graphrag 0.1.1.dev4__tar.gz → 0.1.2.dev48__tar.gz

{graphrag-0.1.1.dev4 → graphrag-0.1.2.dev48}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: graphrag
-Version: 0.1.1.dev4
+Version: 0.1.2.dev48
 Summary: 
 License: MIT
 Author: Alonso Guevara Fernández
@@ -21,7 +21,7 @@ Requires-Dist: devtools (>=0.12.2,<0.13.0)
 Requires-Dist: environs (>=11.0.0,<12.0.0)
 Requires-Dist: fastparquet (>=2024.2.0,<2025.0.0)
 Requires-Dist: graspologic (>=3.4.1,<4.0.0)
-Requires-Dist: lancedb (>=0.9.0,<0.10.0)
+Requires-Dist: lancedb (>=0.10.0,<0.11.0)
 Requires-Dist: nest-asyncio (>=1.6.0,<2.0.0) ; platform_system == "Windows"
 Requires-Dist: networkx (>=3,<4)
 Requires-Dist: nltk (==3.8.1)
@@ -35,8 +35,8 @@ Requires-Dist: pyyaml (>=6.0.1,<7.0.0)
 Requires-Dist: rich (>=13.6.0,<14.0.0)
 Requires-Dist: scipy (==1.12.0)
 Requires-Dist: swifter (>=1.4.0,<2.0.0)
-Requires-Dist: tenacity (>=8.2.3,<9.0.0)
-Requires-Dist: textual (>=0.70.0,<0.71.0)
+Requires-Dist: tenacity (>=8.5.0,<9.0.0)
+Requires-Dist: textual (>=0.72.0,<0.73.0)
 Requires-Dist: tiktoken (>=0.7.0,<0.8.0)
 Requires-Dist: typing-extensions (>=4.12.2,<5.0.0)
 Requires-Dist: uvloop (>=0.19.0,<0.20.0) ; platform_system != "Windows"
@@ -44,9 +44,25 @@ Description-Content-Type: text/markdown
 
 # GraphRAG
 
-👉 [Use the GraphRAG Accelerator solution](https://github.com/Azure-Samples/graphrag-accelerator)
-👉 [Microsoft Research Blog Post](https://www.microsoft.com/en-us/research/blog/graphrag-unlocking-llm-discovery-on-narrative-private-data/)
-👉 [Read the docs](https://microsoft.github.io/graphrag)
+👉 [Use the GraphRAG Accelerator solution](https://github.com/Azure-Samples/graphrag-accelerator) <br/>
+👉 [Microsoft Research Blog Post](https://www.microsoft.com/en-us/research/blog/graphrag-unlocking-llm-discovery-on-narrative-private-data/)<br/>
+👉 [Read the docs](https://microsoft.github.io/graphrag)<br/>
+👉 [GraphRAG Arxiv](https://arxiv.org/pdf/2404.16130)
+
+<div align="left">
+  <a href="https://pypi.org/project/graphrag/">
+    <img alt="PyPI - Version" src="https://img.shields.io/pypi/v/graphrag">
+  </a>
+  <a href="https://pypi.org/project/graphrag/">
+    <img alt="PyPI - Downloads" src="https://img.shields.io/pypi/dm/graphrag">
+  </a>
+  <a href="https://github.com/microsoft/graphrag/issues">
+    <img alt="GitHub Issues" src="https://img.shields.io/github/issues/microsoft/graphrag">
+  </a>
+  <a href="https://github.com/microsoft/graphrag/discussions">
+    <img alt="GitHub Discussions" src="https://img.shields.io/github/discussions/microsoft/graphrag">
+  </a>
+</div>
 
 ## Overview
 
@@ -62,10 +78,13 @@ To get started with the GraphRAG system we recommend trying the [Solution Accele
 
 This repository presents a methodology for using knowledge graph memory structures to enhance LLM outputs. Please note that the provided code serves as a demonstration and is not an officially supported Microsoft offering.
 
