프롬프트 구조화 방식은 프롬프트의 정보를 파싱하는 모델의 기능에 영향을 줄 수 있습니다.
또한 모델이 제공된 정보를 사용하는 방법을 올바르게 해석하는 데 도움이 됩니다. 프롬프트를 구조화하기 위해 프리픽스 또는 XML 태그를 사용하여 프롬프트의 여러 부분 또는 구성요소를 구분할 수 있습니다.
프리픽스를 사용하여 간단한 프롬프트 구조화
프리픽스는 프롬프트의 정보에 라벨을 지정하기 위해 콜론이 뒤에 오는 단어 또는 구문입니다. 다음 예시에서 'TASK:', 'CLASSES:', 'OBJECTS:'는 프리픽스입니다.
TASK:
Classify the OBJECTS.
CLASSES:
- Large
- Small
OBJECTS:
- Rhino
- Mouse
- Snail
- Elephant
- Rhino: Large
- Mouse: Small
- Snail: Small
- Elephant: Large
XML 및 기타 구분 기호를 사용하여 복잡한 프롬프트 구조화
복잡한 프롬프트의 경우 XML 및 기타 구분 기호를 사용하여 프롬프트의 구성요소를 구분합니다. 복잡하고 긴 프롬프트 구성요소에 대해 BEGIN 및 END 또는 {} 섹션 구분 기호를 사용하여 실제 요청 사항과 명확하게 구분할 수 있습니다.
You are a chatbot agent answering customer's questions in a chat.
Your task is to answer the customer's question using the data provided in the <DATA> section.
- You can access order history in the <ORDERS> section including email id and order total
with payment summary.
- Refer to <ORDERLINES> for item level details within each order in <ORDERS>.
Today is 2024-01-29
<DATA>
<ORDERS>
{OrderId|CustomerEmail|CreatedTimestamp|IsCancelled|OrderTotal|PaymentSummary
CC10182|222larabrown@gmail.com|2024-01-19|true|0.0|Not available
CC10183|baklavainthebalkans@gmail.com|2024-01-19|true|0.0|Not available}
{...}
...
</ORDERS>
<ORDERLINES>
OrderId|OrderLineId|CreatedTimestamp|ItemDescription|Quantity|FulfillmentStatus|ExpectedDeliveryDate
|ActualDeliveryDate|ActualShipDate|ExpectedShipDate|TrackingInformation|ShipToAddress|CarrierCode|De
liveryMethod|UnitPrice|OrderLineSubTotal|LineShippingCharge|TotalTaxes|Payments CC10182|1||Shorts|0.
0|unshipped|2024-01-31|2024-02-01|2024-01-30|2024-01-29||||ShipToAddress|115.99|0.0|0.0|0.0|
...
</ORDERLINES>
</DATA>
<INSTRUCTIONS>
- If there is no data that can help answer the question, respond with "I do not have this
information. Please contact customer service".
- You are allowed to ask a follow up question if it will help narrow down the data row customer may
be referring to.
- You can only answer questions related to order history and amount charged for it. Include OrderId
in the response, when applicable.
- For everything else, please redirect to the customer service agent.
- Answer in plain English and no sources are required
- Chat with the customer so far is under the CHAT section.
</INSTRUCTIONS>
QUESTION: How much did I pay for my last order?
ANSWER:
[[["이해하기 쉬움","easyToUnderstand","thumb-up"],["문제가 해결됨","solvedMyProblem","thumb-up"],["기타","otherUp","thumb-up"]],[["이해하기 어려움","hardToUnderstand","thumb-down"],["잘못된 정보 또는 샘플 코드","incorrectInformationOrSampleCode","thumb-down"],["필요한 정보/샘플이 없음","missingTheInformationSamplesINeed","thumb-down"],["번역 문제","translationIssue","thumb-down"],["기타","otherDown","thumb-down"]],["최종 업데이트: 2025-09-04(UTC)"],[],[],null,["# Structure prompts\n\nHow you structure a prompt can affect the model's ability to parse the information in the prompt.\nIt also helps the model correctly interpret how to use the given information. To give structure to a\nprompt, you can use prefixes or XML tags to delimit different parts or components of a prompt.\n\nUse prefixes to structure simple prompts\n----------------------------------------\n\nA prefix is a word or phrase followed by a colon to label the information in a prompt. In the\nfollowing example, \"TASK:, \"CLASSES:,\" and \"OBJECTS:\" are the prefixes.\n\nUse XML and other delimiters to structure complex prompts\n---------------------------------------------------------\n\nFor complex prompts, use XML and other delimiters to separate components of a prompt. You can use\n`BEGIN` and `END` or `{}` section delimiters for complex and\nlengthy prompt components to clearly distinguish them from the actual instructions. \n\n```\nYou are a chatbot agent answering customer's questions in a chat.\nYour task is to answer the customer's question using the data provided in the \u003cDATA\u003e section.\n - You can access order history in the \u003cORDERS\u003e section including email id and order total\n with payment summary.\n - Refer to \u003cORDERLINES\u003e for item level details within each order in \u003cORDERS\u003e.\n\nToday is 2024-01-29\n\n\u003cDATA\u003e\n\u003cORDERS\u003e\n{OrderId|CustomerEmail|CreatedTimestamp|IsCancelled|OrderTotal|PaymentSummary\nCC10182|222larabrown@gmail.com|2024-01-19|true|0.0|Not available\nCC10183|baklavainthebalkans@gmail.com|2024-01-19|true|0.0|Not available}\n{...}\n...\n\u003c/ORDERS\u003e\n\n\u003cORDERLINES\u003e\nOrderId|OrderLineId|CreatedTimestamp|ItemDescription|Quantity|FulfillmentStatus|ExpectedDeliveryDate\n|ActualDeliveryDate|ActualShipDate|ExpectedShipDate|TrackingInformation|ShipToAddress|CarrierCode|De\nliveryMethod|UnitPrice|OrderLineSubTotal|LineShippingCharge|TotalTaxes|Payments CC10182|1||Shorts|0.\n0|unshipped|2024-01-31|2024-02-01|2024-01-30|2024-01-29||||ShipToAddress|115.99|0.0|0.0|0.0|\n...\n\u003c/ORDERLINES\u003e\n\u003c/DATA\u003e\n\n\u003cINSTRUCTIONS\u003e\n- If there is no data that can help answer the question, respond with \"I do not have this\n information. Please contact customer service\".\n- You are allowed to ask a follow up question if it will help narrow down the data row customer may\n be referring to.\n- You can only answer questions related to order history and amount charged for it. Include OrderId\n in the response, when applicable.\n- For everything else, please redirect to the customer service agent. \n- Answer in plain English and no sources are required\n- Chat with the customer so far is under the CHAT section.\n\u003c/INSTRUCTIONS\u003e\n\nQUESTION: How much did I pay for my last order?\nANSWER:\n```\n\nWhat's next\n-----------\n\n- Explore more examples of prompts in the [Prompt gallery](/vertex-ai/generative-ai/docs/prompt-gallery)."]]