用额外提示词辅助解析

自定义提示词让你像给 LLM 下指令一样引导 Parse 的智能体模型。

这些提示词在下列场景中很有用：提升解析器在复杂版式上的表现、按特定格式抽取数据，或对文档做其他形式的转换。

本例演示如何给 Parse 提供附加指令（提示词）来塑造 LLM 解析非结构化文档的方式。我们拿一张麦当劳小票，展示如何忽略文档的某些部分，只解析每样商品的单价与最终应付金额。

设置

把 LlamaCloud API key 设好，SDK 会自动读取：

终端窗口

export LLAMA_CLOUD_API_KEY="llx-..."

不加指令地解析小票

这个例子用的麦当劳小票如下：

把它下载为 mcdonalds_receipt.png：

终端窗口

curl -fL -o mcdonalds_receipt.png https://raw.githubusercontent.com/run-llama/llama_cloud_services/main/examples/parse/parsing_instructions/mcdonalds_receipt.png

先在不加任何特殊指令的情况下解析这张小票 —— 就一次默认的智能体解析：

Python
TypeScript
Go
Java
CLI

终端窗口

pip install "llama-cloud>=2.8"

from llama_cloud import LlamaCloud

client = LlamaCloud()  # reads LLAMA_CLOUD_API_KEY from the environment

# Upload the receipt
file = client.files.create(file="mcdonalds_receipt.png", purpose="parse")

# Parse with no special instructions
vanilla_result = client.parsing.parse(
    file_id=file.id,
    tier="agentic",
    version="latest",
    expand=["markdown"],
    output_options={
        "markdown": {"tables": {"output_tables_as_markdown": True}},
    },
)

print(vanilla_result.markdown.pages[0].markdown)

终端窗口

npm install @llamaindex/llama-cloud

import LlamaCloud from '@llamaindex/llama-cloud';
import fs from 'fs';

const client = new LlamaCloud(); // reads LLAMA_CLOUD_API_KEY from the environment

// Upload the receipt
const file = await client.files.create({
  file: fs.createReadStream('mcdonalds_receipt.png'),
  purpose: 'parse',
});

// Parse with no special instructions
const vanillaResult = await client.parsing.parse({
  file_id: file.id,
  tier: 'agentic',
  version: 'latest',
  expand: ['markdown'],
  output_options: {
    markdown: { tables: { output_tables_as_markdown: true } },
  },
});

console.log(vanillaResult.markdown.pages[0].markdown);

终端窗口

go get github.com/run-llama/llama-parse-go

package main

import (
  "context"
  "fmt"
  "log"
  "os"
  "time"

  llamacloud "github.com/run-llama/llama-parse-go"
)

func main() {
  ctx := context.Background()
  client := llamacloud.NewClient() // reads LLAMA_CLOUD_API_KEY from the environment

  f, err := os.Open("mcdonalds_receipt.png")
  if err != nil {
    log.Fatal(err)
  }
  defer f.Close()

  // Upload the receipt
  file, err := client.Files.New(ctx, llamacloud.FileNewParams{
    File:    f,
    Purpose: "parse",
  })
  if err != nil {
    log.Fatal(err)
  }

  // Parse with no special instructions
  job, err := client.Parsing.New(ctx, llamacloud.ParsingNewParams{
    FileID:  llamacloud.String(file.ID),
    Tier:    llamacloud.ParsingNewParamsTierAgentic,
    Version: llamacloud.ParsingNewParamsVersionLatest,
    OutputOptions: llamacloud.ParsingNewParamsOutputOptions{
      Markdown: llamacloud.ParsingNewParamsOutputOptionsMarkdown{
        Tables: llamacloud.ParsingNewParamsOutputOptionsMarkdownTables{
          OutputTablesAsMarkdown: llamacloud.Bool(true),
        },
      },
    },
  })
  if err != nil {
    log.Fatal(err)
  }

