Merge pull request #16 from lpm0073/next

bug fix to rag() and stronger typing

CHANGELOG.md

@@ -1,9 +1,8 @@
 ## [1.1.2](https://github.com/lpm0073/hybrid-search-retriever/compare/v1.1.1...v1.1.2) (2023-12-01)
 
-
 ### Bug Fixes
 
-* syntax error in examples.prompt ([230b709](https://github.com/lpm0073/hybrid-search-retriever/commit/230b7090c96bdd4d7d8757b182f891ab1b82c6f4))
+- syntax error in examples.prompt ([230b709](https://github.com/lpm0073/hybrid-search-retriever/commit/230b7090c96bdd4d7d8757b182f891ab1b82c6f4))
 
 ## [1.1.1](https://github.com/lpm0073/netec-llm/compare/v1.1.0...v1.1.1) (2023-12-01)
 

README.md

@@ -46,7 +46,7 @@ python3 -m models.examples.training_services_oracle "Oracle database administrat
 python3 -m models.examples.load "./data/"
 
 # example 6 - Retrieval Augmented Generation
-python3 -m models.examples.rag "What is Accounting Based Valuation?"
+python3 -m models.examples.rag "What analytics and accounting courses does Wharton offer?"
 ```
 
 ## Setup

models/__version__.py

@@ -1,2 +1,2 @@
 # -*- coding: utf-8 -*-
-__version__ = "1.1.2"
+__version__ = "1.1.3"

models/const.py

@@ -21,7 +21,7 @@
     OPENAI_CHAT_TEMPERATURE = float(os.environ.get("OPENAI_CHAT_TEMPERATURE", 0.0))
     OPENAI_CHAT_MAX_RETRIES = int(os.environ.get("OPENAI_CHAT_MAX_RETRIES", 3))
     OPENAI_CHAT_CACHE = bool(os.environ.get("OPENAI_CHAT_CACHE", True))
-    DEBUG_MODE = bool(os.environ.get("DEBUG_MODE", False))
+    DEBUG_MODE = os.environ.get("DEBUG_MODE", "False") == "True"
 else:
     raise FileNotFoundError("No .env file found in root directory of repository")
 

models/examples/prompt.py

@@ -2,6 +2,8 @@
 """Sales Support Model (hsr)"""
 import argparse
 
+from langchain.schema import HumanMessage, SystemMessage
+
 from models.hybrid_search_retreiver import HybridSearchRetriever
 
 
@@ -10,9 +12,11 @@
 
 if __name__ == "__main__":
     parser = argparse.ArgumentParser(description="hsr examples")
-    parser.add_argument("system_prompt", type=str, help="A system prompt to send to the model.")
-    parser.add_argument("human_prompt", type=str, help="A human prompt to send to the model.")
+    parser.add_argument("system_message", type=str, help="A system prompt to send to the model.")
+    parser.add_argument("human_message", type=str, help="A human prompt to send to the model.")
     args = parser.parse_args()
 
-    result = hsr.cached_chat_request(args.system_prompt, args.human_prompt)
-    print(result)
+    system_message = SystemMessage(text=args.system_message)
+    human_message = HumanMessage(text=args.human_message)
+    result = hsr.cached_chat_request(system_message=system_message, human_message=human_message)
+    print(result.content)

models/examples/rag.py

@@ -2,6 +2,8 @@
 """Sales Support Model (hsr) Retrieval Augmented Generation (RAG)"""
 import argparse
 
+from langchain.schema import HumanMessage
+
 from models.hybrid_search_retreiver import HybridSearchRetriever
 
 
@@ -12,5 +14,6 @@
     parser.add_argument("prompt", type=str, help="A question about the PDF contents")
     args = parser.parse_args()
 
-    result = hsr.rag(prompt=args.prompt)
+    human_message = HumanMessage(text=args.prompt)
+    result = hsr.rag(human_message=human_message)
     print(result)

models/hybrid_search_retreiver.py

@@ -18,9 +18,12 @@
 
 # document loading
 import glob
+
+# general purpose imports
 import logging
 import os
 import textwrap
+from typing import Union
 
 # pinecone integration
 import pinecone
@@ -38,7 +41,7 @@
 
 # hybrid search capability
 from langchain.retrievers import PineconeHybridSearchRetriever
-from langchain.schema import HumanMessage, SystemMessage
+from langchain.schema import BaseMessage, HumanMessage, SystemMessage
 from langchain.text_splitter import Document
 from langchain.vectorstores.pinecone import Pinecone
 from pinecone_text.sparse import BM25Encoder
@@ -95,14 +98,20 @@ class HybridSearchRetriever:
     text_splitter = TextSplitter()
