How to Seamlessly Use MongoDB Atlas and IBM watsonx.ai LLMs in Your GenAI Applications

1
# watsonx URL
2
3
try:
4
    wxa_url = os.environ["WXA_URL"]
5
6
except KeyError:
7
    wxa_url = getpass.getpass("Please enter your watsonx.ai URL domain (hit enter): ")

1
# watsonx API Key
2
3
try:
4
    wxa_api_key = os.environ["WXA_API_KEY"]
5
except KeyError:
6
    wxa_api_key = getpass.getpass("Please enter your watsonx.ai API key (hit enter): ")

1
# watsonx Project ID
2
3
try:
4
    wxa_project_id = os.environ["WXA_PROJECT_ID"]
5
except KeyError:
6
    wxa_project_id = getpass.getpass("Please enter your watsonx.ai Project ID (hit enter): ")

1
from ibm_watson_machine_learning.foundation_models.utils.enums import ModelTypes
2
from ibm_watson_machine_learning.foundation_models import Model
3
from ibm_watson_machine_learning.metanames import GenTextParamsMetaNames as GenParams
4
from ibm_watson_machine_learning.foundation_models.utils.enums import DecodingMethods
5
6
parameters = {
7
    GenParams.DECODING_METHOD: DecodingMethods.GREEDY,
8
    GenParams.MIN_NEW_TOKENS: 1,
9
    GenParams.MAX_NEW_TOKENS: 100
10
}
11
12
model = Model(
13
    model_id=ModelTypes.GRANITE_13B_INSTRUCT,
14
    params=parameters,
15
    credentials={
16
        "url": wxa_url,
17
        "apikey": wxa_api_key
18
    },
19
    project_id=wxa_project_id
20
)
21
22
from ibm_watson_machine_learning.foundation_models.extensions.langchain import WatsonxLLM
23
24
granite_llm_ibm = WatsonxLLM(model=model)
25
26
# Sample query chosen in the example to evaluate the RAG use case
27
query = "I want to introduce my daughter to science and spark her enthusiasm. What kind of gifts should I get her?"
28
29
# Sample LLM query without RAG framework
30
result = granite_llm_ibm(query)

1
from pymongo import MongoClient
2
3
try:
4
    MONGO_CONN = os.environ["MONGO_CONN"]
5
except KeyError:
6
    MONGO_CONN = getpass.getpass("Please enter your MongoDB connection String (hit enter): ")

1
import wget
2
3
filename = './amazon-products.jsonl'
4
url = "https://github.com/ashwin-gangadhar-mdb/mbd-watson-rag/raw/main/amazon-products.jsonl"
5
6
if not os.path.isfile(filename):
7
    wget.download(url, out=filename)
8
9
# Load the documents using Langchain Document Loader
10
from langchain.document_loaders import JSONLoader
11
12
loader = JSONLoader(file_path=filename, jq_schema=".text",text_content=False,json_lines=True)
13
docs = loader.load()
14
15
# Initialize Embedding for transforming raw documents to vectors**
16
from langchain.embeddings import HuggingFaceEmbeddings
17
from tqdm import tqdm as notebook_tqdm
18
19
embeddings = HuggingFaceEmbeddings()
20
21
# Initialize MongoDB client along with Langchain connector module
22
from langchain.vectorstores import MongoDBAtlasVectorSearch
23
24
client = MongoClient(MONGO_CONN)
25
vcol = client["amazon"]["products"]
26
vectorstore = MongoDBAtlasVectorSearch(vcol, embeddings)
27
28
# Load documents to collection in MongoDB**
29
vectorstore.add_documents(docs)

1
{ 
2
    "mappings": {   
3
        "dynamic": true,   
4
        "fields": {     
5
            "embedding": [       
6
                {         
7
                    "dimensions": 768,         
8
                    "similarity": "cosine",         
9
                    "type": "knnVector"       
10
                }      
11
            ]   
12
        } 
13
    } 
14
} 

1
texts_sim = vectorstore.similarity_search(query, k=3)
2
3
print("Number of relevant texts: " + str(len(texts_sim)))
4
print("First 100 characters of relevant texts.")
5
6
for i in range(len(texts_sim)):
7
    print("Text " + str(i) + ": " + str(texts_sim[i].page_content[0:100]))

1
from langchain.chains import RetrievalQA
2
from langchain.prompts import PromptTemplate
3
4
prompt = PromptTemplate(template="""
5
6
Use the following pieces of context to answer the question at the end. If you don't know the answer, just say that you don't know, don't try to make up an answer.
7
8
{context}
9
10
##Question:{question} \n\
11
12
##Top 3 recommnedations from Products and reason:\n""",input_variables=["context","question"])
13
14
chain_type_kwargs = {"prompt": prompt}
15
retriever = vectorstore.as_retriever(search_type="mmr", search_kwargs={'k': 6, 'lambda_mult': 0.25})
16
17
qa = RetrievalQA.from_chain_type(llm=granite_llm_ibm, chain_type="stuff",
18
                                 retriever=retriever,
19
                                 chain_type_kwargs=chain_type_kwargs)
20
21
res = qa.run(query)
22
23
print(f"{'-'*50}")
24
print("Query: " + query)
25
print(f"Response:\n{res}\n{'-'*50}\n")