Merge branch 'main' into copilot/fix-1943

.semversioner/2.3.0.json

@@ -0,0 +1,34 @@
+{
+  "changes": [
+    {
+      "description": "Remove Dynamic Max Retries support. Refactor typer typing in cli interface",
+      "type": "minor"
+    },
+    {
+      "description": "Update fnllm to latest. Update default graphrag configuration",
+      "type": "minor"
+    },
+    {
+      "description": "A few fixes and enhancements for better reuse and flow.",
+      "type": "patch"
+    },
+    {
+      "description": "Add full llm response to LLM PRovider output",
+      "type": "patch"
+    },
+    {
+      "description": "Fix Drift Reduce Response for non streaming calls",
+      "type": "patch"
+    },
+    {
+      "description": "Fix global search prompt to include missing formatting key",
+      "type": "patch"
+    },
+    {
+      "description": "Upgrade pyarrow dependency to >=17.0.0 to fix CVE-2024-52338",
+      "type": "patch"
+    }
+  ],
+  "created_at": "2025-05-23T21:02:47+00:00",
+  "version": "2.3.0"
+}

.semversioner/next-release/minor-20250514234228008828.json

@@ -1,4 +0,0 @@
-{
-  "type": "minor",
-  "description": "Update fnllm to latest. Update default graphrag configuration"
-}

.semversioner/next-release/minor-20250520014004743256.json

@@ -1,4 +0,0 @@
-{
-  "type": "minor",
-  "description": "Remove Dynamic Max Retries support. Refactor typer typing in cli interface"
-}

.semversioner/next-release/patch-20250515212234042330.json

@@ -1,4 +0,0 @@
-{
-  "type": "patch",
-  "description": "A few fixes and enhancements for better reuse and flow."
-}

.semversioner/next-release/patch-20250519191613388970.json

@@ -1,4 +0,0 @@
-{
-  "type": "patch",
-  "description": "Upgrade pyarrow dependency to >=17.0.0 to fix CVE-2024-52338"
-}

.semversioner/next-release/patch-20250520214513426940.json

@@ -1,4 +0,0 @@
-{
-  "type": "patch",
-  "description": "Fix global search prompt to include missing formatting key"
-}

CHANGELOG.md

@@ -1,6 +1,16 @@
 # Changelog
 Note: version releases in the 0.x.y range may introduce breaking changes.
 
+## 2.3.0
+
+- minor: Remove Dynamic Max Retries support. Refactor typer typing in cli interface
+- minor: Update fnllm to latest. Update default graphrag configuration
+- patch: A few fixes and enhancements for better reuse and flow.
+- patch: Add full llm response to LLM PRovider output
+- patch: Fix Drift Reduce Response for non streaming calls
+- patch: Fix global search prompt to include missing formatting key
+- patch: Upgrade pyarrow dependency to >=17.0.0 to fix CVE-2024-52338
+
 ## 2.2.1
 
 - patch: Fix Community Report prompt tuning response

graphrag/language_model/providers/fnllm/models.py

@@ -83,7 +83,10 @@ class OpenAIChatFNLLM:
         else:
             response = await self.model(prompt, history=history, **kwargs)
         return BaseModelResponse(
-            output=BaseModelOutput(content=response.output.content),
+            output=BaseModelOutput(
+                content=response.output.content,
+                full_response=response.output.raw_model.to_dict(),
+            ),
             parsed_response=response.parsed_json,
             history=response.history,
             cache_hit=response.cache_hit,
@@ -282,7 +285,10 @@ class AzureOpenAIChatFNLLM:
         else:
             response = await self.model(prompt, history=history, **kwargs)
         return BaseModelResponse(
-            output=BaseModelOutput(content=response.output.content),
+            output=BaseModelOutput(
+                content=response.output.content,
+                full_response=response.output.raw_model.to_dict(),
+            ),
             parsed_response=response.parsed_json,
             history=response.history,
             cache_hit=response.cache_hit,

graphrag/language_model/response/base.py

@@ -18,6 +18,11 @@ class ModelOutput(Protocol):
         """Return the textual content of the output."""
         ...
 
