mirror of
https://github.com/microsoft/graphrag.git
synced 2025-07-25 09:51:02 +00:00
9b4f24ebce
* Firsst cut at config cleanup * Reorder top nav * Add query prompts to tuning page * Remove dynamic notebook from nav * Add more thorough yml config descriptions in docs * Further clean out the config * Semver * Add new blog post * Emphasize yaml * Clarify output * Fix unit test * Fix bullet nesting
90 lines
4.1 KiB
Markdown
90 lines
4.1 KiB
Markdown
# Manual Prompt Tuning ⚙️
|
|
|
|
The GraphRAG indexer, by default, will run with a handful of prompts that are designed to work well in the broad context of knowledge discovery.
|
|
However, it is quite common to want to tune the prompts to better suit your specific use case.
|
|
We provide a means for you to do this by allowing you to specify a custom prompt file, which will each use a series of token-replacements internally.
|
|
|
|
Each of these prompts may be overridden by writing a custom prompt file in plaintext. We use token-replacements in the form of `{token_name}`, and the descriptions for the available tokens can be found below.
|
|
|
|
## Indexing Prompts
|
|
|
|
### Entity/Relationship Extraction
|
|
|
|
[Prompt Source](http://github.com/microsoft/graphrag/blob/main/graphrag/prompts/index/entity_extraction.py)
|
|
|
|
#### Tokens
|
|
|
|
- **{input_text}** - The input text to be processed.
|
|
- **{entity_types}** - A list of entity types
|
|
- **{tuple_delimiter}** - A delimiter for separating values within a tuple. A single tuple is used to represent an individual entity or relationship.
|
|
- **{record_delimiter}** - A delimiter for separating tuple instances.
|
|
- **{completion_delimiter}** - An indicator for when generation is complete.
|
|
|
|
### Summarize Entity/Relationship Descriptions
|
|
|
|
[Prompt Source](http://github.com/microsoft/graphrag/blob/main/graphrag/prompts/index/summarize_descriptions.py)
|
|
|
|
#### Tokens
|
|
|
|
- **{entity_name}** - The name of the entity or the source/target pair of the relationship.
|
|
- **{description_list}** - A list of descriptions for the entity or relationship.
|
|
|
|
### Claim Extraction
|
|
|
|
[Prompt Source](http://github.com/microsoft/graphrag/blob/main/graphrag/prompts/index/claim_extraction.py)
|
|
|
|
#### Tokens
|
|
|
|
- **{input_text}** - The input text to be processed.
|
|
- **{tuple_delimiter}** - A delimiter for separating values within a tuple. A single tuple is used to represent an individual entity or relationship.
|
|
- **{record_delimiter}** - A delimiter for separating tuple instances.
|
|
- **{completion_delimiter}** - An indicator for when generation is complete.
|
|
- **{entity_specs}** - A list of entity types.
|
|
- **{claim_description}** - Description of what claims should look like. Default is: `"Any claims or facts that could be relevant to information discovery."`
|
|
|
|
See the [configuration documentation](../config/overview.md) for details on how to change this.
|
|
|
|
### Generate Community Reports
|
|
|
|
[Prompt Source](http://github.com/microsoft/graphrag/blob/main/graphrag/prompts/index/community_report.py)
|
|
|
|
#### Tokens
|
|
|
|
- **{input_text}** - The input text to generate the report with. This will contain tables of entities and relationships.
|
|
|
|
## Query Prompts
|
|
|
|
### Local Search
|
|
|
|
[Prompt Source](http://github.com/microsoft/graphrag/blob/main/graphrag/prompts/query/local_search_system_prompt.py)
|
|
|
|
#### Tokens
|
|
|
|
- **{response_type}** - Describe how the response should look. We default to "multiple paragraphs".
|
|
- **{context_data}** - The data tables from GraphRAG's index.
|
|
|
|
### Global Search
|
|
|
|
[Mapper Prompt Source](http://github.com/microsoft/graphrag/blob/main/graphrag/prompts/query/global_search_map_system_prompt.py)
|
|
|
|
[Reducer Prompt Source](http://github.com/microsoft/graphrag/blob/main/graphrag/prompts/query/global_search_reduce_system_prompt.py)
|
|
|
|
[Knowledge Prompt Source](http://github.com/microsoft/graphrag/blob/main/graphrag/prompts/query/global_search_knowledge_system_prompt.py)
|
|
|
|
Global search uses a map/reduce approach to summarization. You can tune these prompts independently. This search also includes the ability to adjust the use of general knowledge from the model's training.
|
|
|
|
#### Tokens
|
|
|
|
- **{response_type}** - Describe how the response should look (reducer only). We default to "multiple paragraphs".
|
|
- **{context_data}** - The data tables from GraphRAG's index.
|
|
|
|
### Drift Search
|
|
|
|
[Prompt Source](http://github.com/microsoft/graphrag/blob/main/graphrag/prompts/query/drift_search_system_prompt.py)
|
|
|
|
#### Tokens
|
|
|
|
- **{response_type}** - Describe how the response should look. We default to "multiple paragraphs".
|
|
- **{context_data}** - The data tables from GraphRAG's index.
|
|
- **{community_reports}** - The most relevant community reports to include in the summarization.
|
|
- **{query}** - The query text as injected into the context. |