Update to latest fnllm (#1930)

* Update to latest fnllm

* Semver + smoke tests

* Add --method to smoke tests indexing

* format...

* Adjust embeddings limiter

Author: AlonsoGuevara

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

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

graphrag/config/defaults.py

@@ -5,6 +5,7 @@
 
 from dataclasses import dataclass, field
 from pathlib import Path
+from typing import Literal
 
 from graphrag.config.enums import (
     AsyncType,
@@ -275,8 +276,8 @@ class LanguageModelDefaults:
     proxy: None = None
     audience: None = None
     model_supports_json: None = None
-    tokens_per_minute: int = 50_000
-    requests_per_minute: int = 1_000
+    tokens_per_minute: Literal["auto"] = "auto"
+    requests_per_minute: Literal["auto"] = "auto"
     retry_strategy: str = "native"
     max_retries: int = 10
     max_retry_wait: float = 10.0

graphrag/config/init_content.py

@@ -34,8 +34,8 @@
     async_mode: {language_model_defaults.async_mode.value} # or asyncio
     retry_strategy: native
     max_retries: -1                   # set to -1 for dynamic retry logic (most optimal setting based on server response)
-    tokens_per_minute: 0              # set to 0 to disable rate limiting
-    requests_per_minute: 0            # set to 0 to disable rate limiting
+    tokens_per_minute: {language_model_defaults.tokens_per_minute}              # set to null to disable rate limiting
+    requests_per_minute: {language_model_defaults.requests_per_minute}            # set to null to disable rate limiting
   {defs.DEFAULT_EMBEDDING_MODEL_ID}:
     type: {defs.DEFAULT_EMBEDDING_MODEL_TYPE.value} # or azure_openai_embedding
     # api_base: https://<instance>.openai.azure.com
@@ -52,8 +52,8 @@
     async_mode: {language_model_defaults.async_mode.value} # or asyncio
     retry_strategy: native
     max_retries: -1                   # set to -1 for dynamic retry logic (most optimal setting based on server response)
-    tokens_per_minute: 0              # set to 0 to disable rate limiting
-    requests_per_minute: 0            # set to 0 to disable rate limiting
+    tokens_per_minute: {language_model_defaults.tokens_per_minute}              # set to null to disable rate limiting
+    requests_per_minute: {language_model_defaults.requests_per_minute}            # set to null to disable rate limiting
 
 ### Input settings ###
 

graphrag/config/models/language_model_config.py

@@ -3,6 +3,8 @@
 
 """Language model configuration."""
 
+from typing import Literal
+
 import tiktoken
 from pydantic import BaseModel, Field, model_validator
 
@@ -192,11 +194,11 @@ def _validate_deployment_name(self) -> None:
         description="The request timeout to use.",
         default=language_model_defaults.request_timeout,
     )
-    tokens_per_minute: int = Field(
+    tokens_per_minute: int | Literal["auto"] | None = Field(
         description="The number of tokens per minute to use for the LLM service.",
         default=language_model_defaults.tokens_per_minute,
     )
-    requests_per_minute: int = Field(
+    requests_per_minute: int | Literal["auto"] | None = Field(
         description="The number of requests per minute to use for the LLM service.",
         default=language_model_defaults.requests_per_minute,
     )

graphrag/language_model/providers/fnllm/utils.py

@@ -81,7 +81,6 @@ def _create_openai_config(config: LanguageModelConfig, azure: bool) -> OpenAICon
             encoding=encoding_model,
             deployment=config.deployment_name,
             chat_parameters=chat_parameters,
-            sleep_on_rate_limit_recommendation=True,
         )
     return PublicOpenAIConfig(
         api_key=config.api_key,
@@ -98,7 +97,6 @@ def _create_openai_config(config: LanguageModelConfig, azure: bool) -> OpenAICon
         model=config.model,
         encoding=encoding_model,
         chat_parameters=chat_parameters,
-        sleep_on_rate_limit_recommendation=True,
     )
 
 

poetry.lock

@@ -56,7 +56,7 @@ lancedb = "^0.17.0"
 aiofiles = "^24.1.0"
 
 # LLM
-fnllm = {extras = ["azure", "openai"], version = "0.2.3"}
+fnllm = {extras = ["azure", "openai"], version = "^0.3.0"}
 json-repair = "^0.30.3"
 openai = "^1.68.0"
 nltk = "3.9.1"

pyproject.toml

@@ -20,8 +20,8 @@ models:
     api_version: ${GRAPHRAG_API_VERSION}
     deployment_name: ${GRAPHRAG_EMBEDDING_DEPLOYMENT_NAME}
     model: ${GRAPHRAG_EMBEDDING_MODEL}
-    tokens_per_minute: ${GRAPHRAG_EMBEDDING_TPM}
-    requests_per_minute: ${GRAPHRAG_EMBEDDING_RPM}
+    tokens_per_minute: null
+    requests_per_minute: null
     concurrent_requests: 50
     async_mode: threaded
 

tests/fixtures/min-csv/settings.yml

@@ -20,8 +20,8 @@ models:
     api_version: ${GRAPHRAG_API_VERSION}
     deployment_name: ${GRAPHRAG_EMBEDDING_DEPLOYMENT_NAME}
     model: ${GRAPHRAG_EMBEDDING_MODEL}
-    tokens_per_minute: ${GRAPHRAG_EMBEDDING_TPM}
-    requests_per_minute: ${GRAPHRAG_EMBEDDING_RPM}
+    tokens_per_minute: null
+    requests_per_minute: null
     concurrent_requests: 50
     async_mode: threaded
 

tests/fixtures/text/settings.yml

@@ -139,6 +139,8 @@ def __run_indexer(
             root.resolve().as_posix(),
             "--logger",
             "print",
+            "--method",
+            "standard",
         ]
         command = [arg for arg in command if arg]
         log.info("running command ", " ".join(command))