Update FNLLM (#1738)

* Add ModelProvider to Query package.

* Spellcheck + others

* Semver

* Fix tests

* Format

* Fix Pyright

* Fix tests

* Fix for smoke tests

* Update fnllm version

* Semver

* Ruff

Author: AlonsoGuevara

.semversioner/next-release/patch-20250225004511391743.json

@@ -0,0 +1,4 @@
+{
+  "type": "patch",
+  "description": "Update fnllm. Remove unused libs."
+}

graphrag/cli/query.py

@@ -391,7 +391,6 @@ def on_context(context: Any) -> None:
     logger.success(f"DRIFT Search Response:\n{response}")
     # NOTE: we return the response and context data here purely as a complete demonstration of the API.
     # External users should use the API directly to get the response and context data.
-    # TODO: Map/Reduce Drift Search answer to a single response
     return response, context_data
 
 

graphrag/config/defaults.py

@@ -244,7 +244,6 @@ class GlobalSearchDefaults:
     dynamic_search_concurrent_coroutines: int = 16
     dynamic_search_max_level: int = 2
     chat_model_id: str = DEFAULT_CHAT_MODEL_ID
-    embedding_model_id: str = DEFAULT_EMBEDDING_MODEL_ID
 
 
 @dataclass

graphrag/config/init_content.py

@@ -151,7 +151,6 @@
 
 global_search:
   chat_model_id: {graphrag_config_defaults.global_search.chat_model_id}
-  embedding_model_id: {graphrag_config_defaults.global_search.embedding_model_id}
   map_prompt: "prompts/global_search_map_system_prompt.txt"
   reduce_prompt: "prompts/global_search_reduce_system_prompt.txt"
   knowledge_prompt: "prompts/global_search_knowledge_system_prompt.txt"

graphrag/index/operations/summarize_communities/community_reports_extractor.py

@@ -3,7 +3,6 @@
 
 """A module containing 'CommunityReportsResult' and 'CommunityReportsExtractor' models."""
 
-import json
 import logging
 import traceback
 from dataclasses import dataclass
@@ -86,11 +85,7 @@ async def __call__(self, inputs: dict[str, Any]):
                 model_parameters={"max_tokens": self._max_report_length},
             )
 
-            # TODO: Json mode is currently broken on fnllm: https://github.com/microsoft/essex-toolkit/issues/364
-            # once fixed, just assign to output the response.parsed_json
-            output = CommunityReportResponse.model_validate(
-                json.loads(response.output.content)
-            )
+            output = response.parsed_response
         except Exception as e:
             log.exception("error generating community report")
             self._on_error(e, traceback.format_exc(), None)

poetry.lock

@@ -57,12 +57,10 @@ lancedb = "^0.17.0"
 aiofiles = "^24.1.0"
 
 # LLM
-fnllm = {extras = ["azure", "openai"], version = "^0.1.2"}
-httpx = "^0.28.1"
+fnllm = {extras = ["azure", "openai"], version = "^0.2.3"}
 json-repair = "^0.30.3"
 openai = "^1.57.0"
 nltk = "3.9.1"
-tenacity = "^9.0.0"
 tiktoken = "^0.8.0"
 
 # Data-Science