Skip to content

Commit c7c0891

Browse files
jcodellajcodella
committed
udpated LangChain semantic caching example
1 parent bbd17a2 commit c7c0891

File tree

2 files changed

+113
-64
lines changed

2 files changed

+113
-64
lines changed

‎Python/CosmosDB-MongoDB-vCore/Semantic-Caching-Tutorial.ipynb‎

Lines changed: 103 additions & 55 deletions
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,33 @@
11
{
22
"cells": [
3+
{
4+
"cell_type": "markdown",
5+
"id": "74e01527",
6+
"metadata": {},
7+
"source": [
8+
"# Introduction\n",
9+
"\n",
10+
"In this tutorial, we'll demonstrate how to leverage a semantic caching with Azure Cosmos DB for MongoDB and LangChain.\n",
11+
"\n",
12+
"[Learn more here from the LangChain docs.](https://python.langchain.com/docs/integrations/llms/llm_caching#azure-cosmos-db-semantic-cache)"
13+
]
14+
},
15+
{
16+
"cell_type": "code",
17+
"execution_count": null,
18+
"id": "c19c036d",
19+
"metadata": {},
20+
"outputs": [],
21+
"source": [
22+
"# ! pip install langchain\n",
23+
"# ! pip install langchain_openai\n",
24+
"# ! pip install langchain_community\n",
25+
"# ! pip install pymongo\n",
26+
"# ! pip install python-dotenv\n",
27+
"# ! pip install azure-core\n",
28+
"# ! pip install azure-cosmos"
29+
]
30+
},
331
{
432
"cell_type": "code",
533
"execution_count": null,
@@ -17,13 +45,13 @@
1745
" CosmosDBVectorSearchType,\n",
1846
")\n",
1947
"\n",
20-
"from langchain_openai import OpenAIEmbeddings, AzureOpenAIEmbeddings\n",
21-
"from langchain_openai import OpenAI, AzureOpenAI\n",
22-
"from langchain.globals import set_llm_cache\n",
48+
"from langchain.chains import ConversationalRetrievalChain, LLMChain\n",
49+
"from langchain.prompts import PromptTemplate\n",
2350
"\n",
24-
"from azure.core.exceptions import AzureError\n",
25-
"from azure.core.credentials import AzureKeyCredential\n",
51+
"from langchain_openai import AzureChatOpenAI, AzureOpenAIEmbeddings\n",
52+
"from langchain.globals import set_llm_cache\n",
2653
"\n",
54+
"import pymongo\n",
2755
"from dotenv import load_dotenv"
2856
]
2957
},
@@ -42,7 +70,8 @@
4270
"\n",
4371
"COSMOS_MONGO_USER = config['cosmos_db_mongo_user']\n",
4472
"COSMOS_MONGO_PWD = config['cosmos_db_mongo_pwd']\n",
45-
"COSMOS_MONGO_SERVER = config['cosmos_db_mongo_server'] "
73+
"COSMOS_MONGO_SERVER = config['cosmos_db_mongo_server']\n",
74+
"DIMENSIONS = int(config['openai_embeddings_dimensions']) "
4675
]
4776
},
4877
{
@@ -57,7 +86,8 @@
5786
"\n",
5887
"CONNECTION_STRING = (\"mongodb+srv://\"+COSMOS_MONGO_USER+\":\"+COSMOS_MONGO_PWD+\"@\"+COSMOS_MONGO_SERVER+\"?tls=true&authMechanism=SCRAM-SHA-256&retrywrites=false&maxIdleTimeMS=120000\")\n",
5988
"\n",
60-
"DB_NAME, COLLECTION_NAME = NAMESPACE.split(\".\")"
89+
"DB_NAME, COLLECTION_NAME = NAMESPACE.split(\".\")\n",
90+
"mongo_client = pymongo.MongoClient(CONNECTION_STRING)\n"
6191
]
6292
},
6393
{
@@ -67,24 +97,66 @@
6797
"metadata": {},
6898
"outputs": [],
6999
"source": [
70-
"# Requires model version 0301 or more recent\n",
71-
"llm = AzureOpenAI(\n",
72-
" deployment_name=config['openai_completions_deployment'],\n",
73-
" model_name=config['openai_completions_model'],\n",
74-
" openai_api_key=config['openai_api_key'],\n",
75-
" azure_endpoint=config['openai_api_endpoint'],\n",
76-
" openai_api_version=config['openai_api_version'],\n",
77-
" n=2,\n",
78-
" best_of=2,\n",
79-
" cache=True)\n",
80-
"\n",
81-
"\n",
82-
"embeddings = AzureOpenAIEmbeddings(\n",
83-
" azure_deployment=config['openai_embeddings_deployment'],\n",
84-
" model=config['openai_embeddings_model'],\n",
85-
" openai_api_key=config['openai_api_key'],\n",
86-
" azure_endpoint=config['openai_api_endpoint'],\n",
87-
" openai_api_version=config['openai_api_version'])"
100+
"def init_llm_chain():\n",
101+
"\n",
102+
" # Clear old cache if it exists\n",
103+
" mongo_client[DB_NAME][COLLECTION_NAME].drop_indexes()\n",
104+
" mongo_client[DB_NAME].drop_collection(COLLECTION_NAME)\n",
105+
"\n",
106+
" # Define a template for the LLM prompt\n",
107+
" prompt_template = \"\"\"\n",
108+
" You are an upbeat AI assistant who is excited to help answer questions. \n",
109+
"\n",
110+
" Question: {question}\n",
111+