+⚠️ *Warning: GraphRAG indexing can be an expensive operation, please read all of the documentation to understand the process and costs involved, and start small.*
+
 ## Diving Deeper
 
 - To learn about our contribution guidelines, see [CONTRIBUTING.md](./CONTRIBUTING.md)
 - To start developing _GraphRAG_, see [DEVELOPING.md](./DEVELOPING.md)
+- Join the conversation and provide feedback in the [GitHub Discussions tab!](https://github.com/microsoft/graphrag/discussions)
 
 ## Prompt Tuning
 

{graphrag-0.1.1.dev4 → graphrag-0.1.2.dev48}/README.md

@@ -1,8 +1,24 @@
 # GraphRAG
 
-👉 [Use the GraphRAG Accelerator solution](https://github.com/Azure-Samples/graphrag-accelerator)
-👉 [Microsoft Research Blog Post](https://www.microsoft.com/en-us/research/blog/graphrag-unlocking-llm-discovery-on-narrative-private-data/)
-👉 [Read the docs](https://microsoft.github.io/graphrag)
+👉 [Use the GraphRAG Accelerator solution](https://github.com/Azure-Samples/graphrag-accelerator) <br/>
+👉 [Microsoft Research Blog Post](https://www.microsoft.com/en-us/research/blog/graphrag-unlocking-llm-discovery-on-narrative-private-data/)<br/>
+👉 [Read the docs](https://microsoft.github.io/graphrag)<br/>
+👉 [GraphRAG Arxiv](https://arxiv.org/pdf/2404.16130)
+
+<div align="left">
+  <a href="https://pypi.org/project/graphrag/">
+    <img alt="PyPI - Version" src="https://img.shields.io/pypi/v/graphrag">
+  </a>
+  <a href="https://pypi.org/project/graphrag/">
+    <img alt="PyPI - Downloads" src="https://img.shields.io/pypi/dm/graphrag">
+  </a>
+  <a href="https://github.com/microsoft/graphrag/issues">
+    <img alt="GitHub Issues" src="https://img.shields.io/github/issues/microsoft/graphrag">
+  </a>
+  <a href="https://github.com/microsoft/graphrag/discussions">
+    <img alt="GitHub Discussions" src="https://img.shields.io/github/discussions/microsoft/graphrag">
+  </a>
+</div>
 
 ## Overview
 
@@ -18,10 +34,13 @@ To get started with the GraphRAG system we recommend trying the [Solution Accele
 
 This repository presents a methodology for using knowledge graph memory structures to enhance LLM outputs. Please note that the provided code serves as a demonstration and is not an officially supported Microsoft offering.
 
+⚠️ *Warning: GraphRAG indexing can be an expensive operation, please read all of the documentation to understand the process and costs involved, and start small.*
+
 ## Diving Deeper
 
 - To learn about our contribution guidelines, see [CONTRIBUTING.md](./CONTRIBUTING.md)
 - To start developing _GraphRAG_, see [DEVELOPING.md](./DEVELOPING.md)
+- Join the conversation and provide feedback in the [GitHub Discussions tab!](https://github.com/microsoft/graphrag/discussions)
 
 ## Prompt Tuning
 

{graphrag-0.1.1.dev4 → graphrag-0.1.2.dev48}/graphrag/config/create_graphrag_config.py