+    @property
+    def full_response(self) -> dict[str, Any] | None:
+        """Return the complete JSON response returned by the model."""
+        ...
+
 
 class ModelResponse(Protocol, Generic[T]):
     """Protocol for LLM response."""
@@ -43,6 +48,10 @@ class BaseModelOutput(BaseModel):
 
     content: str = Field(..., description="The textual content of the output.")
     """The textual content of the output."""
+    full_response: dict[str, Any] | None = Field(
+        None, description="The complete JSON response returned by the LLM provider."
+    )
+    """The complete JSON response returned by the LLM provider."""
 
 
 class BaseModelResponse(BaseModel, Generic[T]):

graphrag/language_model/response/base.pyi

@@ -0,0 +1,50 @@
+# Copyright (c) 2025 Microsoft Corporation.
+# Licensed under the MIT License
+
+from typing import Any, Generic, Protocol, TypeVar
+
+from pydantic import BaseModel
+
+_T = TypeVar("_T", bound=BaseModel, covariant=True)
+
+class ModelOutput(Protocol):
+    @property
+    def content(self) -> str: ...
+    @property
+    def full_response(self) -> dict[str, Any] | None: ...
+
+class ModelResponse(Protocol, Generic[_T]):
+    @property
+    def output(self) -> ModelOutput: ...
+    @property
+    def parsed_response(self) -> _T | None: ...
+    @property
+    def history(self) -> list[Any]: ...
+
+class BaseModelOutput(BaseModel):
+    content: str
+    full_response: dict[str, Any] | None
+
+    def __init__(
+        self,
+        content: str,
+        full_response: dict[str, Any] | None = None,
+    ) -> None: ...
+
+class BaseModelResponse(BaseModel, Generic[_T]):
+    output: BaseModelOutput
+    parsed_response: _T | None
+    history: list[Any]
+    tool_calls: list[Any]
+    metrics: Any | None
+    cache_hit: bool | None
+
+    def __init__(
+        self,
+        output: BaseModelOutput,
+        parsed_response: _T | None = None,
+        history: list[Any] = ...,  # default provided by Pydantic
+        tool_calls: list[Any] = ...,  # default provided by Pydantic
+        metrics: Any | None = None,
+        cache_hit: bool | None = None,
+    ) -> None: ...

graphrag/query/structured_search/drift_search/search.py

@@ -386,7 +386,7 @@ class DRIFTSearch(BaseSearch[DRIFTSearchContextBuilder]):
         model_response = await self.model.achat(
             prompt=query,
             history=search_messages,
-            model_parameters=llm_kwargs,
+            model_parameters=llm_kwargs.get("model_params", {}),
         )
 
         reduced_response = model_response.output.content

pyproject.toml

@@ -1,7 +1,7 @@
 [tool.poetry]
 name = "graphrag"
 # Maintainers: do not change the version here manually, use ./scripts/release.sh
-version = "2.2.1"
+version = "2.3.0"
 description = "GraphRAG: A graph-based retrieval-augmented generation (RAG) system."
 authors = [
     "Alonso Guevara Fernández <alonsog@microsoft.com>",

tests/integration/language_model/test_factory.py

@@ -33,7 +33,11 @@ async def test_create_custom_chat_model():
         def chat(
             self, prompt: str, history: list | None = None, **kwargs: Any
         ) -> ModelResponse:
-            return BaseModelResponse(output=BaseModelOutput(content="content"))
+            return BaseModelResponse(
+                output=BaseModelOutput(
+                    content="content", full_response={"key": "value"}
+                )
+            )
 
         async def achat_stream(
             self, prompt: str, history: list | None = None, **kwargs: Any
@@ -49,9 +53,11 @@ async def test_create_custom_chat_model():
     assert isinstance(model, CustomChatModel)
     response = await model.achat("prompt")
     assert response.output.content == "content"
+    assert response.output.full_response is None
 
     response = model.chat("prompt")
     assert response.output.content == "content"
+    assert response.output.full_response == {"key": "value"}
 
 
 async def test_create_custom_embedding_llm():