File size: 4,727 Bytes
872912a 4317be9 872912a 9e4cf5c 872912a 9e4cf5c 872912a 9e4cf5c 872912a 9e4cf5c 872912a 9e4cf5c 872912a 4317be9 872912a c1b2280 872912a 5c29f08 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 |
# Import the necessary Libraries
import os
import uuid
import json
import gradio as gr
from openai import OpenAI
from langchain_community.embeddings.sentence_transformer import SentenceTransformerEmbeddings
from langchain_community.vectorstores import Chroma
from huggingface_hub import CommitScheduler
from pathlib import Path
from dotenv import load_dotenv
# Create Client
load_dotenv()
os.environ["ANYSCALE_API_KEY"]=os.getenv("ANYSCALE_API_KEY")
client = OpenAI(
base_url="https://api.endpoints.anyscale.com/v1",
api_key=os.environ['ANYSCALE_API_KEY']
)
embedding_model = SentenceTransformerEmbeddings(model_name='thenlper/gte-large')
# Define the embedding model and the vectorstore
collection_name = 'report-10k-2024'
vectorstore_persisted = Chroma(
collection_name=collection_name,
persist_directory='./dataset-10k',
embedding_function=embedding_model
)
# Load the persisted vectorDB
retriever = vectorstore_persisted.as_retriever(
search_type='similarity',
search_kwargs={'k': 5}
)
# Prepare the logging functionality
log_file = Path("logs/") / f"data_{uuid.uuid4()}.json"
log_folder = log_file.parent
scheduler = CommitScheduler(
repo_id="RAG-investment-recommendation-log",
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 AI assistant to help Finsights Grey Inc., an innovative financial technology firm, develop a Retrieval-Augmented Generation (RAG) system to automate the extraction, summarization, and analysis of information from 10-K reports. Your knowledge base was last updated in August 2023.
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 10-K report relevant to the user query.
User questions will begin with the token: ###Question.
Please answer only using the context provided in the input. Do not mention anything about the context in your final answer.
If the answer is not found in the context, respond, "I don't know".
Here is an example of how to structure your response:
Answer:
[Answer]
Source
[Source]
"""
# Define the user message template
qna_user_message_template = """
###Context
Here are some documents that are relevant to the question.
{context}
```
{question}
```
"""
# 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
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='mistralai/Mixtral-8x7B-Instruct-v0.1',
messages=prompt,
temperature=0
)
prediction = response.choices[0].message.content
except Exception as e:
prediction = 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.
with gr.Blocks() as demo:
with gr.Row():
question = gr.Textbox(label="Enter your question")
company = gr.Radio(["aws", "IBM", "google", "meta", "msft"], label="Select a company")
submit = gr.Button("Submit")
output = gr.Textbox(label="Output")
submit.click(
fn=predict,
inputs=[question, company],
outputs=output
)
demo.launch()
# Create the interface
# For the inputs parameter of Interface, provide [textbox, company]
demo.queue()
demo.launch(auth=("demouser", os.getenv('PASSWD'))) |