@@ -112,6 +112,9 @@ def create_graphrag_config(
                 proxy=reader.str("proxy") or base.proxy,
                 model=reader.str("model") or base.model,
                 max_tokens=reader.int(Fragment.max_tokens) or base.max_tokens,
+                temperature=reader.float(Fragment.temperature) or base.temperature,
+                top_p=reader.float(Fragment.top_p) or base.top_p,
+                n=reader.int(Fragment.n) or base.n,
                 model_supports_json=reader.bool(Fragment.model_supports_json)
                 or base.model_supports_json,
                 request_timeout=reader.float(Fragment.request_timeout)
@@ -134,13 +137,27 @@ def create_graphrag_config(
         config: LLMConfigInput, base: LLMParameters
     ) -> LLMParameters:
         with reader.use(config.get("llm")):
+            api_type = reader.str(Fragment.type) or defs.EMBEDDING_TYPE
+            api_type = LLMType(api_type) if api_type else defs.LLM_TYPE
             api_key = reader.str(Fragment.api_key) or base.api_key
-            api_base = reader.str(Fragment.api_base) or base.api_base
-            api_version = reader.str(Fragment.api_version) or base.api_version
+
+            # In a unique events where:
+            # - same api_bases for LLM and embeddings (both Azure)
+            # - different api_bases for LLM and embeddings (both Azure)
+            # - LLM uses Azure OpenAI, while embeddings uses base OpenAI (this one is important)
+            # - LLM uses Azure OpenAI, while embeddings uses third-party OpenAI-like API
+            api_base = (
+                reader.str(Fragment.api_base) or base.api_base
+                if _is_azure(api_type)
+                else reader.str(Fragment.api_base)
+            )
+            api_version = (
+                reader.str(Fragment.api_version) or base.api_version
+                if _is_azure(api_type)
+                else reader.str(Fragment.api_version)
+            )
             api_organization = reader.str("organization") or base.organization
             api_proxy = reader.str("proxy") or base.proxy
-            api_type = reader.str(Fragment.type) or defs.EMBEDDING_TYPE
-            api_type = LLMType(api_type) if api_type else defs.LLM_TYPE
             cognitive_services_endpoint = (
                 reader.str(Fragment.cognitive_services_endpoint)
                 or base.cognitive_services_endpoint
@@ -246,6 +263,10 @@ def create_graphrag_config(
                     type=llm_type,
                     model=reader.str(Fragment.model) or defs.LLM_MODEL,
                     max_tokens=reader.int(Fragment.max_tokens) or defs.LLM_MAX_TOKENS,
+                    temperature=reader.float(Fragment.temperature)
+                    or defs.LLM_TEMPERATURE,
+                    top_p=reader.float(Fragment.top_p) or defs.LLM_TOP_P,
+                    n=reader.int(Fragment.n) or defs.LLM_N,
                     model_supports_json=reader.bool(Fragment.model_supports_json),
                     request_timeout=reader.float(Fragment.request_timeout)
                     or defs.LLM_REQUEST_TIMEOUT,
@@ -420,7 +441,7 @@ def create_graphrag_config(
 
         community_report_config = values.get("community_reports") or {}
         with (
-            reader.envvar_prefix(Section.community_report),
+            reader.envvar_prefix(Section.community_reports),
             reader.use(community_report_config),
         ):
             community_reports_model = CommunityReportsConfig(
@@ -474,6 +495,10 @@ def create_graphrag_config(
                 or defs.LOCAL_SEARCH_TOP_K_MAPPED_ENTITIES,
                 top_k_relationships=reader.int("top_k_relationships")
                 or defs.LOCAL_SEARCH_TOP_K_RELATIONSHIPS,
+                temperature=reader.float("llm_temperature")
+                or defs.LOCAL_SEARCH_LLM_TEMPERATURE,
+                top_p=reader.float("llm_top_p") or defs.LOCAL_SEARCH_LLM_TOP_P,
+                n=reader.int("llm_n") or defs.LOCAL_SEARCH_LLM_N,
                 max_tokens=reader.int(Fragment.max_tokens)
                 or defs.LOCAL_SEARCH_MAX_TOKENS,
                 llm_max_tokens=reader.int("llm_max_tokens")
@@ -485,6 +510,10 @@ def create_graphrag_config(
             reader.envvar_prefix(Section.global_search),
         ):
             global_search_model = GlobalSearchConfig(
+                temperature=reader.float("llm_temperature")
+                or defs.GLOBAL_SEARCH_LLM_TEMPERATURE,
+                top_p=reader.float("llm_top_p") or defs.GLOBAL_SEARCH_LLM_TOP_P,
+                n=reader.int("llm_n") or defs.GLOBAL_SEARCH_LLM_N,
                 max_tokens=reader.int(Fragment.max_tokens)
                 or defs.GLOBAL_SEARCH_MAX_TOKENS,
                 data_max_tokens=reader.int("data_max_tokens")
@@ -550,16 +579,19 @@ class Fragment(str, Enum):
     max_retries = "MAX_RETRIES"
     max_retry_wait = "MAX_RETRY_WAIT"
     max_tokens = "MAX_TOKENS"
+    temperature = "TEMPERATURE"
+    top_p = "TOP_P"
+    n = "N"
     model = "MODEL"
     model_supports_json = "MODEL_SUPPORTS_JSON"
     prompt_file = "PROMPT_FILE"
     request_timeout = "REQUEST_TIMEOUT"
-    rpm = "RPM"
+    rpm = "REQUESTS_PER_MINUTE"
     sleep_recommendation = "SLEEP_ON_RATE_LIMIT_RECOMMENDATION"
     storage_account_blob_url = "STORAGE_ACCOUNT_BLOB_URL"
     thread_count = "THREAD_COUNT"
     thread_stagger = "THREAD_STAGGER"
-    tpm = "TPM"
+    tpm = "TOKENS_PER_MINUTE"
     type = "TYPE"
 