" If you don't know the answer, just say that you don't know, don't try to make up an answer.\n",
112+
" \"\"\"\n",
113+
" chatbot_prompt = PromptTemplate(\n",
114+
" template = prompt_template, input_variables = [\"question\", \"context\"])\n",
115+
"\n",
116+
" # Requires model version 0301 or more recent\n",
117+
" # Point to completions model deployed in Azure OpenAI\n",
118+
" llm = AzureChatOpenAI(\n",
119+
" deployment_name=config['openai_completions_deployment'],\n",
120+
" model_name=config['openai_completions_model'],\n",
121+
" api_key=config['openai_api_key'],\n",
122+
" azure_endpoint=config['openai_api_endpoint'],\n",
123+
" api_version=config['openai_api_version'],\n",
124+
" cache=True,\n",
125+
" n=1)\n",
126+
"\n",
127+
" # Point to embeddings model deployed in Azure OpenAI\n",
128+
" embeddings = AzureOpenAIEmbeddings(\n",
129+
" azure_deployment=config['openai_embeddings_deployment'],\n",
130+
" model=config['openai_embeddings_model'],\n",
131+
" api_key=config['openai_api_key'],\n",
132+
" azure_endpoint=config['openai_api_endpoint'],\n",
133+
" dimensions=DIMENSIONS)\n",
134+
"\n",
135+
" # Setup simple LLM chain\n",
136+
" llm_chain = LLMChain(llm = llm, prompt=chatbot_prompt)\n",
137+
"\n",
138+
" # Setup semantic cache for LLM\n",
139+
" num_lists = 1\n",
140+
" similarity_algorithm = CosmosDBSimilarityType.COS\n",
141+
" kind = CosmosDBVectorSearchType.VECTOR_IVF\n",
142+
"\n",
143+
" score_threshold = 0.9\n",
144+
"\n",
145+
" sem_cache = AzureCosmosDBSemanticCache(\n",
146+
" cosmosdb_connection_string=CONNECTION_STRING,\n",
147+
" cosmosdb_client=None,\n",
148+
" embedding=embeddings,\n",
149+
" database_name=DB_NAME,\n",
150+
" collection_name=COLLECTION_NAME,\n",
151+
" num_lists=num_lists,\n",
152+
" similarity=similarity_algorithm,\n",
153+
" kind=kind,\n",
154+
" dimensions=DIMENSIONS,\n",
155+
" score_threshold=score_threshold)\n",
156+
"\n",
157+
" set_llm_cache(sem_cache)\n",
158+
"\n",
159+
" return llm_chain"
88160
]
89161
},
90162
{
@@ -96,32 +168,8 @@
96168
},
97169
"outputs": [],
98170
"source": [
99-
"# Default value for these params\n",
100-
"num_lists = 1\n",
101-
"dimensions = 1536\n",
102-
"similarity_algorithm = CosmosDBSimilarityType.COS\n",
103-
"kind = CosmosDBVectorSearchType.VECTOR_IVF\n",
104-
"m = 16\n",
105-
"ef_construction = 64\n",
106-
"ef_search = 40\n",
107-
"score_threshold = 0.9\n",
108-
"\n",
109-
"set_llm_cache(\n",
110-
" AzureCosmosDBSemanticCache(\n",
111-
" cosmosdb_connection_string=CONNECTION_STRING,\n",
112-
" cosmosdb_client=None,\n",
113-
" embedding=embeddings,\n",
114-
" database_name=DB_NAME,\n",
115-
" collection_name=COLLECTION_NAME,\n",
116-
" num_lists=num_lists,\n",
117-
" similarity=similarity_algorithm,\n",
118-
" kind=kind,\n",
119-
" dimensions=dimensions,\n",
120-
" m=m,\n",
121-
" ef_construction=ef_construction,\n",
122-
" ef_search=ef_search,\n",
123-
" score_threshold=score_threshold )\n",
124-
")"
171+
"# Initialize llm chain\n",
172+
"llm_chain = init_llm_chain()"
125173
]
126174
},
127175
{
@@ -133,7 +181,7 @@
133181
"source": [
134182
"%%time\n",
135183
"# The first time, the quesiton/response is not yet cachced in Cosmos DB, so retrieval should be slower\n",
136-
"llm(\"Tell me about beer making\")"
184+
"llm_chain.invoke(\"Tell me something interesting about beer making\")"
137185
]
138186
},
139187
{
@@ -145,7 +193,7 @@
145193
"source": [
146194
"%%time\n",
147195
"# This quesiton/response is not yet cachced in Cosmos DB, so retrieval should be slower\n",
148-
"llm(\"Tell me a joke?\")"
196+
"llm_chain(\"Tell me a joke about tomatoes and food.\")"
149197
]
150198
},
151199
{
@@ -157,7 +205,7 @@
157205
"source": [
158206
"%%time\n",
159207
"# The second time, the quesiton/response is cached in Cosmos DB, so retrieval should be faster\n",
160-
"llm(\"Tell me about beer making\")"
208+
"llm_chain(\"Tell me something interesting about beer making\")"
161209
]
162210
},
163211
{
@@ -169,7 +217,7 @@
169217
"source": [
170218
"%%time\n",
171219
"# This question is semantically similar to the previous one within the score_threshold amount, so retrieval should be faster\n",
172-
"llm(\"How do I make beer?\")"
220+
"llm_chain(\"How do I make beer?\")"
173221
]
174222
}
175223
],

