Move prompts

.semversioner/next-release/patch-20241114012244853718.json

@@ -0,0 +1,4 @@
+{
+  "type": "patch",
+  "description": "Centralized prompts and export all for easier injection."
+}

docs/prompt_tuning/manual_prompt_tuning.md

@@ -8,7 +8,7 @@ Each of these prompts may be overridden by writing a custom prompt file in plain
 
 ## Entity/Relationship Extraction
 
-[Prompt Source](http://github.com/microsoft/graphrag/blob/main/graphrag/index/graph/extractors/graph/prompts.py)
+[Prompt Source](http://github.com/microsoft/graphrag/blob/main/graphrag/prompts/entity_extraction.py)
 
 ### Tokens (values provided by extractor)
 
@@ -20,7 +20,7 @@ Each of these prompts may be overridden by writing a custom prompt file in plain
 
 ## Summarize Entity/Relationship Descriptions
 
-[Prompt Source](http://github.com/microsoft/graphrag/blob/main/graphrag/index/graph/extractors/summarize/prompts.py)
+[Prompt Source](http://github.com/microsoft/graphrag/blob/main/graphrag/prompts/summarize_descriptions.py)
 
 ### Tokens (values provided by extractor)
 
@@ -29,7 +29,7 @@ Each of these prompts may be overridden by writing a custom prompt file in plain
 
 ## Claim Extraction
 
-[Prompt Source](http://github.com/microsoft/graphrag/blob/main/graphrag/index/graph/extractors/claims/prompts.py)
+[Prompt Source](http://github.com/microsoft/graphrag/blob/main/graphrag/prompts/claim_extraction.py)
 
 ### Tokens (values provided by extractor)
 
@@ -47,7 +47,7 @@ See the [configuration documentation](../config/overview.md) for details on how
 
 ## Generate Community Reports
 
-[Prompt Source](http://github.com/microsoft/graphrag/blob/main/graphrag/index/graph/extractors/community_reports/prompts.py)
+[Prompt Source](http://github.com/microsoft/graphrag/blob/main/graphrag/prompts/community_report.py)
 
 ### Tokens (values provided by extractor)
 

docs/query/global_search.md

@@ -56,11 +56,11 @@ Below are the key parameters of the [GlobalSearch class](https://github.com/micr
 
 * `llm`: OpenAI model object to be used for response generation
 * `context_builder`: [context builder](https://github.com/microsoft/graphrag/blob/main//graphrag/query/structured_search/global_search/community_context.py) object to be used for preparing context data from community reports
-* `map_system_prompt`: prompt template used in the `map` stage. Default template can be found at [map_system_prompt](https://github.com/microsoft/graphrag/blob/main//graphrag/query/structured_search/global_search/map_system_prompt.py)
-* `reduce_system_prompt`: prompt template used in the `reduce` stage, default template can be found at [reduce_system_prompt](https://github.com/microsoft/graphrag/blob/main//graphrag/query/structured_search/global_search/reduce_system_prompt.py)
+* `map_system_prompt`: prompt template used in the `map` stage. Default template can be found at [map_system_prompt](https://github.com/microsoft/graphrag/blob/main//graphrag/prompts/query/global_search_map_system_prompt.py)
+* `reduce_system_prompt`: prompt template used in the `reduce` stage, default template can be found at [reduce_system_prompt](https://github.com/microsoft/graphrag/blob/main//graphrag/prompts/query/global_search_reduce_system_prompt.py)
 * `response_type`: free-form text describing the desired response type and format (e.g., `Multiple Paragraphs`, `Multi-Page Report`)
 * `allow_general_knowledge`: setting this to True will include additional instructions to the `reduce_system_prompt` to prompt the LLM to incorporate relevant real-world knowledge outside of the dataset. Note that this may increase hallucinations, but can be useful for certain scenarios. Default is False
-*`general_knowledge_inclusion_prompt`: instruction to add to the `reduce_system_prompt` if `allow_general_knowledge` is enabled. Default instruction can be found at [general_knowledge_instruction](https://github.com/microsoft/graphrag/blob/main//graphrag/query/structured_search/global_search/reduce_system_prompt.py)
+*`general_knowledge_inclusion_prompt`: instruction to add to the `reduce_system_prompt` if `allow_general_knowledge` is enabled. Default instruction can be found at [general_knowledge_instruction](https://github.com/microsoft/graphrag/blob/main//graphrag/prompts/query/global_search_knowledge_system_prompt.py)
 * `max_data_tokens`: token budget for the context data
 * `map_llm_params`: a dictionary of additional parameters (e.g., temperature, max_tokens) to be passed to the LLM call at the `map` stage
 * `reduce_llm_params`: a dictionary of additional parameters (e.g., temperature, max_tokens) to passed to the LLM call at the `reduce` stage