 
@@ -570,7 +602,7 @@ class Section(str, Enum):
     cache = "CACHE"
     chunk = "CHUNK"
     claim_extraction = "CLAIM_EXTRACTION"
-    community_report = "COMMUNITY_REPORT"
+    community_reports = "COMMUNITY_REPORTS"
     embedding = "EMBEDDING"
     entity_extraction = "ENTITY_EXTRACTION"
     graphrag = "GRAPHRAG"

{graphrag-0.1.1.dev4 → graphrag-0.1.2.dev48}/graphrag/config/defaults.py

@@ -23,6 +23,9 @@ ENCODING_MODEL = "cl100k_base"
 LLM_TYPE = LLMType.OpenAIChat
 LLM_MODEL = "gpt-4-turbo-preview"
 LLM_MAX_TOKENS = 4000
+LLM_TEMPERATURE = 0
+LLM_TOP_P = 1
+LLM_N = 1
 LLM_REQUEST_TIMEOUT = 180.0
 LLM_TOKENS_PER_MINUTE = 0
 LLM_REQUESTS_PER_MINUTE = 0
@@ -42,19 +45,19 @@ EMBEDDING_TARGET = TextEmbeddingTarget.required
 
 CACHE_TYPE = CacheType.file
 CACHE_BASE_DIR = "cache"
-CHUNK_SIZE = 300
+CHUNK_SIZE = 1200
 CHUNK_OVERLAP = 100
 CHUNK_GROUP_BY_COLUMNS = ["id"]
 CLAIM_DESCRIPTION = (
     "Any claims or facts that could be relevant to information discovery."
 )
-CLAIM_MAX_GLEANINGS = 0
+CLAIM_MAX_GLEANINGS = 1
 CLAIM_EXTRACTION_ENABLED = False
 MAX_CLUSTER_SIZE = 10
 COMMUNITY_REPORT_MAX_LENGTH = 2000
 COMMUNITY_REPORT_MAX_INPUT_LENGTH = 8000
 ENTITY_EXTRACTION_ENTITY_TYPES = ["organization", "person", "geo", "event"]
-ENTITY_EXTRACTION_MAX_GLEANINGS = 0
+ENTITY_EXTRACTION_MAX_GLEANINGS = 1
 INPUT_FILE_TYPE = InputFileType.text
 INPUT_TYPE = InputType.file
 INPUT_BASE_DIR = "input"
@@ -87,9 +90,15 @@ LOCAL_SEARCH_CONVERSATION_HISTORY_MAX_TURNS = 5
 LOCAL_SEARCH_TOP_K_MAPPED_ENTITIES = 10
 LOCAL_SEARCH_TOP_K_RELATIONSHIPS = 10
 LOCAL_SEARCH_MAX_TOKENS = 12_000
+LOCAL_SEARCH_LLM_TEMPERATURE = 0
+LOCAL_SEARCH_LLM_TOP_P = 1
+LOCAL_SEARCH_LLM_N = 1
 LOCAL_SEARCH_LLM_MAX_TOKENS = 2000
 
 # Global Search
+GLOBAL_SEARCH_LLM_TEMPERATURE = 0
+GLOBAL_SEARCH_LLM_TOP_P = 1
+GLOBAL_SEARCH_LLM_N = 1
 GLOBAL_SEARCH_MAX_TOKENS = 12_000
 GLOBAL_SEARCH_DATA_MAX_TOKENS = 12_000
 GLOBAL_SEARCH_MAP_MAX_TOKENS = 1000

{graphrag-0.1.1.dev4 → graphrag-0.1.2.dev48}/graphrag/config/models/claim_extraction_config.py

