Prompting - Miscellaneous
Chain of Summaries
Cookbook
Instructor - Basics
- Basic use
- Basic use via mixin
- Handling errors with `Maybe` helper class
- Modes
- Making some fields optional
- Private vs public object field
- Automatic correction based on validation results
- Using attributes
- Using LLM API connections from config file
- Validation
- Custom validation using Symfony Validator
- Validation across multiple fields
- Validation with LLM
Instructor - Advanced
- Context caching (structured output)
- Customize parameters of LLM driver
- Custom prompts
- Using structured data as an input
- Extracting arguments of function or method
- Streaming partial updates during inference
- Providing example inputs and outputs
- Extracting scalar values
- Extracting sequences of objects
- Streaming
- Structures
Instructor - Troubleshooting
Instructor - LLM API Support
Instructor - Extras
- Extraction of complex objects
- Extraction of complex objects (Anthropic)
- Extraction of complex objects (Cohere)
- Extraction of complex objects (Gemini)
- Image processing - car damage detection
- Image to data (OpenAI)
- Image to data (Anthropic)
- Image to data (Gemini)
- Generating JSON Schema from PHP classes
- Generating JSON Schema dynamically
- Create tasks from meeting transcription
- Translating UI text fields
- Web page to PHP objects
Polyglot - LLM Basics
Polyglot - LLM Advanced
Polyglot - LLM Troubleshooting
Polyglot - LLM API Support
Polyglot - LLM Extras
Prompting - Zero-Shot Prompting
Prompting - Few-Shot Prompting
Prompting - Thought Generation
Prompting - Miscellaneous
- Arbitrary properties
- Consistent values of arbitrary properties
- Chain of Summaries
- Chain of Thought
- Single label classification
- Multiclass classification
- Entity relationship extraction
- Handling errors
- Limiting the length of lists
- Reflection Prompting
- Restating instructions
- Ask LLM to rewrite instructions
- Expanding search queries
- Summary with Keywords
- Reusing components
- Using CoT to improve interpretation of component data
Prompting - Miscellaneous
Chain of Summaries
Overview
This is an example of summarization with increasing amount of details. Instructor is provided with data structure containing instructions on how to create increasingly detailed summaries of the project report.
It starts with generating an overview of the project, followed by X iterations of increasingly detailed summaries. Each iteration should contain all the information from the previous summary, plus a few additional facts from the content which are most relevant and missing from the previous iteration.
Example
<?php
require 'examples/boot.php';
use Cognesy\Instructor\Instructor;
$report = <<<EOT
[2021-09-01]
Acme Insurance project to implement SalesTech CRM solution is currently
in RED status due to delayed delivery of document production system, led
by 3rd party vendor - Alfatech. Customer (Acme) is discussing the resolution
with the vendor. Due to dependencies it will result in delay of the
ecommerce track by 2 sprints. System integrator (SysCorp) are working
to absorb some of the delay by deploying extra resources to speed up
development when the doc production is done. Another issue is that the
customer is not able to provide the test data for the ecommerce track.
SysCorp notified it will impact stabilization schedule unless resolved by
the end of the month. Steerco has been informed last week about the
potential impact of the issues, but insists on maintaining release schedule
due to marketing campaign already ongoing. Customer executives are asking
us - SalesTech team - to confirm SysCorp's assessment of the situation.
We're struggling with that due to communication issues - SysCorp team has
not shown up on 2 recent calls. Lack of insight has been escalated to
SysCorp's leadership team yesterday, but we've got no response yet. The
previously reported Integration Proxy connectivity issue which was blocking
policy track has been resolved on 2021-08-30 - the track is now GREEN.
Production deployment plan has been finalized on Aug 15th and awaiting
customer approval.
EOT;
/** Executive level summary of the project */
class Summary {
/** current summary iteration, not bigger than 3 */
public int $iteration = 0;
/** @var string[] 1-3 facts most relevant from executive perspective and missing from the summary (avoid technical details) */
public array $missingFacts = [];
/** denser summary in 1-3 sentences, which covers every fact from the previous summary plus the missing ones */
public string $expandedSummary = '';
}
/** Increasingly denser, expanded summaries */
class ChainOfSummaries {
/** simplified, executive view with no details, just a single statement of overall situation */
public string $overview;
/** @var Summary[] contains at least 3 gradually more expanded summaries of the content */
public array $summaries;
}
$summaries = (new Instructor)
->request(
messages: $report,
responseModel: ChainOfSummaries::class,
prompt: 'Generate a denser summary based on the provided content.',
toolName: 'summarizer',
toolDescription: 'Generates a summary based on the provided content.',
options: [
'max_tokens' => 4096,
])
->get();
print("\n# Summaries with increasing density:\n\n");
print("Overview:\n");
print("{$summaries->overview}\n\n");
foreach ($summaries->summaries as $summary) {
print("Expanded summary - iteration #{$summary->iteration}:\n");
print("{$summary->expandedSummary}\n\n");
}
?>