docs/query/local_search.md

@@ -50,7 +50,7 @@ Below are the key parameters of the [LocalSearch class](https://github.com/micro
 
 * `llm`: OpenAI model object to be used for response generation
 * `context_builder`: [context builder](https://github.com/microsoft/graphrag/blob/main//graphrag/query/structured_search/local_search/mixed_context.py) object to be used for preparing context data from collections of knowledge model objects
-* `system_prompt`: prompt template used to generate the search response. Default template can be found at [system_prompt](https://github.com/microsoft/graphrag/blob/main//graphrag/query/structured_search/local_search/system_prompt.py)
+* `system_prompt`: prompt template used to generate the search response. Default template can be found at [system_prompt](https://github.com/microsoft/graphrag/blob/main//graphrag/prompts/query/local_search_system_prompt.py)
 * `response_type`: free-form text describing the desired response type and format (e.g., `Multiple Paragraphs`, `Multi-Page Report`)
 * `llm_params`: a dictionary of additional parameters (e.g., temperature, max_tokens) to be passed to the LLM call
 * `context_builder_params`: a dictionary of additional parameters to be passed to the [`context_builder`](https://github.com/microsoft/graphrag/blob/main//graphrag/query/structured_search/local_search/mixed_context.py) object when building context for the search prompt

docs/query/question_generation.md

@@ -13,7 +13,7 @@ Below are the key parameters of the [Question Generation class](https://github.c
 
 * `llm`: OpenAI model object to be used for response generation
 * `context_builder`: [context builder](https://github.com/microsoft/graphrag/blob/main//graphrag/query/structured_search/local_search/mixed_context.py) object to be used for preparing context data from collections of knowledge model objects, using the same context builder class as in local search
-* `system_prompt`: prompt template used to generate candidate questions. Default template can be found at [system_prompt](https://github.com/microsoft/graphrag/blob/main//graphrag/query/question_gen/system_prompt.py)
+* `system_prompt`: prompt template used to generate candidate questions. Default template can be found at [system_prompt](https://github.com/microsoft/graphrag/blob/main//graphrag/prompts/query/question_gen_system_prompt.py)
 * `llm_params`: a dictionary of additional parameters (e.g., temperature, max_tokens) to be passed to the LLM call
 * `context_builder_params`: a dictionary of additional parameters to be passed to the [`context_builder`](https://github.com/microsoft/graphrag/blob/main//graphrag/query/structured_search/local_search/mixed_context.py) object when building context for the question generation prompt
 * `callbacks`: optional callback functions, can be used to provide custom event handlers for LLM's completion streaming events

graphrag/api/query.py

@@ -98,13 +98,24 @@ async def global_search(
         dynamic_community_selection=dynamic_community_selection,
     )
     _entities = read_indexer_entities(nodes, entities, community_level=community_level)
+    map_prompt = _load_search_prompt(config.root_dir, config.global_search.map_prompt)
+    reduce_prompt = _load_search_prompt(
+        config.root_dir, config.global_search.reduce_prompt
+    )
+    knowledge_prompt = _load_search_prompt(
+        config.root_dir, config.global_search.knowledge_prompt
+    )
+
     search_engine = get_global_search_engine(
         config,
         reports=reports,
         entities=_entities,
         communities=_communities,
         response_type=response_type,
         dynamic_community_selection=dynamic_community_selection,
+        map_system_prompt=map_prompt,
+        reduce_system_prompt=reduce_prompt,
+        general_knowledge_inclusion_prompt=knowledge_prompt,
     )
     result: SearchResult = await search_engine.asearch(query=query)
     response = result.response
@@ -156,13 +167,24 @@ async def global_search_streaming(
         dynamic_community_selection=dynamic_community_selection,
     )
     _entities = read_indexer_entities(nodes, entities, community_level=community_level)
+    map_prompt = _load_search_prompt(config.root_dir, config.global_search.map_prompt)
+    reduce_prompt = _load_search_prompt(
+        config.root_dir, config.global_search.reduce_prompt
+    )
+    knowledge_prompt = _load_search_prompt(
+        config.root_dir, config.global_search.knowledge_prompt
+    )
+
     search_engine = get_global_search_engine(
         config,
         reports=reports,
         entities=_entities,
         communities=_communities,
         response_type=response_type,
         dynamic_community_selection=dynamic_community_selection,
+        map_system_prompt=map_prompt,
+        reduce_system_prompt=reduce_prompt,
+        general_knowledge_inclusion_prompt=knowledge_prompt,
     )
     search_result = search_engine.astream_search(query=query)
 