@@ -43,7 +43,9 @@ class ClaimExtractionConfig(LLMConfig):
             "type": ExtractClaimsStrategyType.graph_intelligence,
             "llm": self.llm.model_dump(),
             **self.parallelization.model_dump(),
-            "extraction_prompt": (Path(root_dir) / self.prompt).read_text()
+            "extraction_prompt": (Path(root_dir) / self.prompt)
+            .read_bytes()
+            .decode(encoding="utf-8")
             if self.prompt
             else None,
             "claim_description": self.description,

{graphrag-0.1.1.dev4 → graphrag-0.1.2.dev48}/graphrag/config/models/community_reports_config.py

@@ -38,7 +38,9 @@ class CommunityReportsConfig(LLMConfig):
             "type": CreateCommunityReportsStrategyType.graph_intelligence,
             "llm": self.llm.model_dump(),
             **self.parallelization.model_dump(),
-            "extraction_prompt": (Path(root_dir) / self.prompt).read_text()
+            "extraction_prompt": (Path(root_dir) / self.prompt)
+            .read_bytes()
+            .decode(encoding="utf-8")
             if self.prompt
             else None,
             "max_report_length": self.max_length,

{graphrag-0.1.1.dev4 → graphrag-0.1.2.dev48}/graphrag/config/models/entity_extraction_config.py

@@ -38,7 +38,9 @@ class EntityExtractionConfig(LLMConfig):
             "type": ExtractEntityStrategyType.graph_intelligence,
             "llm": self.llm.model_dump(),
             **self.parallelization.model_dump(),
-            "extraction_prompt": (Path(root_dir) / self.prompt).read_text()
+            "extraction_prompt": (Path(root_dir) / self.prompt)
+            .read_bytes()
+            .decode(encoding="utf-8")
             if self.prompt
             else None,
             "max_gleanings": self.max_gleanings,

{graphrag-0.1.1.dev4 → graphrag-0.1.2.dev48}/graphrag/config/models/global_search_config.py

@@ -11,6 +11,18 @@ import graphrag.config.defaults as defs
 class GlobalSearchConfig(BaseModel):
     """The default configuration section for Cache."""
 
+    temperature: float | None = Field(
+        description="The temperature to use for token generation.",
+        default=defs.GLOBAL_SEARCH_LLM_TEMPERATURE,
+    )
+    top_p: float | None = Field(
+        description="The top-p value to use for token generation.",
+        default=defs.GLOBAL_SEARCH_LLM_TOP_P,
+    )
+    n: int | None = Field(
+        description="The number of completions to generate.",
+        default=defs.GLOBAL_SEARCH_LLM_N,
+    )
     max_tokens: int = Field(
         description="The maximum context size in tokens.",
         default=defs.GLOBAL_SEARCH_MAX_TOKENS,

{graphrag-0.1.1.dev4 → graphrag-0.1.2.dev48}/graphrag/config/models/llm_parameters.py

@@ -25,6 +25,18 @@ class LLMParameters(BaseModel):
         description="The maximum number of tokens to generate.",
         default=defs.LLM_MAX_TOKENS,
     )
+    temperature: float | None = Field(
+        description="The temperature to use for token generation.",
+        default=defs.LLM_TEMPERATURE,
+    )
+    top_p: float | None = Field(
+        description="The top-p value to use for token generation.",
+        default=defs.LLM_TOP_P,
+    )
+    n: int | None = Field(
+        description="The number of completions to generate.",
+        default=defs.LLM_N,
+    )
     request_timeout: float = Field(
         description="The request timeout to use.", default=defs.LLM_REQUEST_TIMEOUT
     )

{graphrag-0.1.1.dev4 → graphrag-0.1.2.dev48}/graphrag/config/models/local_search_config.py

@@ -31,6 +31,18 @@ class LocalSearchConfig(BaseModel):
         description="The top k mapped relations.",
         default=defs.LOCAL_SEARCH_TOP_K_RELATIONSHIPS,
     )
+    temperature: float | None = Field(
+        description="The temperature to use for token generation.",
+        default=defs.LOCAL_SEARCH_LLM_TEMPERATURE,
+    )
+    top_p: float | None = Field(
+        description="The top-p value to use for token generation.",
+        default=defs.LOCAL_SEARCH_LLM_TOP_P,
+    )
+    n: int | None = Field(
+        description="The number of completions to generate.",
+        default=defs.LOCAL_SEARCH_LLM_N,
+    )
     max_tokens: int = Field(
         description="The maximum tokens.", default=defs.LOCAL_SEARCH_MAX_TOKENS
     )