     bm25_encoder = BM25Encoder().default()
 
-    def cached_chat_request(self, system_message: str, human_message: str) -> SystemMessage:
+    def cached_chat_request(
+        self, system_message: Union[str, SystemMessage], human_message: Union[str, HumanMessage]
+    ) -> BaseMessage:
         """Cached chat request."""
-        messages = [
-            SystemMessage(content=system_message),
-            HumanMessage(content=human_message),
-        ]
+        if not isinstance(system_message, SystemMessage):
+            logging.debug("Converting system message to SystemMessage")
+            system_message = SystemMessage(content=str(system_message))
+
+        if not isinstance(human_message, HumanMessage):
+            logging.debug("Converting human message to HumanMessage")
+            human_message = HumanMessage(content=str(human_message))
+        messages = [system_message, human_message]
         # pylint: disable=not-callable
-        retval = self.chat(messages).content
+        retval = self.chat(messages)
         return retval
 
     def prompt_with_template(self, prompt: PromptTemplate, concept: str, model: str = DEFAULT_MODEL_NAME) -> str:
@@ -158,10 +167,10 @@ def load(self, filepath: str):
 
         logging.debug("Finished loading PDFs")
 
-    def rag(self, prompt: str):
+    def rag(self, human_message: Union[str, HumanMessage]):
         """
         Embedded prompt.
-        1. Retrieve prompt: Given a user input, relevant splits are retrieved
+        1. Retrieve human message prompt: Given a user input, relevant splits are retrieved
            from storage using a Retriever.
         2. Generate: A ChatModel / LLM produces an answer using a prompt that includes
            the question and the retrieved data
@@ -174,33 +183,32 @@ def rag(self, prompt: str):
         The typical workflow is to use the embeddings to retrieve relevant documents,
         and then use the text of these documents as part of the prompt for GPT-3.
         """
+        if not isinstance(human_message, HumanMessage):
+            logging.debug("Converting human_message to HumanMessage")
+            human_message = HumanMessage(content=human_message)
+
         retriever = PineconeHybridSearchRetriever(
             embeddings=self.openai_embeddings, sparse_encoder=self.bm25_encoder, index=self.pinecone_index
         )
-        documents = retriever.get_relevant_documents(query=prompt)
+        documents = retriever.get_relevant_documents(query=human_message.content)
         logging.debug("Retrieved %i related documents from Pinecone", len(documents))
 
         # Extract the text from the documents
         document_texts = [doc.page_content for doc in documents]
         leader = textwrap.dedent(
-            """\
-            \n\nYou can assume that the following is true.
+            """You are a helpful assistant.
+            You can assume that all of the following is true.
             You should attempt to incorporate these facts
-            into your response:\n\n
+            into your responses:\n\n
         """
         )
+        system_message = f"{leader} {'. '.join(document_texts)}"
 
-        # Create a prompt that includes the document texts
-        prompt_with_relevant_documents = f"{prompt + leader} {'. '.join(document_texts)}"
-
-        logging.debug("Prompt contains %i words", len(prompt_with_relevant_documents.split()))
-        logging.debug("Prompt: %s", prompt_with_relevant_documents)
-
-        # Get a response from the GPT-3.5-turbo model
-        response = self.cached_chat_request(
-            system_message="You are a helpful assistant.", human_message=prompt_with_relevant_documents
-        )
+        logging.debug("System messages contains %i words", len(system_message.split()))
+        logging.debug("Prompt: %s", system_message)
+        system_message = SystemMessage(content=system_message)
+        response = self.cached_chat_request(system_message=system_message, human_message=human_message)
 
