pdf uploader

app.py

@@ -11,14 +11,14 @@ from langchain_core.output_parsers import StrOutputParser
 from langchain_ollama import ChatOllama
 from langchain_core.documents import Document
 from langchain_community.llms import HuggingFacePipeline
-
+from langchain_community.document_loaders import PyPDFLoader
+from langchain.text_splitter import RecursiveCharacterTextSplitter
 
 # Initialize in-memory ChromaDB client
-# This client runs entirely within the app.py script.
 client = chromadb.Client()
 
 # Load your embeddings model
-model_kwargs = {"device": "cpu"} # Hugging Face Spaces typically use CPU for free tiers
+model_kwargs = {"device": "cpu"}
 encode_kwargs = {"normalize_embeddings": True}
 embeddings = HuggingFaceEmbeddings(
     model_name="sentence-transformers/paraphrase-distilroberta-base-v1",
@@ -27,18 +27,28 @@ embeddings = HuggingFaceEmbeddings(
 )
 
 # Initialize the vector DB using the in-memory client
-# You'll need to embed your documents here. In a real-world app, you'd load them from a file.
-# For a demo, let's create a dummy document.
 vectorDB = Chroma(
     client=client,
     collection_name="embeddings",
     embedding_function=embeddings,
 )
-# Example of adding a document. You would replace this with your actual documents.
-sample_doc = "This is a sample document about the history of artificial intelligence. It was created to demonstrate the RAG pipeline."
-vectorDB.add_documents([Document(page_content=sample_doc, metadata={"docId": "my_doc_id"})])
 
-# Your existing functions without the HttpClient call
+# Function to process and ingest a PDF file
+def process_pdf(file_path):
+    # Use PyPDFLoader to load the PDF
+    loader = PyPDFLoader(file_path)
+    documents = loader.load()
+    
+    # Split the documents for better retrieval
+    text_splitter = RecursiveCharacterTextSplitter(chunk_size=400, chunk_overlap=30)
+    texts = text_splitter.split_documents(documents)
+    
+    # Ingest into the vector store
+    # Note: A docId is added to group these documents
+    vectorDB.add_documents(texts, [{"docId": "uploaded_doc"}] * len(texts))
+    gr.Info("PDF processed and ready for questions!")
+
+# Your existing functions
 def using_ollama_model(retriever, query, results, conversation_history):
     history_text = ""
     for item in conversation_history:
@@ -60,8 +70,7 @@ def using_ollama_model(retriever, query, results, conversation_history):
     )
     
     doc_texts = "\\n".join([doc.page_content for doc in results])
-    model_id = "meta-llama/Llama-3.2-1B-Instruct" # was llama3.2
-    # llm = ChatOllama(model="llama3.2", temperature=0.4, num_predict=512)
+    model_id = "meta-llama/Llama-3.2-1B-Instruct"
     tokenizer = AutoTokenizer.from_pretrained(model_id)
     model = AutoModelForCausalLM.from_pretrained(model_id)
     pipe = pipeline(
@@ -75,10 +84,8 @@ def using_ollama_model(retriever, query, results, conversation_history):
         repetition_penalty=1.2
     )
 
-     # Use the pipeline with LangChain's HuggingFacePipeline
     llm = HuggingFacePipeline(pipeline=pipe)
 
-    
     rag_chain = template | llm | StrOutputParser()
     
     answer = rag_chain.invoke({"history": history_text, "results": doc_texts, "query": query})
@@ -107,22 +114,48 @@ def retrievingReponse(docId, query, conversation_history):
     llm_result = using_ollama_model(retriever, query, results, conversation_history)
     return llm_result
 
-# Gradio interface
-def gradio_rag_wrapper(query, history):
+# The revised Gradio wrapper function
+def gradio_rag_wrapper(message, history):
+    # Check if a file has been uploaded
+    # 'message' is a dictionary due to `multimodal=True`
+    uploaded_files = message.get("files", [])
+    
+    # Process the PDF if it exists
+    if uploaded_files:
+        for file_path in uploaded_files:
+            process_pdf(file_path)
+            # Return a message to confirm the upload
+            return "PDF uploaded and processed. You can now ask questions about the content."
+
+    # Process the text query
+    text_query = message.get("text", "")
+    if not text_query.strip():
+        # Handle cases where only a file was uploaded
+        return "Please upload a document or enter a text query."
+
     rag_history = []
     for user_msg, bot_msg in history:
-        rag_history.append({"question": user_msg, "answer": bot_msg})
+        # Note: You need to extract the text from user messages which may contain files
+        user_text = user_msg.get("text", "") if isinstance(user_msg, dict) else user_msg
+        
+        rag_history.append({"question": user_text, "answer": bot_msg})
 
-    docId = "my_doc_id"
-    response = retrievingReponse(docId, query, rag_history)
+    docId = "uploaded_doc" # Use the docId from the uploaded file
+    response = retrievingReponse(docId, text_query, rag_history)
     
     return response
 
+# Create the Gradio interface with multimodal input
 demo = gr.ChatInterface(
     fn=gradio_rag_wrapper,
+    multimodal=True, # This enables file upload
     title="Contextual RAG Chatbot on Hugging Face Spaces",
-    description="Ask questions about the document to get answers.",
+    description="Upload a PDF file to start chatting!",
+    textbox=gr.MultimodalTextbox(file_types=[".pdf"]), # Restrict file types
 )
 
 if __name__ == "__main__":
-    demo.launch()
+    # Create a dummy doc for initial testing if no PDF is uploaded
+    vectorDB.add_documents([Document(page_content="This is a sample document about the history of artificial intelligence. It was created to demonstrate the RAG pipeline.", metadata={"docId": "uploaded_doc"})])
+    
+    demo.launch()