‎Python/CosmosDB-MongoDB-vCore/example.env‎

Lines changed: 10 additions & 9 deletions
Original file line numberDiff line numberDiff line change
@@ -1,11 +1,12 @@
1-
cosmos_db_mongo_user = "sample"
2-
cosmos_db_mongo_pwd = "sample"
3-
cosmos_db_mongo_server = "sample_endpoint.com/"
4-
openai_api_endpoint = "sample_endpoint.com/"
5-
openai_api_key =
1+
cosmos_db_mongo_user = ""
2+
cosmos_db_mongo_pwd = ""
3+
cosmos_db_mongo_server = "yourcdbresource.mongocluster.cosmos.azure.com/"
4+
openai_api_endpoint = "https://your_endpoint.openai.azure.com/"
5+
openai_api_key = """
66
openai_api_type = "azure"
77
openai_api_version = "2023-05-15"
8-
openai_embeddings_deployment = "deployment_name"
9-
openai_embeddings_model = "model_type"
10-
openai_completions_deployment = "deployment_name"
11-
openai_completions_model = "model_type"
8+
openai_embeddings_deployment = "embedding-deployment-name"
9+
openai_embeddings_model = "embedding-model-type"
10+
openai_embeddings_dimensions = 1536
11+
openai_completions_deployment = "completions-deployment-name"
12+
openai_completions_model = "completions-model-type"

0 commit comments

Comments
 (0)