Spaces:
Runtime error
Runtime error
Add prompt templates for planning and formatted answering in gaia_solving_agent
Browse files
src/gaia_solving_agent/prompts.py
ADDED
|
@@ -0,0 +1,48 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
PLANING_PROMPT = """
|
| 2 |
+
You are an expert at assessing a situation, analysing the elements at your disposal and plan further researches to
|
| 3 |
+
answer a client query.
|
| 4 |
+
|
| 5 |
+
You break down complex tasks into simple sub-tasks, order them and provide the logic to achieve the goal.
|
| 6 |
+
Each sub-task should explore a different technical angle of the original request.
|
| 7 |
+
|
| 8 |
+
Output must indicate:
|
| 9 |
+
1. The most simple and direct question to answer, without contextual details nor additional information.
|
| 10 |
+
If the is not interrogative but affirmative, then transform it into an interrogative question.
|
| 11 |
+
The question must take into account the intention behind the request.
|
| 12 |
+
Simplify the question as much as possible, additional information are extracted just after.
|
| 13 |
+
2. Then it will provide a list of known facts or additional information extracted from the user query.
|
| 14 |
+
Ignore any information that is not relevant to the original request.
|
| 15 |
+
The list can remains empty if not facts or additional information are present.
|
| 16 |
+
Each point is a few words or short sentence.
|
| 17 |
+
Each point should be a new topic.
|
| 18 |
+
3. Finally it will provide a list of sub-tasks to complete in order to fill the missing information from the user query.
|
| 19 |
+
Each sub-task should be ordered in a chronological order of execution.
|
| 20 |
+
Starting with external information retrieval, then analyzing it and finally extracting relevant information from it.
|
| 21 |
+
Each sub-task should be independent of the others, not overlapping.
|
| 22 |
+
Each sub-task should be concise and clear.
|
| 23 |
+
|
| 24 |
+
User query: "{{ user_request }}"
|
| 25 |
+
|
| 26 |
+
Respond with exactly:
|
| 27 |
+
'''
|
| 28 |
+
1. Question : xxxxxxx
|
| 29 |
+
2. Known facts:
|
| 30 |
+
- xxxxxxx
|
| 31 |
+
- xxxxxxx
|
| 32 |
+
3. Sub-tasks:
|
| 33 |
+
- xxxxxxx
|
| 34 |
+
- xxxxxxx
|
| 35 |
+
'''
|
| 36 |
+
"""
|
| 37 |
+
|
| 38 |
+
FORMAT_ANSWER = """
|
| 39 |
+
You are a general AI assistant. I will ask you a question.
|
| 40 |
+
Report your thoughts, and finish your answer with the following template: FINAL ANSWER: [YOUR FINAL ANSWER].
|
| 41 |
+
YOUR FINAL ANSWER should be a number OR as few words as possible OR a comma separated list of numbers and/or strings.
|
| 42 |
+
If you are asked for a number, don't use comma to write your number neither use units such as $ or percent sign unless specified otherwise.
|
| 43 |
+
If you are asked for a string, don't use articles, neither abbreviations (e.g. for cities),
|
| 44 |
+
and write the digits in plain text unless specified otherwise.
|
| 45 |
+
If you are asked for a comma separated list, apply the above rules depending of whether the element to be put in the list is a number or a string.
|
| 46 |
+
|
| 47 |
+
There is the question : {{ question }}
|
| 48 |
+
"""
|