Refactoring implementation for ODA RAG POC for NL2SQL use case

LLM/oda_examples/oda-oci-data-science-oracledb-23ai-llm/README.md

@@ -102,10 +102,10 @@ The key prerequisites that you would need to set up before you can proceed to ru
 
 - Notebook session used to initiate the distributed training and to access the fine-tuned model. For more information, see [Notebook session](https://docs.oracle.com/en-us/iaas/data-science/using/manage-notebook-sessions.htm).
 
-- Install the latest version of [Oracle Accelerated Data Science (ADS)](https://accelerated-data-science.readthedocs.io/en/latest/index.html)
+- Install "PyTorch 2.0 for GPU on Python 3.9" conda and install required dependencies mentioned in requirements.txt
 
   ```
-  pip install oracle-ads[opctl] -U
+  pip install -r requirements.txt
   ```
 
 ## Task 1: Deploy Required Models

LLM/oda_examples/oda-oci-data-science-oracledb-23ai-nl2sql-llm/Langchain_NL2SQL_Model_Deployment.ipynb

@@ -251,7 +251,8 @@
     }
    ],
    "source": [
-    "#Copy required python scripts to artifact dir \n",
+    "#Copy required python scripts to artifact dir\n",
+    "#mkdir langchain_nl2sql_model\n",
     "#cp config.py config_private.py oci_utils.py oracle_vector_db.py langchain_nl2sql_model/\n",
     "langchain_model.prepare(\n",
     "        inference_conda_env=CONDA_PACK_PATH,\n",

LLM/oda_examples/oda-oci-data-science-oracledb-23ai-nl2sql-llm/README.md

@@ -102,10 +102,10 @@ The key prerequisites that you would need to set up before you can proceed to ru
 
 - Notebook session used to initiate the distributed training and to access the fine-tuned model. For more information, see [Notebook session](https://docs.oracle.com/en-us/iaas/data-science/using/manage-notebook-sessions.htm).
 
-- Install the latest version of [Oracle Accelerated Data Science (ADS)](https://accelerated-data-science.readthedocs.io/en/latest/index.html)
+- Install "PyTorch 2.0 for GPU on Python 3.9" conda and install required dependencies mentioned in requirements.txt
 
   ```
-  pip install oracle-ads[opctl] -U
+  pip install -r requirements.txt
   ```
 
 ## Task 1: Deploy Required Models

LLM/oda_examples/oda-oci-data-science-oracledb-23ai-nl2sql-llm/langchain_nl2sql_model_score.py

@@ -4,16 +4,16 @@
 
 from langchain_community.llms import OCIModelDeploymentVLLM
 import ads
-from oracle_vector_db import oracle_query, test_oracle_query
+from oracle_vector_db import oracle_query, oracle_query
 from ads.model.generic_model import GenericModel
 import ads
-from oracle_vector_db import oracle_query, test_oracle_query
+from oracle_vector_db import oracle_query, oracle_query
 from langchain.chains import RetrievalQA
 from typing import List
 from langchain_core.callbacks import CallbackManagerForRetrieverRun
 from langchain_core.documents import Document
 from langchain_core.retrievers import BaseRetriever
-from oracle_vector_db import oracle_query, test_oracle_query
+from oracle_vector_db import oracle_query, oracle_query
 from config import COMMAND_MD_ENDPOINT, EMBEDDING_MD_ENDPOINT, RERANKER_MD_ENDPOINT, TOP_K
 from pprint import pprint
 
@@ -57,50 +57,48 @@
 - You should NEVER generate SQL queries with JOIN, since the Schema only includes 1 table.
 - Use Context section to get additional details while bulding the query
 - Do not return multiple queries in response. Just respond with single SQL query and nothing else
-- Return only single query and append response between ###
+- Return only single query and append response between backtick character
+- For date column use TO_DATE function in where clause. e.g. Nomination_date > TO_DATE('2001-01-01', 'YYYY-MM-DD');
 
 Question: {question}
 Oracle SQL: """
 