         logging.debug("Response:")
         logging.debug("------------------------------------------------------")
-        return response
+        return response.content

models/tests/test_examples.py

@@ -6,6 +6,7 @@
 from unittest.mock import MagicMock, patch
 
 import pytest  # pylint: disable=unused-import
+from langchain.schema import HumanMessage, SystemMessage
 
 from models.examples.prompt import hsr as prompt_hrs
 from models.examples.rag import hsr as rag_hsr
@@ -14,7 +15,7 @@
 from models.prompt_templates import NetecPromptTemplates
 
 
-HUMAN_PROMPT = 'return the word "SUCCESS" in upper case.'
+HUMAN_MESSAGE = 'return the word "SUCCESS" in upper case.'
 
 
 class TestExamples:
@@ -25,44 +26,47 @@ def test_prompt(self, mock_parse_args):
         """Test prompt example."""
         mock_args = MagicMock()
         mock_args.system_prompt = "you are a helpful assistant"
-        mock_args.human_prompt = HUMAN_PROMPT
+        mock_args.human_prompt = HUMAN_MESSAGE
         mock_parse_args.return_value = mock_args
 
-        result = prompt_hrs.cached_chat_request(mock_args.system_prompt, mock_args.human_prompt)
-        assert result == "SUCCESS"
+        system_message = SystemMessage(content="you are a helpful assistant")
+        human_message = HumanMessage(content=HUMAN_MESSAGE)
+        result = prompt_hrs.cached_chat_request(system_message=system_message, human_message=human_message)
+        assert result.content == "SUCCESS"
 
     @patch("argparse.ArgumentParser.parse_args")
     def test_rag(self, mock_parse_args):
         """Test RAG example."""
         mock_args = MagicMock()
-        mock_args.human_prompt = HUMAN_PROMPT
+        mock_args.human_message = HUMAN_MESSAGE
         mock_parse_args.return_value = mock_args
 
-        result = rag_hsr.rag(mock_args.human_prompt)
+        human_message = HumanMessage(content=mock_args.human_message)
+        result = rag_hsr.rag(human_message=human_message)
         assert result == "SUCCESS"
 
     @patch("argparse.ArgumentParser.parse_args")
     def test_training_services(self, mock_parse_args):
         """Test training services templates."""
         mock_args = MagicMock()
-        mock_args.human_prompt = HUMAN_PROMPT
+        mock_args.human_message = HUMAN_MESSAGE
         mock_parse_args.return_value = mock_args
 
         templates = NetecPromptTemplates()
         prompt = templates.training_services
 
-        result = training_services_hsr.prompt_with_template(prompt=prompt, concept=mock_args.human_prompt)
+        result = training_services_hsr.prompt_with_template(prompt=prompt, concept=mock_args.human_message)
         assert "SUCCESS" in result
 
     @patch("argparse.ArgumentParser.parse_args")
     def test_oracle_training_services(self, mock_parse_args):
         """Test oracle training services."""
         mock_args = MagicMock()
-        mock_args.human_prompt = HUMAN_PROMPT
+        mock_args.human_message = HUMAN_MESSAGE
         mock_parse_args.return_value = mock_args
 
         templates = NetecPromptTemplates()
         prompt = templates.oracle_training_services
 
-        result = training_services_oracle_hsr.prompt_with_template(prompt=prompt, concept=mock_args.human_prompt)
+        result = training_services_oracle_hsr.prompt_with_template(prompt=prompt, concept=mock_args.human_message)
         assert "SUCCESS" in result

models/tests/test_openai.py

@@ -19,4 +19,4 @@ def test_03_test_openai_connectivity(self):
         retval = hsr.cached_chat_request(
             "your are a helpful assistant", "please return the value 'CORRECT' in all upper case."
         )
-        assert retval == "CORRECT"
+        assert retval.content == "CORRECT"

requirements.txt

@@ -21,6 +21,7 @@ codespell==2.2.6
 # ------------
 langchain==0.0.343
 langchainhub==0.1.14
+langchain-experimental==0.0.43
 openai==1.3.5
 pinecone-client==2.2.4
 pinecone-text==0.7.0