{graphrag-0.1.1.dev4 → graphrag-0.1.2.dev48}/graphrag/config/models/summarize_descriptions_config.py

@@ -34,7 +34,9 @@ class SummarizeDescriptionsConfig(LLMConfig):
             "type": SummarizeStrategyType.graph_intelligence,
             "llm": self.llm.model_dump(),
             **self.parallelization.model_dump(),
-            "summarize_prompt": (Path(root_dir) / self.prompt).read_text()
+            "summarize_prompt": (Path(root_dir) / self.prompt)
+            .read_bytes()
+            .decode(encoding="utf-8")
             if self.prompt
             else None,
             "max_summary_length": self.max_length,

{graphrag-0.1.1.dev4 → graphrag-0.1.2.dev48}/graphrag/index/__main__.py

@@ -28,9 +28,10 @@ if __name__ == "__main__":
     )
     parser.add_argument(
         "--root",
-        help="If no configuration is defined, the root directory to use for input data and output data",
+        help="If no configuration is defined, the root directory to use for input data and output data. Default value: the current directory",
         # Only required if config is not defined
         required=False,
+        default=".",
         type=str,
     )
     parser.add_argument(
@@ -62,9 +63,16 @@ if __name__ == "__main__":
         help="Create an initial configuration in the given path.",
         action="store_true",
     )
-
+    parser.add_argument(
+        "--overlay-defaults",
+        help="Overlay default configuration values on a provided configuration file (--config).",
+        action="store_true",
+    )
     args = parser.parse_args()
 
+    if args.overlay_defaults and not args.config:
+        parser.error("--overlay-defaults requires --config")
+
     index_cli(
         root=args.root,
         verbose=args.verbose or False,
@@ -76,5 +84,6 @@ if __name__ == "__main__":
         emit=args.emit,
         dryrun=args.dryrun or False,
         init=args.init or False,
+        overlay_defaults=args.overlay_defaults or False,
         cli=True,
     )

{graphrag-0.1.1.dev4 → graphrag-0.1.2.dev48}/graphrag/index/cli.py

@@ -69,7 +69,7 @@ def redact(input: dict) -> str:
 
 
 def index_cli(
-    root: str | None,
+    root: str,
     init: bool,
     verbose: bool,
     resume: str | None,
@@ -79,19 +79,24 @@ def index_cli(
     config: str | None,
     emit: str | None,
     dryrun: bool,
+    overlay_defaults: bool,
     cli: bool = False,
 ):
     """Run the pipeline with the given config."""
-    root = root or ""
     run_id = resume or time.strftime("%Y%m%d-%H%M%S")
     _enable_logging(root, run_id, verbose)
     progress_reporter = _get_progress_reporter(reporter)
     if init:
         _initialize_project_at(root, progress_reporter)
         sys.exit(0)
-    pipeline_config: str | PipelineConfig = config or _create_default_config(
-        root, verbose, dryrun or False, progress_reporter
-    )
+    if overlay_defaults:
+        pipeline_config: str | PipelineConfig = _create_default_config(
+            root, config, verbose, dryrun or False, progress_reporter
+        )
+    else:
+        pipeline_config: str | PipelineConfig = config or _create_default_config(
+            root, None, verbose, dryrun or False, progress_reporter
+        )
     cache = NoopPipelineCache() if nocache else None
     pipeline_emit = emit.split(",") if emit else None
     encountered_errors = False
@@ -180,11 +185,11 @@ def _initialize_project_at(path: str, reporter: ProgressReporter) -> None:
 
     dotenv = root / ".env"
     if not dotenv.exists():
-        with settings_yaml.open("w") as file:
-            file.write(INIT_YAML)
+        with settings_yaml.open("wb") as file:
+            file.write(INIT_YAML.encode(encoding="utf-8", errors="strict"))
 