  // expand is a GET parameter — request markdown when you fetch the result
  getParams := llamacloud.ParsingGetParams{Expand: []string{"markdown"}}
  result, err := client.Parsing.Get(ctx, job.ID, getParams)
  if err != nil {
    log.Fatal(err)
  }
  for result.Job.Status != "COMPLETED" && result.Job.Status != "FAILED" && result.Job.Status != "CANCELLED" {
    time.Sleep(2 * time.Second)
    result, err = client.Parsing.Get(ctx, job.ID, getParams)
    if err != nil {
      log.Fatal(err)
    }
  }
  if result.Job.Status != "COMPLETED" {
    log.Fatalf("parse ended as %s", result.Job.Status)
  }

  if len(result.Markdown.Pages) == 0 {
    log.Fatal("parse completed with no markdown pages")
  }

  fmt.Println(result.Markdown.Pages[0].Markdown)
}

implementation("ai.llamaindex:llama-cloud:1.3.0")

import ai.llamaindex.llamacloud.client.LlamaCloudClient;
import ai.llamaindex.llamacloud.client.okhttp.LlamaCloudOkHttpClient;
import ai.llamaindex.llamacloud.models.files.FileCreateParams;
import ai.llamaindex.llamacloud.models.files.FileCreateResponse;
import ai.llamaindex.llamacloud.models.parsing.ParsingCreateParams;
import ai.llamaindex.llamacloud.models.parsing.ParsingCreateResponse;
import ai.llamaindex.llamacloud.models.parsing.ParsingGetParams;
import ai.llamaindex.llamacloud.models.parsing.ParsingGetResponse;
import java.nio.file.Paths;

public class ParseReceipt {
    public static void main(String[] args) throws Exception {
        // reads LLAMA_CLOUD_API_KEY from the environment
        LlamaCloudClient client = LlamaCloudOkHttpClient.fromEnv();

        // Upload the receipt
        FileCreateResponse file = client.files().create(FileCreateParams.builder()
                .file(Paths.get("mcdonalds_receipt.png"))
                .purpose("parse")
                .build());

        // Parse with no special instructions
        ParsingCreateResponse job = client.parsing().create(ParsingCreateParams.builder()
                .fileId(file.id())
                .tier(ParsingCreateParams.Tier.AGENTIC)
                .version(ParsingCreateParams.Version.LATEST)
                .outputOptions(ParsingCreateParams.OutputOptions.builder()
                        .markdown(ParsingCreateParams.OutputOptions.Markdown.builder()
                                .tables(ParsingCreateParams.OutputOptions.Markdown.Tables.builder()
                                        .outputTablesAsMarkdown(true)
                                        .build())
                                .build())
                        .build())
                .build());

        // expand is a query parameter — request markdown when you fetch the result
        ParsingGetParams getParams = ParsingGetParams.builder()
                .jobId(job.id())
                .addExpand("markdown")
                .build();

        ParsingGetResponse result = client.parsing().get(getParams);
        while (!result.job().status().equals(ParsingGetResponse.Job.Status.COMPLETED)
                && !result.job().status().equals(ParsingGetResponse.Job.Status.FAILED)
                && !result.job().status().equals(ParsingGetResponse.Job.Status.CANCELLED)) {
            Thread.sleep(2000);
            result = client.parsing().get(getParams);
        }
        if (!result.job().status().equals(ParsingGetResponse.Job.Status.COMPLETED)) {
            throw new RuntimeException("parse ended as " + result.job().status());
        }

        System.out.println(result.markdown().get().pages().get(0).asMarkdownResult().markdown());
    }
}

终端窗口

go install github.com/run-llama/llama-parse-cli/cmd/llp@latest

终端窗口

# Upload the receipt
FILE_ID=$(llp files create \
  --file mcdonalds_receipt.png \
  --purpose parse | jq -r '.id')

# Start a parse job with no special instructions
JOB_ID=$(llp parsing create \
  --file-id "$FILE_ID" \
  --tier agentic \
  --version latest \
  --output-options.markdown '{tables: {output_tables_as_markdown: true}}' | jq -r '.id')

# Poll until the job reaches a terminal status
while true; do
  STATUS=$(llp parsing get --job-id "$JOB_ID" | jq -er '.job.status') || break
  case "$STATUS" in COMPLETED|FAILED|CANCELLED) break ;; esac
  sleep 2
done

