initial commit

.gitignore

@@ -0,0 +1 @@
+.env

app.py

@@ -1,27 +1,51 @@
 from gradio_client import Client
 import gradio as gr
+import requests
+from langchain.chains import RetrievalQA
+from langchain.vectorstores import Pinecone
+import pinecone
+from langchain.qa import RetrievalQA
+from langchain.vectorstore import vectorstore
+import os
+from langchain.embeddings.huggingface import HuggingFaceEmbeddings
 
-chat_client = Client("https://mosaicml-mpt-30b-chat.hf.space/", serialize = False)
-retrieval = Client("https://ishaan-mital-ncert-helper-vector-db.hf.space/--replicas/149bg26k5/")
+API_URL = "https://api-inference.huggingface.co/models/HuggingFaceH4/zephyr-7b-beta"
+# retrieval = Client("https://ishaan-mital-ncert-helper-vector-db.hf.space/--replicas/149bg26k5/")
 
-init_prompt ="## Instruction: You are an AI language model and must return truthful responses as per the information. Do not answer with any information which isn't completely verified and correct. Do not lie. Do not present information where you don't know the answer. Do not include incorrect extra information. Your name is NCERT-Helper. You are a helpful and truthful chatbot. You can help create questions(MCQs) for students in their 11th and 12th standard CBSE Board."
-info="Information: \n"
-# q_prompt="\n ##Instruction: Please provide an appropriate response to the following in less than 3 lines: \n"  
-chatbot = [["", None]]
+embed_model_id = 'sentence-transformers/all-MiniLM-L6-v2'
+embed_model = HuggingFaceEmbeddings(
+    model_name=embed_model_id,
+)
+pinecone.init(
+    api_key=os.environ.get('PINECONE_API_KEY'),
+    environment=os.environ.get('PINECONE_ENVIRONMENT')
+)
+
+index_name = 'llama-rag'
+index = pinecone.Index(index_name)
+text_field = 'text'  # field in metadata that contains text content
+
+vectorstore = Pinecone(
+    index, embed_model.embed_query, text_field
+)
+
+def call_llm_api(input_text):
+    headers = {"Authorization": f"Bearer {os.environ.get('API_KEY')}"}
+    payload = {"imput": input_text}
+    response = requests.post(API_URL, headers=headers, json=payload)
+    return response.json()  # Adjust as needed based on your API response format
+	
+rag_pipeline = RetrievalQA.from_chain_type(
+    llm=call_llm_api, chain_type='stuff',
+    retriever=vectorstore.as_retriever()
+)
 
 
 def main(question):
     global chatbot
-    information = retrieval.predict(question, api_name = "/predict")
-    answer=chat_client.predict(
-        info +information+question, # str  in 'Type an input and press Enter' Textbox component
-        chatbot,
-        fn_index=1
-    )
-    chatbot = answer[1]
-    return answer[1][0][1]
+    return rag_pipeline(question)
 
 demo = gr.Interface(main, inputs = "text", outputs = "text")
 
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch(share=True)

requirements.txt

@@ -1,4 +1,9 @@
 hugchat
 gradio
 gradio_client
-gtts
+gtts
+pydantic==1.10.9
+langchain
+pinecone-client==2.2.2
+faiss-cpu
+sentence_transformers