revert gaia prompts to the main version

examples/gaia_agent/prompts.py

@@ -1,82 +1,35 @@
-SYSTEM_PROMPT = """You are an expert AI Agent trained to assist users with complex information processing tasks.
+SYSTEM_PROMPT = """You are an expert AI Agent trained to assist user with complex information processing tasks.
 Your role is to understand user queries and respond in a helpful and accurate manner.
 Keep your replies concise and direct. Prioritize clarity and avoid over-elaboration.
-Do not express emotions or opinions about user questions."""
+Do not express your emotions or opinions about the user question."""
 
-FORMAT = """
-Output only a single JSON dict or a single JSON list. DO NOT OUTPUT ANYTHING BESIDES THE JSON! DO NOT PLACE ANY COMMENTS INSIDE THE JSON. It will break the system that processes the output.
-"""
-
-PLAN = f"""Write a concise multi-step plan explaining which steps should be performed to find the answer for the given task.
-Remember that you can use web search, browser, python code execution and access the youtube videos to reach your goals.
-Be specific about how each step should be performed. Only describe the intended actions here, do not perform them yet.
-Consider that next steps may depend on results of previous steps, so include conditional branching using "if" statements where needed.
-Respond with the thought kind="plan_thought". {FORMAT}"""
-
-START = f"""Let's start executing the plan step by step, using the allowed steps described earlier. {FORMAT}"""
-
-REFLECT_AND_ACT = f"""
-Produce a JSON list with two steps in it.
-The first step should be the reasoning thought produced according to the following rules:
-    1. Summarize the last observation.
-    2. If the last action interacted with the page, describe how it affected its content.
-    3. If the last observation is a webpage, explain the observed effect of the last action on the page.
-    4. Check if the action led to the desired outcome. Then explain its effect on the task and the plan.
-    5. If there are any errors, describe them and suggest reasons for the error and possible solutions.
-    6. Produce a list of things to do to accomplish the current step of the plan.
-    7. After that, propose the immediate next step.
-    8. If you see a cookie consent form, accept it first.
-    9. Quote the relevant part of the observation verbatim when the action depends on it, for example when interacting with the page.
-    10. Do not hallucinate unseen elements or results; only use the information from the observation.
-    11. If the web search snippets do not contain the information you are looking for, always inspect the most relevant search result.
-    12. Do not overuse the web search, you can use it only a few times.
-    13. If you cannot see some part of the element on the page, always scroll the page.
-    14. If the intended element is not visible on the page, scroll the page!
-    15. Always close all popups before interacting with the newly loaded page!
-The second step should be an action that performs the proposed step.
-Remember that your response should be a list and not a dictionary.
-{FORMAT}"""
+short_format_instruction = (
+    "DO NOT OUTPUT ANYTHING BESIDES THE JSON. It will break the system that processes the output."
+)
 
-VERIFY = f"""
-Produce a JSON list with two steps in it.
-The first step should be the reasoning thought produced according to the following rules:
-    - Recite the task and mention all the restrictions and requirements for the final answer found in the task description.
-    - Check if the task expected to have singular or multiple answers.
-    - Verify that the answer meets all following requirements from this list.
-    - Verify that the answer should be a number OR as few words as possible OR a comma-separated list of numbers and/or strings.
-    - Verify that the answer MUST follow any formatting instructions or requirements specified in the original task (e.g., alphabetization, sequencing, units, rounding, decimal places, etc.)
-    - If asked for a number, express it numerically, don't use commas, do not add anything after the number, don't include units such as $ or percent signs unless specified otherwise in the question.
-    - If asked for a string, don't use articles or abbreviations (e.g. for cities), unless specified otherwise. Don't output any final sentence punctuation such as '.', '!', or '?'.
-    - If asked for a comma-separated list, apply the above rules depending on whether the elements are numbers or strings.
-The second step should be an verified_answer.
-Remember that your response should be a list and not a dictionary.
-{FORMAT}"""
+PLAN = f'What steps should I do to answer the question above? Be specific about how each step should be done. Respond with the thought kind="plan_thought". {short_format_instruction}'
+START = f"""Let's start executing the plan step by step, using allowed steps described earlier. {short_format_instruction}"""
+THINK_AFTER_OBSERVATION = f""""Lets think step by step about the observation, how it affects the plan and what should be done next. {short_format_instruction}"""
 
 ALLOWED_STEPS = """
-You are allowed to produce ONLY steps with the following JSON schemas:
-{allowed_steps}
-Do not reproduce the schema when producing steps; use it as a reference.
-"""
-
-ACT_STEPS = """
-You can use the following tools: search the web, read web pages or documents, python code for computations and modeling, and reasoning.
-You are allowed to produce ONLY steps with the following JSON schemas:
+You can use the following tools: search the web, read web page or document, python code for computations and modeling, and reasoning.
+You are allowed to produce ONLY steps with the following json schemas:
 {allowed_steps}
-Do not reproduce the schema when producing steps; use it as a reference.
+Do not reproduce schema when producing the steps, use it as a reference.
 """
 
 ALLOWED_STEPS_CODE = """
-You can use the following tools: search the web, read web pages or documents, python code, and reasoning.
-You are allowed to produce ONLY steps with the following JSON schemas:
+You can use the following tools: search the web, read web page or document, python code, and reasoning.
+You are allowed to produce ONLY steps with the following json schemas:
 {allowed_steps}
-If you want to create a python code step, output a python code block inside backticks, like this:
+If you want to create a python code step, just output python code block inside the backticks, like this:
 ```python
 a = 1
 result = a*2
 print(result)
 ```
 Note that the last code line should print the result.
-Do not reproduce the schema when producing steps; use it as a reference.
+Do not reproduce schema when producing the steps, use it as a reference.
 """
 
 FACTS_SURVEY = f"""Before we begin executing the plan, please answer the following pre-survey to the best of your ability.
@@ -94,6 +47,6 @@
     3. Facts to derive
     4. Educated guesses
 
-Respond with facts_survey_thought! Strictly follow the provided JSON schema of the step, do not replicate the schema fields itself!
-{FORMAT}
+Respond with facts_survey_thought.
+{short_format_instruction}
 """