@@ -238,6 +260,7 @@ async def local_search(
 
     _entities = read_indexer_entities(nodes, entities, community_level)
     _covariates = read_indexer_covariates(covariates) if covariates is not None else []
+    prompt = _load_search_prompt(config.root_dir, config.local_search.prompt)
 
     search_engine = get_local_search_engine(
         config=config,
@@ -248,6 +271,7 @@ async def local_search(
         covariates={"claims": _covariates},
         description_embedding_store=description_embedding_store,  # type: ignore
         response_type=response_type,
+        system_prompt=prompt,
     )
 
     result: SearchResult = await search_engine.asearch(query=query)
@@ -312,6 +336,7 @@ async def local_search_streaming(
 
     _entities = read_indexer_entities(nodes, entities, community_level)
     _covariates = read_indexer_covariates(covariates) if covariates is not None else []
+    prompt = _load_search_prompt(config.root_dir, config.local_search.prompt)
 
     search_engine = get_local_search_engine(
         config=config,
@@ -322,6 +347,7 @@ async def local_search_streaming(
         covariates={"claims": _covariates},
         description_embedding_store=description_embedding_store,  # type: ignore
         response_type=response_type,
+        system_prompt=prompt,
     )
     search_result = search_engine.astream_search(query=query)
 
@@ -401,14 +427,15 @@ async def drift_search(
     _entities = read_indexer_entities(nodes, entities, community_level)
     _reports = read_indexer_reports(community_reports, nodes, community_level)
     read_indexer_report_embeddings(_reports, full_content_embedding_store)
-
+    prompt = _load_search_prompt(config.root_dir, config.drift_search.prompt)
     search_engine = get_drift_search_engine(
         config=config,
         reports=_reports,
         text_units=read_indexer_text_units(text_units),
         entities=_entities,
         relationships=read_indexer_relationships(relationships),
         description_embedding_store=description_embedding_store,  # type: ignore
+        local_system_prompt=prompt,
     )
 
     result: SearchResult = await search_engine.asearch(query=query)
@@ -551,3 +578,17 @@ def _reformat_context_data(context_data: dict) -> dict:
             continue
         final_format[key] = records
     return final_format
+
+
+def _load_search_prompt(root_dir: str, prompt_config: str | None) -> str | None:
+    """
+    Load the search prompt from disk if configured.
+
+    If not, leave it empty - the search functions will load their defaults.
+
+    """
+    if prompt_config:
+        prompt_file = Path(root_dir) / prompt_config
+        if prompt_file.exists():
+            return prompt_file.read_bytes().decode(encoding="utf-8")
+    return None

