Update app.py

Creating the app

app.py

@@ -1,63 +1,173 @@
 import gradio as gr
 from huggingface_hub import InferenceClient
+%%writefile app.py
 
-"""
-For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
-"""
-client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
 
+## Setup
+# Import the necessary Libraries
+import os
+import uuid
+import joblib
+import json
+
+import gradio as gr
+import pandas as pd
+
+from huggingface_hub import CommitScheduler
+from pathlib import Path
+from langchain_community.embeddings.sentence_transformer import (
+    SentenceTransformerEmbeddings
+)
+from langchain_community.vectorstores import Chroma
+from google.colab import userdata, drive
+
+from openai import OpenAI
+
+# Create Client
+client = OpenAI(
+    base_url="https://api.endpoints.anyscale.com/v1",
+    api_key=anyscale_api_key
+)
+
+# Define the embedding model and the vectorstore
+embedding_model_name = 'thenlper/gte-large'
+embedding_model = SentenceTransformerEmbeddings(model_name=embedding_model_name)
+collection_name_qna = 'report_10K_db'
+persisted_vectordb_location = 'report_10K_db'
+# Load the persisted vectorDB
+vectorstore_persisted = Chroma(
+    collection_name=collection_name_qna,
+    persist_directory=persisted_vectordb_location,
+    embedding_function=embedding_model
+)
+
+# Prepare the logging functionality
+
+log_file = Path("logs/") / f"data_{uuid.uuid4()}.json"
+log_folder = log_file.parent
+
+scheduler = CommitScheduler(
+    repo_id="---------",
+    repo_type="dataset",
+    folder_path=log_folder,
+    path_in_repo="data",
+    every=2
+)
+
+# Define the Q&A system message
+qna_system_message = """
+You are an assistant to a financial technology services firm who answers user queries on 10-K reports.
+User input will have the context required by you to answer user questions.
+This context will begin with the token: ###Context.
+The context contains references to specific portions of a document relevant to the user query.
 
-def respond(
-    message,
-    history: list[tuple[str, str]],
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-):
-    messages = [{"role": "system", "content": system_message}]
+User questions will begin with the token: ###Question.
 
-    for val in history:
-        if val[0]:
-            messages.append({"role": "user", "content": val[0]})
-        if val[1]:
-            messages.append({"role": "assistant", "content": val[1]})
+When crafting your response:
+1. Select only context relevant to answer the question.
+2. Include the source links in your response.
+3. User questions will begin with the token: ###Question.
+4. If the question is irrelevant to 10-K respond with - "I am an assistant for 10-K reports. I can only help you with 
 
-    messages.append({"role": "user", "content": message})
+Please adhere to the following guidelines:
+- Your response should only be about the question asked and nothing else.
+- Answer only using the context provided.
+- Do not mention anything about the context in your final answer.
+- If the answer is not found in the context, it is very very important for you to respond with "I don't know. Please check the docs @ '/content/dataset/'"
+- Always quote the source when you use the context. Cite the relevant source at the end of your response under the section - Source:
+- Do not make up sources. Use the links provided in the sources section of the context and nothing else. You are prohibited from providing other links/sources.
 
-    response = ""
+Please answer only using the context provided in the input. Do not mention anything about the context in your final answer.
 
-    for message in client.chat_completion(
-        messages,
-        max_tokens=max_tokens,
-        stream=True,
-        temperature=temperature,
-        top_p=top_p,
-    ):
-        token = message.choices[0].delta.content
+Here is an example of how to structure your response:
 
-        response += token
-        yield response
+Answer:
+[Answer]
+
+Source:
+[Source]
 
 """
-For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
+# Define the user message template
+qna_user_message_template = """
+###Context
+Here are some documents and their source links that are relevant to the question mentioned below.
+{context}
+
+###Question
+{question}
 """
-demo = gr.ChatInterface(
-    respond,
-    additional_inputs=[
-        gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
-        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
-        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
-        gr.Slider(
-            minimum=0.1,
-            maximum=1.0,
-            value=0.95,
-            step=0.05,
-            label="Top-p (nucleus sampling)",
-        ),
-    ],
-)
+# Define the predict function that runs when 'Submit' is clicked or when a API request is made
+def predict(user_input,company):
+
+    filter = "dataset/"+company+"-10-k-2023.pdf"
+    relevant_document_chunks = vectorstore_persisted.similarity_search(user_input, k=5, filter={"source":filter})
+
+    # Create context_for_query
+    user_input = user_input
+    relevant_document_chunks = vectorstore_persisted.similarity_search(user_input,k=5,filter={"source":company})
+    context_list = [d.page_content for d in relevant_document_chunks]
+    context_for_query = ". ".join(context_list)
+
+    # Create messages
+    prompt = [
+              {'role':'system', 'content': qna_system_message},
+              {'role': 'user', 'content': qna_user_message_template.format(
+                  context=context_for_query,
+                  question=user_input
+                  )
+              }
+              ]
+
+    # Get response from the LLM
+    try:
+        response = client.chat.completions.create(
+            model=model_name,
+            messages=prompt,
+            temperature=0
+        )
+
+        prediction = response.choices[0].message.content.strip()
+    except Exception as e:
+        prediction = f'Sorry, I encountered the following error: \n {e}'
 
+    # While the prediction is made, log both the inputs and outputs to a local log file
+    # While writing to the log file, ensure that the commit scheduler is locked to avoid parallel
+    # access
+
+    with scheduler.lock:
+        with log_file.open("a") as f:
+            f.write(json.dumps(
+                {
+                    'user_input': user_input,
+                    'retrieved_context': context_for_query,
+                    'model_response': prediction
+                }
+            ))
+            f.write("\n")
+
+    return prediction
+
+# Set-up the Gradio UI
+# Add text box and radio button to the interface
+# The radio button is used to select the company 10k report in which the context needs to be retrieved.
+
+textbox = gr.Textbox(placeholder="Enter your Query.",lines=6)
+company = gr.Radio(["/content/dataset/Meta-10-k-2023.pdf","/content/dataset/aws-10-k-2023.pdf","/content/dataset/google-10-k-2023.pdf","/content/dataset/IBM-10-k-2023.pdf","/content/dataset/msft-10-k-2023.pdf"], label="Companies Reports")
+
+
+# Create the interface
+# For the inputs parameter of Interface provide [textbox,company]
+demo = gr.Interface(
+    fn=predict,
+    inputs=[textbox,company],
+    outputs="text",
+    title="Information from 10-K reports",
+    description="system to streamline the extraction and analysis of key information from 10-K reports",
+    allow_flagging="auto",
+    concurrency_limit=12
+)
 
 if __name__ == "__main__":
+    demo.queue()
     demo.launch()