added code

app.py

@@ -1 +1,103 @@
-app.
+# for setting/extracting environment variables such as API keysimport os
+
+### 1. For Web Scraping
+# for querying Financial Modelling Prep API
+from urllib.request import urlopen
+import json
+
+### 2. For Converting Scraped Text Into a Vector Store of Chunked Documents
+# for tokenizing texts and splitting them into chunks of documents
+from transformers import GPT2TokenizerFast
+from langchain.text_splitter import RecursiveCharacterTextSplitter
+# for turning documents into embeddings before putting them in vector store
+from langchain.embeddings import HuggingFaceEmbeddings
+# for vector store for documents
+from langchain.vectorstores import Chroma
+
+### 3. For Querying LLM
+# for loading HuggingFace LLM models from the hub
+from langchain.llms import HuggingFaceHub
+# for querying LLM conveniently using the context
+from langchain.chains.question_answering import load_qa_chain
+
+### 4. For Gradio App UI
+import gradio as gr
+
+os.environ['FMP_API_KEY'] = 'your_api_key'
+fmp_api_key = os.environ['FMP_API_KEY']
+
+if os.path.exists("chromadb_earnings_transcripts_extracted"):
+ os.system("rm -r chromadb_earnings_transcripts_extracted")
+if os.path.exists("earnings_transcripts_chromadb.zip"):
+ os.system("rm earnings_transcripts_chromadb.zip")
+
+os.system("wget https://github.com/damianboh/test_earnings_calls/raw/main/earnings_transcripts_chromadb.zip")
+os.system("unzip earnings_transcripts_chromadb.zip -d chromadb_earnings_transcripts_extracted")
+
+# initialize the default model for embedding the tokenized texts, the articles are stored in this embedded form in the vector database
+hf_embeddings = HuggingFaceEmbeddings()
+
+chroma_db = Chroma(persist_directory='chromadb_earnings_transcripts_extracted/chromadb_earnings_transcripts',embedding_function=hf_embeddings)
+
+# Load the huggingface inference endpoint of an LLM model
+# Name of the LLM model we are using, feel free to try others!
+model = "mistralai/Mistral-7B-Instruct-v0.1"
+
+# This is an inference endpoint API from huggingface, the model is not run locally, it is run on huggingface
+hf_llm = HuggingFaceHub(repo_id=model,model_kwargs={'temperature':0.5,"max_new_tokens":300})
+
+def source_question_answer(query,vectorstore:Chroma=chroma_db,llm:HuggingFaceHub=hf_llm):
+ """
+ Return answer to the query
+ """
+ input_docs = vectorstore.similarity_search(query,k=4)
+ qa_chain = load_qa_chain(llm, chain_type="stuff")
+ query = f"[INST]According to the earnings calls transcripts earlier, {query}[INST]"
+
+ response = qa_chain.run(input_documents=input_docs, question=query)
+ source_docs_1 = input_docs[0].page_content
+ source_docs_2 = input_docs[1].page_content
+ source_docs_3 = input_docs[2].page_content
+ source_docs_4 = input_docs[3].page_content
+
+ source_title_1 = input_docs[0].metadata['title']
+ source_title_2 = input_docs[1].metadata['title']
+ source_title_3 = input_docs[2].metadata['title']
+ source_title_4 = input_docs[3].metadata['title']
+
+ return response, source_docs_1, source_docs_2, source_docs_3, source_docs_4, source_title_1, source_title_2, source_title_3, source_title_4
+
+with gr.Blocks() as app:
+ with gr.Row():
+ gr.HTML("<h1>Chat with Tesla 2023 Earnings Calls Transcripts</h1>")
+
+ with gr.Row():
+ query = gr.Textbox("Is Elon happy about Tesla?", placeholder="Enter question here...", label="Enter question")
+ btn = gr.Button("Ask Question")
+
+ with gr.Row():
+ gr.HTML("<h3>Answer</h3>")
+
+ with gr.Row():
+ answer = gr.Textbox(label="Answer")
+
+ with gr.Row():
+ gr.HTML("<h3>Sources Referenced from Tesla 2023 Earnings Calls Transcripts</h3>")
+
+ with gr.Row():
+ with gr.Column():
+ source_title_1 = gr.Markdown()
+ source1 = gr.Textbox(label="Source Text 1")
+ with gr.Column():
+ source_title_2 = gr.Markdown()
+ source2 = gr.Textbox(label="Source Text 2")
+
+ with gr.Row():
+ with gr.Column():
+ source_title_3 = gr.Markdown()
+ source3 = gr.Textbox(label="Source Text 3")
+ with gr.Column():
+ source_title_4 = gr.Markdown()
+ source4 = gr.Textbox(label="Source Text 4")
+
+app.launch()

requirements.txt

@@ -0,0 +1,5 @@
+gradio==3.45.2
+chromadb==0.4.13
+langchain==0.0.305
+transformers==4.33.3
+sentence-transformers==2.2.2