graphrag/cli/initialize.py

@@ -5,14 +5,24 @@
 
 from pathlib import Path
 
-from graphrag.index.graph.extractors.claims.prompts import CLAIM_EXTRACTION_PROMPT
-from graphrag.index.graph.extractors.community_reports.prompts import (
-    COMMUNITY_REPORT_PROMPT,
-)
-from graphrag.index.graph.extractors.graph.prompts import GRAPH_EXTRACTION_PROMPT
-from graphrag.index.graph.extractors.summarize.prompts import SUMMARIZE_PROMPT
 from graphrag.index.init_content import INIT_DOTENV, INIT_YAML
 from graphrag.logging import ReporterType, create_progress_reporter
+from graphrag.prompts.index.claim_extraction import CLAIM_EXTRACTION_PROMPT
+from graphrag.prompts.index.community_report import (
+    COMMUNITY_REPORT_PROMPT,
+)
+from graphrag.prompts.index.entity_extraction import GRAPH_EXTRACTION_PROMPT
+from graphrag.prompts.index.summarize_descriptions import SUMMARIZE_PROMPT
+from graphrag.prompts.query.drift_search_system_prompt import DRIFT_LOCAL_SYSTEM_PROMPT
+from graphrag.prompts.query.global_search_knowledge_system_prompt import (
+    GENERAL_KNOWLEDGE_INSTRUCTION,
+)
+from graphrag.prompts.query.global_search_map_system_prompt import MAP_SYSTEM_PROMPT
+from graphrag.prompts.query.global_search_reduce_system_prompt import (
+    REDUCE_SYSTEM_PROMPT,
+)
+from graphrag.prompts.query.local_search_system_prompt import LOCAL_SEARCH_SYSTEM_PROMPT
+from graphrag.prompts.query.question_gen_system_prompt import QUESTION_SYSTEM_PROMPT
 
 
 def initialize_project_at(path: Path) -> None:
@@ -40,28 +50,21 @@ def initialize_project_at(path: Path) -> None:
     if not prompts_dir.exists():
         prompts_dir.mkdir(parents=True, exist_ok=True)
 
-    entity_extraction = prompts_dir / "entity_extraction.txt"
-    if not entity_extraction.exists():
-        with entity_extraction.open("wb") as file:
-            file.write(
-                GRAPH_EXTRACTION_PROMPT.encode(encoding="utf-8", errors="strict")
-            )
-
-    summarize_descriptions = prompts_dir / "summarize_descriptions.txt"
-    if not summarize_descriptions.exists():
-        with summarize_descriptions.open("wb") as file:
-            file.write(SUMMARIZE_PROMPT.encode(encoding="utf-8", errors="strict"))
-
-    claim_extraction = prompts_dir / "claim_extraction.txt"
-    if not claim_extraction.exists():
-        with claim_extraction.open("wb") as file:
-            file.write(
-                CLAIM_EXTRACTION_PROMPT.encode(encoding="utf-8", errors="strict")
-            )
+    prompts = {
+        "entity_extraction": GRAPH_EXTRACTION_PROMPT,
+        "summarize_descriptions": SUMMARIZE_PROMPT,
+        "claim_extraction": CLAIM_EXTRACTION_PROMPT,
+        "community_report": COMMUNITY_REPORT_PROMPT,
+        "drift_search_system_prompt": DRIFT_LOCAL_SYSTEM_PROMPT,
+        "global_search_map_system_prompt": MAP_SYSTEM_PROMPT,
+        "global_search_reduce_system_prompt": REDUCE_SYSTEM_PROMPT,
+        "global_search_knowledge_system_prompt": GENERAL_KNOWLEDGE_INSTRUCTION,
+        "local_search_system_prompt": LOCAL_SEARCH_SYSTEM_PROMPT,
+        "question_gen_system_prompt": QUESTION_SYSTEM_PROMPT,
+    }
 
-    community_report = prompts_dir / "community_report.txt"
-    if not community_report.exists():
-        with community_report.open("wb") as file:
-            file.write(
-                COMMUNITY_REPORT_PROMPT.encode(encoding="utf-8", errors="strict")
-            )
+    for name, content in prompts.items():
+        prompt_file = prompts_dir / f"{name}.txt"
+        if not prompt_file.exists():
+            with prompt_file.open("wb") as file:
+                file.write(content.encode(encoding="utf-8", errors="strict"))

graphrag/config/__init__.py

@@ -51,6 +51,7 @@
     ClaimExtractionConfig,
     ClusterGraphConfig,
     CommunityReportsConfig,
+    DRIFTSearchConfig,
     EmbedGraphConfig,
     EntityExtractionConfig,
     GlobalSearchConfig,
@@ -85,6 +86,7 @@
     "ClusterGraphConfigInput",
     "CommunityReportsConfig",
     "CommunityReportsConfigInput",
+    "DRIFTSearchConfig",
     "EmbedGraphConfig",
     "EmbedGraphConfigInput",
     "EntityExtractionConfig",