Refactor code agent system prompt #1153
Conversation
2f30c4e to
3113c0f
Compare
| @@ -210,6 +201,16 @@ def __init__( | |||
| self.agent_name = self.__class__.__name__ | |||
| self.model = model | |||
| self.prompt_templates = prompt_templates or EMPTY_PROMPT_TEMPLATES | |||
| if prompt_templates is not None: | |||
There was a problem hiding this comment.
These asserts should guide users trying to customize their agent's prompts.
| @@ -113,18 +113,15 @@ system_prompt: |- | |||
| If no tool call is needed, use final_answer tool to return your answer. | |||
| 4. Never re-do a tool call that you previously did with the exact same parameters. | |||
| Now Begin! If you solve the task correctly, you will receive a reward of $1,000,000. | |||
There was a problem hiding this comment.
Removing the monetary reward because putting an objective like this can help exploits.
| @@ -158,7 +158,7 @@ def parse_json_blob(json_blob: str) -> Tuple[Dict[str, str], str]: | |||
| json_data = json.loads(json_data, strict=False) | |||
| return json_data, json_blob[:first_accolade_index] | |||
| except IndexError: | |||
| raise ValueError("The JSON blob you used is invalid") | |||
| raise ValueError("The model output does not contain any JSON blob.") | |||
There was a problem hiding this comment.
Clarify the error.
| Takes inputs: {{tool.inputs}} | ||
| Returns an output of type: {{tool.output_type}} | ||
| {%- endfor %} | ||
| def {{ tool.name }}({% for arg_name, arg_info in tool.inputs.items() %}{{ arg_name }}: {{ arg_info.type }}{% if not loop.last %}, {% endif %}{% endfor %}) -> {{tool.output_type}}: |
There was a problem hiding this comment.
Make tool definitions look like python functions: should be easier to read for the model.
| initial_plan="", | ||
| update_facts_pre_messages="", | ||
| update_facts_post_messages="", | ||
| update_plan_pre_messages="", | ||
| update_plan_post_messages="", |
There was a problem hiding this comment.
NIT: In a future PR, it would be great to migrate PlanningPromptTemplate, ManagedAgentPromptTemplate, FinalAnswerPromptTemplate, and PromptTemplates to pydantic.BaseModel to enable default initialization. Ultimately, this will allow for a simpler declaration:
EMPTY_PROMPT_TEMPLATES = PromptTemplates()| missing_keys = set(EMPTY_PROMPT_TEMPLATES.keys()) - set(prompt_templates.keys()) | ||
| assert not missing_keys, ( | ||
| f"Some prompt templates are missing from your custom `prompt_templates`: {missing_keys}" | ||
| ) | ||
| for key in EMPTY_PROMPT_TEMPLATES.keys(): | ||
| for subkey in EMPTY_PROMPT_TEMPLATES[key]: | ||
| assert subkey in prompt_templates[key], ( | ||
| f"Some prompt templates are missing from your custom `prompt_templates`: {subkey} under {key}" | ||
| ) |
There was a problem hiding this comment.
fyi: A Pydantic PromptTemplates model enables automatic validation. The code could simply be:
PromptTemplates(**prompt_templates) # It raises an error if any field or type doesn't match `PromptTemplates`Alternatively, the function could accept only a PromptTemplates BaseModel instance as an argument.
There was a problem hiding this comment.
It will also make the EMPTY_PROMPT_TEMPLATES variable unnecessary.
There was a problem hiding this comment.
@albertvillanova pointer for discussion when you come back.
| @@ -113,18 +113,15 @@ system_prompt: |- | |||
| If no tool call is needed, use final_answer tool to return your answer. | |||
| 4. Never re-do a tool call that you previously did with the exact same parameters. | |||
| Now Begin! If you solve the task correctly, you will receive a reward of $1,000,000. | |||
Co-authored-by: A-Mahla <89754740+A-Mahla@users.noreply.github.com>
No description provided.