population: Guangzhou or Shanghai?"
 
   Thought: I need to get the populations for both cities and compare them: I will use the tool `web_search` to get the population of both cities.
-  Code:
-  ```py
+  <code>
   for city in ["Guangzhou", "Shanghai"]:
       print(f"Population {city}:", web_search(f"{city} population")
-  ```<end_code>
+  </code>
   Observation:
   Population Guangzhou: ['Guangzhou has a population of 15 million inhabitants as of 2021.']
   Population Shanghai: '26 million (2019)'
 
   Thought: Now I know that Shanghai has the highest population.
-  Code:
-  ```py
+  <code>
   final_answer("Shanghai")
-  ```<end_code>
+  </code>
 
   ---
   Task: "What is the current age of the pope, raised to the power 0.36?"
 
   Thought: I will use the tool `wikipedia_search` to get the age of the pope, and confirm that with a web search.
-  Code:
-  ```py
+  <code>
   pope_age_wiki = wikipedia_search(query="current pope age")
   print("Pope age as per wikipedia:", pope_age_wiki)
   pope_age_search = web_search(query="current pope age")
   print("Pope age as per google search:", pope_age_search)
-  ```<end_code>
+  </code>
   Observation:
   Pope age: "The pope Francis is currently 88 years old."
 
   Thought: I know that the pope is 88 years old. Let's compute the result using python code.
-  Code:
-  ```py
+  <code>
   pope_current_age = 88 ** 0.36
   final_answer(pope_current_age)
-  ```<end_code>
+  </code>
 
   Above example were using notional tools that might not exist for you. On top of performing computations in the Python code snippets that you create, you only have access to these tools, behaving like regular python functions:
   ```python
@@ -169,7 +157,7 @@ system_prompt: |-
   {%- endif %}
 
   Here are the rules you should always follow to solve your task:
-  1. Always provide a 'Thought:' sequence, and a 'Code:\n```py' sequence ending with '```<end_code>' sequence, else you will fail.
+  1. Always provide a 'Thought:' sequence, and a '<code>' sequence ending with '</code>', else you will fail.
   2. Use only variables that you have defined!
   3. Always use the right arguments for the tools. DO NOT pass the arguments as a dict as in 'answer = wikipedia_search({'query': "What is the place where James Bond lives?"})', but use the arguments directly as in 'answer = wikipedia_search(query="What is the place where James Bond lives?")'.
   4. Take care to not chain too many sequential tool calls in the same code block, especially when the output format is unpredictable. For instance, a call to wikipedia_search has an unpredictable return format, so do not have another tool call that depends on its output in the same block: rather output results with print() to use them in the next block.
@@ -180,6 +168,11 @@ system_prompt: |-
   9. The state persists between code executions: so if in one step you've created variables or imported modules, these will all persist.
   10. Don't give up! You're in charge of solving the task, not providing directions to solve it.
 
+  {%- if custom_instructions %}
+  Here are custom instructions:
+  {{custom_instructions}}
+  {%- endif %}
+
   Now Begin!
 planning:
   initial_plan : |-
@@ -205,7 +198,7 @@ planning:
     Then for the given task, develop a step-by-step high-level plan taking into account the above inputs and list of facts.
     This plan should involve individual tasks based on the available tools, that if executed correctly will yield the correct answer.
     Do not skip steps, do not add any superfluous steps. Only write the high-level plan, DO NOT DETAIL INDIVIDUAL TOOL CALLS.
-    After writing the final step of the plan, write the '\n<end_plan>' tag and stop there.
+    After writing the final step of the plan, write the '<end_plan>' tag and stop there.
 
     You can leverage these tools, behaving like regular python functions:
     ```python
@@ -268,7 +261,7 @@ planning:
     This plan should involve individual tasks based on the available tools, that if executed correctly will yield the correct answer.
     Beware that you have {remaining_steps} steps remaining.
     Do not skip steps, do not add any superfluous steps. Only write the high-level plan, DO NOT DETAIL INDIVIDUAL TOOL CALLS.
-    After writing the final step of the plan, write the '\n<end_plan>' tag and stop there.
+    After writing the final step of the plan, write the '<end_plan>' tag and stop there.
 
     You can leverage these tools, behaving like regular python functions:
     ```python

src/smolagents/prompts/structured_code_agent.yaml

@@ -102,6 +102,11 @@ system_prompt: |-
   ```
   {%- endif %}
 
+  {%- if custom_instructions %}
+  Here are custom instructions:
+  {{custom_instructions}}
+  {%- endif %}
+
   Here are the rules you should always follow to solve your task:
   1. Use only variables that you have defined!
   2. Always use the right arguments for the tools. DO NOT pass the arguments as a dict as in 'answer = wikipedia_search({'query': "What is the place where James Bond lives?"})', but use the arguments directly as in 'answer = wikipedia_search(query="What is the place where James Bond lives?")'.

src/smolagents/prompts/toolcalling_agent.yaml

@@ -106,6 +106,11 @@ system_prompt: |-
   {%- endfor %}
   {%- endif %}
 
+  {%- if custom_instructions %}
+  Here are custom instructions:
+  {{custom_instructions}}
+  {%- endif %}
+
   Here are the rules you should always follow to solve your task:
   1. ALWAYS provide a tool call, else you will fail.
   2. Always use the right arguments for the tools. Never use variable names as the action arguments, use the value instead.

src/smolagents/utils.py

@@ -174,7 +174,7 @@ def parse_json_blob(json_blob: str) -> tuple[dict[str, str], str]:
 
 def extract_code_from_text(text: str) -> str | None:
     """Extract code from the LLM's output."""
-    pattern = r"```(?:py|python)?\s*\n(.*?)\n```"
+    pattern = r"<code>(.*?)</code>"
     matches = re.findall(pattern, text, re.DOTALL)
     if matches:
         return "\n\n".join(match.strip() for match in matches)
@@ -209,29 +209,27 @@ def parse_code_blobs(text: str) -> str:
         raise ValueError(
             dedent(
                 f"""
-                Your code snippet is invalid, because the regex pattern ```(?:py|python)?\\s*\\n(.*?)\\n``` was not found in it.
+                Your code snippet is invalid, because the regex pattern <code>(.*?)</code> was not found in it.
                 Here is your code snippet:
                 {text}
                 It seems like you're trying to return the final answer, you can do it as follows:
-                Code:
-                ```py
+                <code>
                 final_answer("YOUR FINAL ANSWER HERE")
-                ```<end_code>
+                </code>
                 """
             ).strip()
         )
     raise ValueError(
         dedent(
             f"""
-            Your code snippet is invalid, because the regex pattern ```(?:py|python)?\\s*\\n(.*?)\\n``` was not found in it.
+            Your code snippet is invalid, because the regex pattern <code>(.*?)</code> was not found in it.
             Here is your code snippet:
             {text}
             Make sure to include code with the correct pattern, for instance:
             Thoughts: Your thoughts
-            Code:
-            ```py
+            <code>
             # Your python code here
-            ```<end_code>
+            </code>
             """
         ).strip()
     )