-relational_data_summary_prompt_template = """We have implmented the natural language question to SQL model which retrieves records from relational database.
-The data retrieved is provided below in the Context section and the corresponding question is provided in the Question section.
-By considering the question and context data can you please summarize the answer. The first row in the context data represents the column headers, and the subsequent rows represent the data.
-Just convert context data to user understandable answer
- 
+relational_data_summary_prompt_template = """In the context section, we provide a question and its corresponding answer. The answer contains a list of rows fetched from a database, where the first row is the column headers, and the subsequent rows are the data.
+Your task is to summarize the answer in a single, human-understandable sentence. If there is data in the context section, assume it is correct and summarize it directly without evaluating its content.
+Do not assume there is only one result unless specified.
+
 ### Context:
 Question: {question}
 Ans:
 {context_data}
 
-### In-Context Examples:
-Question: list or view partners who renewal last year
-Answer: Partner2 has renewal last year
+Can you please summarize the answer based on the data provided in the context? Your summary should be a single, concise sentence.
 
-Can you please summarize the ans mentioned in context section according to question in context? Just return the summarized text. Dont mention 'Based on the context data provided' 
-Ans:"""
+Summary:
+"""
 
 import json
 import oci
 import requests
 
 from langchain_community.llms import OCIModelDeploymentVLLM
 import ads
-from oracle_vector_db import oracle_query, test_oracle_query
+from oracle_vector_db import oracle_query, oracle_query
 from ads.model.generic_model import GenericModel
 import ads
-from oracle_vector_db import oracle_query, test_oracle_query
+from oracle_vector_db import oracle_query, oracle_query
 from langchain.chains import RetrievalQA
 from typing import List
 from langchain_core.callbacks import CallbackManagerForRetrieverRun
 from langchain_core.documents import Document
 from langchain_core.retrievers import BaseRetriever
-from oracle_vector_db import oracle_query, test_oracle_query
+from oracle_vector_db import oracle_query, oracle_query
 from config import COMMAND_MD_ENDPOINT, EMBEDDING_MD_ENDPOINT, RERANKER_MD_ENDPOINT, TOP_K
 from pprint import pprint
 from langchain import PromptTemplate
 from config_private import DB_USER, DB_PWD, DB_HOST_IP, DB_SERVICE
 import oracledb
-
+import re
 
 ads.set_auth("resource_principal")
 command_md = OCIModelDeploymentVLLM(
@@ -122,7 +120,7 @@ def _get_relevant_documents(
         prediction = requests.post(EMBEDDING_MD_ENDPOINT, data=f'["{query}"]', auth=rps)
 
         #Search in DB
-        q_result = test_oracle_query(prediction.json()['embeddings'][0], TOP_K, True, False)        
+        q_result = oracle_query(prediction.json()['embeddings'][0], TOP_K, True, False)        
         text_list = []
         for n, id, sim in zip(q_result.nodes, q_result.ids, q_result.similarities):
             text_list.append(n.text)
@@ -156,7 +154,7 @@ def load_model(model_file_name=model_name):
 
 def get_data_from_DB(query):
     DSN = f"{DB_HOST_IP}/{DB_SERVICE}"
-    connection = oracledb.connect(user=DB_USER, password=DB_PWD, dsn=DSN)
+    connection = oracledb.connect(user=DB_USER, password=DB_PWD, dsn=DSN, config_dir="/opt/oracle/config")
     # Creating a cursor object
     cursor = connection.cursor()
 
@@ -170,7 +168,7 @@ def get_data_from_DB(query):
     for row in cursor:
         row_values = [str(value).replace(',', ' ') for value in row]
         result += ','.join(row_values) + '\n'
-    #print(f"Query Result: {result}")
+    print(f"Query Result: {result}")
     # Closing the cursor and connection
     cursor.close()
     connection.close()
@@ -197,6 +195,13 @@ def predict(data, model=load_model()):
         query_str = op.split("Question:")[0]
     else:
         query_str = op
+
+    match = re.search(r'`([^`]*)`', op)
+    if match:
+        query_str = match.group(1)
+    else:
+        query_str = match
+
     query = query_str.replace(';', '')
     result = get_data_from_DB(query)