# Print the markdown for the first page
llp parsing get --job-id "$JOB_ID" --expand markdown \
  | jq -r 'if .job.status == "COMPLETED" then .markdown.pages[0].markdown
    else "Parse job failed: \(.job.error_message // .)" end'

结果是 Parse 从小票还原出的完整 markdown：

> Rate us HIGHLY SATISFIED and
> Receive ONE FREE ITEM
> Purchase any sandwich and receive an item of equal or lesser value
> Go to www.mcdvoice.com within 7 days and tell us about your visit.
> Validation Code:
> Expires 30 days after receipt date.
> Valid at participating US McDonald's.
> Survey Code:
> 31278-01121-21018-20481-00081-0

## McDonald's Restaurant #31278
2378 PINE RD NW
RICE, MN 56367-9740
TEL# 320 393 4600

| KS# 1           | 12/08/2022 08:48 PM |
|-----------------|---------------------|
| Side1           | Order 12            |

| Item                     | Price |
|--------------------------|-------|
| 1 Happy Meal 6 Pc        | 4.89  |
| - 1 Creamy Ranch Cup     |       |
| - 1 Extra Kids Fry       |       |
| - 1 Wreck It Ralph 2     |       |
| - 1 S Coke               |       |
| 1 Snack Oreo McFlurry    | 2.69  |

| Subtotal                 | 7.58  |
| Tax                      | 0.52  |
| Take-Out Total           | 8.10  |

| Cash Tendered            | 10.00 |
| Change                   | 1.90  |

> McDonalds Restaurant Rice
> ***NOW ACCEPTING APPLICATIONS***
> text to #36453
> apply31278

加指令解析小票

现在我们通过提供一段附加提示词来改变输出。自定义提示词放在 agentic_options 里 —— 在 cost_effective、agentic 与 agentic_plus 三个档位可用。下面每个标签页都复用上一页步骤里的 client 与已上传文件：

Python
TypeScript
Go
Java
CLI

parsing_instruction = (
    "The provided document is a McDonald's receipt. "
    "Provide ONLY each line item (item name and price) and the final amount to be paid."
)

result_with_prompt = client.parsing.parse(
    file_id=file.id,
    tier="agentic",
    version="latest",
    expand=["markdown"],
    output_options={
        "markdown": {"tables": {"output_tables_as_markdown": True}},
    },
    agentic_options={"custom_prompt": parsing_instruction},
)

print(result_with_prompt.markdown.pages[0].markdown)

const parsingInstruction =
  "The provided document is a McDonald's receipt. " +
  'Provide ONLY each line item (item name and price) and the final amount to be paid.';

const resultWithPrompt = await client.parsing.parse({
  file_id: file.id,
  tier: 'agentic',
  version: 'latest',
  expand: ['markdown'],
  output_options: {
    markdown: { tables: { output_tables_as_markdown: true } },
  },
  agentic_options: { custom_prompt: parsingInstruction },
});

console.log(resultWithPrompt.markdown.pages[0].markdown);

// Reuses ctx, client, and the uploaded file from above.
instruction := "The provided document is a McDonald's receipt. " +
  "Provide ONLY each line item (item name and price) and the final amount to be paid."

job, err := client.Parsing.New(ctx, llamacloud.ParsingNewParams{
  FileID:  llamacloud.String(file.ID),
  Tier:    llamacloud.ParsingNewParamsTierAgentic,
  Version: llamacloud.ParsingNewParamsVersionLatest,
  OutputOptions: llamacloud.ParsingNewParamsOutputOptions{
    Markdown: llamacloud.ParsingNewParamsOutputOptionsMarkdown{
      Tables: llamacloud.ParsingNewParamsOutputOptionsMarkdownTables{
        OutputTablesAsMarkdown: llamacloud.Bool(true),
      },
    },
  },
  AgenticOptions: llamacloud.ParsingNewParamsAgenticOptions{
    CustomPrompt: llamacloud.String(instruction),
  },
})
if err != nil {
  log.Fatal(err)
}