-    with dotenv.open("w") as file:
-        file.write(INIT_DOTENV)
+    with dotenv.open("wb") as file:
+        file.write(INIT_DOTENV.encode(encoding="utf-8", errors="strict"))
 
     prompts_dir = root / "prompts"
     if not prompts_dir.exists():
@@ -192,34 +197,48 @@ def _initialize_project_at(path: str, reporter: ProgressReporter) -> None:
 
     entity_extraction = prompts_dir / "entity_extraction.txt"
     if not entity_extraction.exists():
-        with entity_extraction.open("w") as file:
-            file.write(GRAPH_EXTRACTION_PROMPT)
+        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("w") as file:
-            file.write(SUMMARIZE_PROMPT)
+        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("w") as file:
-            file.write(CLAIM_EXTRACTION_PROMPT)
+        with claim_extraction.open("wb") as file:
+            file.write(
+                CLAIM_EXTRACTION_PROMPT.encode(encoding="utf-8", errors="strict")
+            )
 
     community_report = prompts_dir / "community_report.txt"
     if not community_report.exists():
-        with community_report.open("w") as file:
-            file.write(COMMUNITY_REPORT_PROMPT)
+        with community_report.open("wb") as file:
+            file.write(
+                COMMUNITY_REPORT_PROMPT.encode(encoding="utf-8", errors="strict")
+            )
 
 
 def _create_default_config(
-    root: str, verbose: bool, dryrun: bool, reporter: ProgressReporter
+    root: str,
+    config: str | None,
+    verbose: bool,
+    dryrun: bool,
+    reporter: ProgressReporter,
 ) -> PipelineConfig:
-    """Create a default config if none is provided."""
+    """Overlay default values on an existing config or create a default config if none is provided."""
+    if config and not Path(config).exists():
+        msg = f"Configuration file {config} does not exist"
+        raise ValueError
+
     if not Path(root).exists():
         msg = f"Root directory {root} does not exist"
         raise ValueError(msg)
 
-    parameters = _read_config_parameters(root, reporter)
+    parameters = _read_config_parameters(root, config, reporter)
     log.info(
         "using default configuration: %s",
         redact(parameters.model_dump()),
@@ -237,27 +256,35 @@ def _create_default_config(
     return result
 
 
-def _read_config_parameters(root: str, reporter: ProgressReporter):
+def _read_config_parameters(root: str, config: str | None, reporter: ProgressReporter):
     _root = Path(root)
-    settings_yaml = _root / "settings.yaml"
+    settings_yaml = (
+        Path(config)
+        if config and Path(config).suffix in [".yaml", ".yml"]
+        else _root / "settings.yaml"
+    )
     if not settings_yaml.exists():
         settings_yaml = _root / "settings.yml"
-    settings_json = _root / "settings.json"
+    settings_json = (
+        Path(config)
+        if config and Path(config).suffix == ".json"
+        else _root / "settings.json"
+    )
 
     if settings_yaml.exists():
         reporter.success(f"Reading settings from {settings_yaml}")
-        with settings_yaml.open("r") as file:
+        with settings_yaml.open("rb") as file:
             import yaml
 
-            data = yaml.safe_load(file)
+            data = yaml.safe_load(file.read().decode(encoding="utf-8", errors="strict"))
             return create_graphrag_config(data, root)
 
     if settings_json.exists():
         reporter.success(f"Reading settings from {settings_json}")
-        with settings_json.open("r") as file:
+        with settings_json.open("rb") as file:
             import json
 
-            data = json.loads(file.read())
+            data = json.loads(file.read().decode(encoding="utf-8", errors="strict"))
             return create_graphrag_config(data, root)
 
     reporter.success("Reading settings from environment variables")

{graphrag-0.1.1.dev4 → graphrag-0.1.2.dev48}/graphrag/index/graph/extractors/claims/claim_extractor.py

@@ -10,6 +10,7 @@ from typing import Any
 
 import tiktoken
 
+import graphrag.config.defaults as defs
 from graphrag.index.typing import ErrorHandlerFn
 from graphrag.llm import CompletionLLM
 
@@ -80,7 +81,9 @@ class ClaimExtractor:
         self._input_resolved_entities_key = (
             input_resolved_entities_key or "resolved_entities"
         )
-        self._max_gleanings = max_gleanings if max_gleanings is not None else 0
+        self._max_gleanings = (
+            max_gleanings if max_gleanings is not None else defs.CLAIM_MAX_GLEANINGS
+        )
         self._on_error = on_error or (lambda _e, _s, _d: None)
 
