Fix: Correct environment variable handling for limited Azure OpenAI G…

…PT and embedding access by properly differentiating endpoints.

nano_graphrag/_llm.py

@@ -4,6 +4,7 @@
 
 import aioboto3
 from openai import AsyncOpenAI, AsyncAzureOpenAI, APIConnectionError, RateLimitError
+from azure.core.exceptions import HttpResponseError
 
 from tenacity import (
     retry,
@@ -19,6 +20,7 @@
 global_openai_async_client = None
 global_azure_openai_async_client = None
 global_amazon_bedrock_async_client = None
+global_azure_openai_emb_async_client = None
 
 
 def get_openai_async_client_instance():
@@ -34,6 +36,12 @@ def get_azure_openai_async_client_instance():
         global_azure_openai_async_client = AsyncAzureOpenAI()
     return global_azure_openai_async_client
 
+def get_azure_openai_emb_async_client_instance():
+    global global_azure_openai_emb_async_client
+    if global_azure_openai_emb_async_client is None:
+        global_azure_openai_emb_async_client = AsyncAzureOpenAI(azure_endpoint=os.environ["AZURE_ENDPOINT_EMB"])
+    return global_azure_openai_emb_async_client
+
 
 def get_amazon_bedrock_async_client_instance():
     global global_amazon_bedrock_async_client
@@ -287,8 +295,11 @@ async def azure_gpt_4o_mini_complete(
     retry=retry_if_exception_type((RateLimitError, APIConnectionError)),
 )
 async def azure_openai_embedding(texts: list[str]) -> np.ndarray:
-    azure_openai_client = get_azure_openai_async_client_instance()
+
+    azure_openai_client = get_azure_openai_emb_async_client_instance()
+
     response = await azure_openai_client.embeddings.create(
-        model="text-embedding-3-small", input=texts, encoding_format="float"
+            model="text-embedding-3-small", input=texts, encoding_format="float"
     )
+
     return np.array([dp.embedding for dp in response.data])