// expand is a GET parameter — request markdown when you fetch the result
getParams := llamacloud.ParsingGetParams{Expand: []string{"markdown"}}
result, err := client.Parsing.Get(ctx, job.ID, getParams)
if err != nil {
  log.Fatal(err)
}
for result.Job.Status != "COMPLETED" && result.Job.Status != "FAILED" && result.Job.Status != "CANCELLED" {
  time.Sleep(2 * time.Second)
  result, err = client.Parsing.Get(ctx, job.ID, getParams)
  if err != nil {
    log.Fatal(err)
  }
}
if result.Job.Status != "COMPLETED" {
  log.Fatalf("parse ended as %s", result.Job.Status)
}

if len(result.Markdown.Pages) == 0 {
  log.Fatal("parse completed with no markdown pages")
}

fmt.Println(result.Markdown.Pages[0].Markdown)

// Reuses client and the uploaded file from above.
String instruction = "The provided document is a McDonald's receipt. "
        + "Provide ONLY each line item (item name and price) and the final amount to be paid.";

ParsingCreateResponse job = client.parsing().create(ParsingCreateParams.builder()
        .fileId(file.id())
        .tier(ParsingCreateParams.Tier.AGENTIC)
        .version(ParsingCreateParams.Version.LATEST)
        .outputOptions(ParsingCreateParams.OutputOptions.builder()
                .markdown(ParsingCreateParams.OutputOptions.Markdown.builder()
                        .tables(ParsingCreateParams.OutputOptions.Markdown.Tables.builder()
                                .outputTablesAsMarkdown(true)
                                .build())
                        .build())
                .build())
        .agenticOptions(ParsingCreateParams.AgenticOptions.builder()
                .customPrompt(instruction)
                .build())
        .build());

// expand is a query parameter — request markdown when you fetch the result
ParsingGetParams getParams = ParsingGetParams.builder()
        .jobId(job.id())
        .addExpand("markdown")
        .build();

ParsingGetResponse result = client.parsing().get(getParams);
while (!result.job().status().equals(ParsingGetResponse.Job.Status.COMPLETED)
        && !result.job().status().equals(ParsingGetResponse.Job.Status.FAILED)
        && !result.job().status().equals(ParsingGetResponse.Job.Status.CANCELLED)) {
    Thread.sleep(2000);
    result = client.parsing().get(getParams);
}
if (!result.job().status().equals(ParsingGetResponse.Job.Status.COMPLETED)) {
    throw new RuntimeException("parse ended as " + result.job().status());
}

System.out.println(result.markdown().get().pages().get(0).asMarkdownResult().markdown());

终端窗口

# Reuses $FILE_ID from above. The custom prompt goes on --agentic-options.custom-prompt.
JOB_ID=$(llp parsing create \
  --file-id "$FILE_ID" \
  --tier agentic \
  --version latest \
  --output-options.markdown '{tables: {output_tables_as_markdown: true}}' \
  --agentic-options.custom-prompt "The provided document is a McDonald's receipt. Provide ONLY each line item (item name and price) and the final amount to be paid." \
  | jq -r '.id')

# Poll until the job reaches a terminal status
while true; do
  STATUS=$(llp parsing get --job-id "$JOB_ID" | jq -er '.job.status') || break
  case "$STATUS" in COMPLETED|FAILED|CANCELLED) break ;; esac
  sleep 2
done

# Print the markdown for the first page
llp parsing get --job-id "$JOB_ID" --expand markdown \
  | jq -r 'if .job.status == "COMPLETED" then .markdown.pages[0].markdown
    else "Parse job failed: \(.job.error_message // .)" end'

这段提示词把输出收窄到只剩逐项明细与总额：

* Happy Meal 6 Pc 4.89
* Snack Oreo McFlurry 2.69

Take-Out Total 8.10

索引

数据加载

查询

存储

成本分析

应用

聊天机器人

问答

结构化数据

智能体

Chat Engines

Query Engine

连接器

文档与节点

摄取流水线

节点解析器

大语言模型

提示词

回调

节点后处理器

响应合成器

检索器

路由

结构化输出

现有数据

Presets and Modes

能力

起步

能力

Examples

Azure Deployment

Databases and Queues

LLM Integrations

设置

不加指令地解析小票

加指令解析小票

友情链接