Llama-4-RAG / main.py
Abid Ali Awan
first commit
355b607
raw
history blame
15.2 kB
import os
import zipfile
from typing import Dict, List, Optional, Union
import gradio as gr
from groq import Groq
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_core.output_parsers import StrOutputParser
from langchain_core.prompts import PromptTemplate
from langchain_core.runnables import RunnablePassthrough
from langchain_groq import ChatGroq
from langchain_huggingface import HuggingFaceEmbeddings
from langchain_core.vectorstores import InMemoryVectorStore
# Retrieve API key for Groq from the environment variables
GROQ_API_KEY = os.environ.get("GROQ_API_KEY")
# Initialize the Groq client
client = Groq(api_key=GROQ_API_KEY)
# Initialize the LLM
llm = ChatGroq(model="meta-llama/llama-4-scout-17b-16e-instruct", api_key=GROQ_API_KEY)
# Initialize the embedding model
embed_model = HuggingFaceEmbeddings(model_name="mixedbread-ai/mxbai-embed-large-v1")
# General constants for the UI
TITLE = """<h1 align="center">✨ Llama 4 RAG Application</h1>"""
AVATAR_IMAGES = (
None,
"https://huggingface.co/datasets/huggingface/brand-assets/resolve/main/hf-logo.png",
)
# List of supported text extensions (alphabetically sorted)
TEXT_EXTENSIONS = [
".bat",
".c",
".cfg",
".conf",
".cpp",
".cs",
".css",
".docx",
".go",
".h",
".html",
".ini",
".java",
".js",
".json",
".jsx",
".md",
".php",
".ps1",
".py",
".rb",
".rs",
".sh",
".toml",
".ts",
".tsx",
".txt",
".xml",
".yaml",
".yml",
]
# Global variables
EXTRACTED_FILES = {}
VECTORSTORE = None
RAG_CHAIN = None
# Initialize the text splitter
text_splitter = RecursiveCharacterTextSplitter(
chunk_size=1000, chunk_overlap=100, separators=["\n\n", "\n"]
)
# Define the RAG prompt template
template = """You are an expert assistant tasked with answering questions based on the provided documents.
Use only the given context to generate your answer.
If the answer cannot be found in the context, clearly state that you do not know.
Be detailed and precise in your response, but avoid mentioning or referencing the context itself.
Context:
{context}
Question:
{question}
Answer:"""
# Create the PromptTemplate
rag_prompt = PromptTemplate.from_template(template)
def extract_text_from_zip(zip_file_path: str) -> Dict[str, str]:
"""
Extract text content from files in a ZIP archive.
Parameters:
zip_file_path (str): Path to the ZIP file.
Returns:
Dict[str, str]: Dictionary mapping filenames to their text content.
"""
text_contents = {}
with zipfile.ZipFile(zip_file_path, "r") as zip_ref:
for file_info in zip_ref.infolist():
# Skip directories
if file_info.filename.endswith("/"):
continue
# Skip binary files and focus on text files
file_ext = os.path.splitext(file_info.filename)[1].lower()
if file_ext in TEXT_EXTENSIONS:
try:
with zip_ref.open(file_info) as file:
content = file.read().decode("utf-8", errors="replace")
text_contents[file_info.filename] = content
except Exception as e:
text_contents[file_info.filename] = (
f"Error extracting file: {str(e)}"
)
return text_contents
def extract_text_from_single_file(file_path: str) -> Dict[str, str]:
"""
Extract text content from a single file.
Parameters:
file_path (str): Path to the file.
Returns:
Dict[str, str]: Dictionary mapping filename to its text content.
"""
text_contents = {}
filename = os.path.basename(file_path)
file_ext = os.path.splitext(filename)[1].lower()
if file_ext in TEXT_EXTENSIONS:
try:
with open(file_path, "r", encoding="utf-8", errors="replace") as file:
content = file.read()
text_contents[filename] = content
except Exception as e:
text_contents[filename] = f"Error reading file: {str(e)}"
return text_contents
def upload_files(
files: Optional[List[str]], chatbot: List[Union[dict, gr.ChatMessage]]
):
"""
Process uploaded files (ZIP or single text files): extract text content and append a message to the chat.