         # Construct the looping arguments

{graphrag-0.1.1.dev4 → graphrag-0.1.2.dev48}/graphrag/index/graph/extractors/graph/graph_extractor.py

@@ -14,6 +14,7 @@ from typing import Any
 import networkx as nx
 import tiktoken
 
+import graphrag.config.defaults as defs
 from graphrag.index.typing import ErrorHandlerFn
 from graphrag.index.utils import clean_str
 from graphrag.llm import CompletionLLM
@@ -78,7 +79,11 @@ class GraphExtractor:
         )
         self._entity_types_key = entity_types_key or "entity_types"
         self._extraction_prompt = prompt or GRAPH_EXTRACTION_PROMPT
-        self._max_gleanings = max_gleanings if max_gleanings is not None else 0
+        self._max_gleanings = (
+            max_gleanings
+            if max_gleanings is not None
+            else defs.ENTITY_EXTRACTION_MAX_GLEANINGS
+        )
         self._on_error = on_error or (lambda _e, _s, _d: None)
 
         # Construct the looping arguments

{graphrag-0.1.1.dev4 → graphrag-0.1.2.dev48}/graphrag/index/init_content.py

@@ -24,6 +24,9 @@ llm:
   # max_retry_wait: {defs.LLM_MAX_RETRY_WAIT}
   # sleep_on_rate_limit_recommendation: true # whether to sleep when azure suggests wait-times
   # concurrent_requests: {defs.LLM_CONCURRENT_REQUESTS} # the number of parallel inflight requests that may be made
+  # temperature: {defs.LLM_TEMPERATURE} # temperature for sampling
+  # top_p: {defs.LLM_TOP_P} # top-p sampling
+  # n: {defs.LLM_N} # Number of completions to generate
 
 parallelization:
   stagger: {defs.PARALLELIZATION_STAGGER}
@@ -90,7 +93,7 @@ entity_extraction:
   ## async_mode: override the global async_mode settings for this task
   prompt: "prompts/entity_extraction.txt"
   entity_types: [{",".join(defs.ENTITY_EXTRACTION_ENTITY_TYPES)}]
-  max_gleanings: 0
+  max_gleanings: {defs.ENTITY_EXTRACTION_MAX_GLEANINGS}
 
 summarize_descriptions:
   ## llm: override the global llm settings for this task
@@ -108,7 +111,7 @@ claim_extraction:
   description: "{defs.CLAIM_DESCRIPTION}"
   max_gleanings: {defs.CLAIM_MAX_GLEANINGS}
 
-community_report:
+community_reports:
   ## llm: override the global llm settings for this task
   ## parallelization: override the global parallelization settings for this task
   ## async_mode: override the global async_mode settings for this task
@@ -141,9 +144,15 @@ local_search:
   # conversation_history_max_turns: {defs.LOCAL_SEARCH_CONVERSATION_HISTORY_MAX_TURNS}
   # top_k_mapped_entities: {defs.LOCAL_SEARCH_TOP_K_MAPPED_ENTITIES}
   # top_k_relationships: {defs.LOCAL_SEARCH_TOP_K_RELATIONSHIPS}
+  # llm_temperature: {defs.LOCAL_SEARCH_LLM_TEMPERATURE} # temperature for sampling
+  # llm_top_p: {defs.LOCAL_SEARCH_LLM_TOP_P} # top-p sampling
+  # llm_n: {defs.LOCAL_SEARCH_LLM_N} # Number of completions to generate
   # max_tokens: {defs.LOCAL_SEARCH_MAX_TOKENS}
 
 global_search:
+  # llm_temperature: {defs.GLOBAL_SEARCH_LLM_TEMPERATURE} # temperature for sampling
+  # llm_top_p: {defs.GLOBAL_SEARCH_LLM_TOP_P} # top-p sampling
+  # llm_n: {defs.GLOBAL_SEARCH_LLM_N} # Number of completions to generate
   # max_tokens: {defs.GLOBAL_SEARCH_MAX_TOKENS}
   # data_max_tokens: {defs.GLOBAL_SEARCH_DATA_MAX_TOKENS}
   # map_max_tokens: {defs.GLOBAL_SEARCH_MAP_MAX_TOKENS}