Parameters:
files (Optional[List[str]]): List of file paths.
chatbot (List[Union[dict, gr.ChatMessage]]): The conversation history.
Returns:
List[Union[dict, gr.ChatMessage]]: Updated conversation history.
"""
global EXTRACTED_FILES, VECTORSTORE, RAG_CHAIN
# Handle multiple file uploads
if len(files) > 1:
total_files_processed = 0
total_files_extracted = 0
file_types = set()
# Process each file
for file in files:
filename = os.path.basename(file)
file_ext = os.path.splitext(filename)[1].lower()
# Process based on file type
if file_ext == ".zip":
extracted_files = extract_text_from_zip(file)
file_types.add("zip")
else:
extracted_files = extract_text_from_single_file(file)
file_types.add("text")
if extracted_files:
total_files_extracted += len(extracted_files)
# Store the extracted content in the global variable
EXTRACTED_FILES[filename] = extracted_files
total_files_processed += 1
# Create a summary message for multiple files
file_types_str = (
"files"
if len(file_types) > 1
else ("ZIP files" if "zip" in file_types else "text files")
)
# Create a list of uploaded file names
file_list = "\n".join([f"- {os.path.basename(file)}" for file in files])
chatbot.append(
gr.ChatMessage(
role="user",
content=f"<p>πŸ“š Multiple {file_types_str} uploaded ({total_files_processed} files)</p><p>Extracted {total_files_extracted} text file(s) in total</p><p>Uploaded files:</p><pre>{file_list}</pre>",
)
)
# Handle single file upload
elif len(files) == 1:
file = files[0]
filename = os.path.basename(file)
file_ext = os.path.splitext(filename)[1].lower()
# Process based on file type
if file_ext == ".zip":
extracted_files = extract_text_from_zip(file)
file_type_msg = "πŸ“¦ ZIP file"
else:
extracted_files = extract_text_from_single_file(file)
file_type_msg = "πŸ“„ File"
if not extracted_files:
chatbot.append(
gr.ChatMessage(
role="user",
content=f"<p>{file_type_msg} uploaded: {filename}, but no text content was found or the file format is not supported.</p>",
)
)
else:
file_list = "\n".join([f"- {name}" for name in extracted_files.keys()])
chatbot.append(
gr.ChatMessage(
role="user",
content=f"<p>{file_type_msg} uploaded: {filename}</p><p>Extracted {len(extracted_files)} text file(s):</p><pre>{file_list}</pre>",
)
)
# Store the extracted content in the global variable
EXTRACTED_FILES[filename] = extracted_files
# Process the extracted files and create vector embeddings
if EXTRACTED_FILES:
# Prepare documents for processing
all_texts = []
for filename, files in EXTRACTED_FILES.items():
for file_path, content in files.items():
all_texts.append(
{"page_content": content, "metadata": {"source": file_path}}
)
# Create document objects
from langchain_core.documents import Document
documents = [
Document(page_content=item["page_content"], metadata=item["metadata"])
for item in all_texts
]
# Split the documents into chunks
chunks = text_splitter.split_documents(documents)
# Create the vector store
VECTORSTORE = InMemoryVectorStore.from_documents(
documents=chunks,
embedding=embed_model,
)
# Create the retriever
retriever = VECTORSTORE.as_retriever()
# Create the RAG chain
RAG_CHAIN = (
{"context": retriever, "question": RunnablePassthrough()}
| rag_prompt
| llm
| StrOutputParser()
)
# Add a confirmation message
chatbot.append(
gr.ChatMessage(
role="assistant",
content="Documents processed and indexed. You can now ask questions about the content.",
)
)
return chatbot
def user(text_prompt: str, chatbot: List[gr.ChatMessage]):
"""
Append a new user text message to the chat history.
Parameters:
text_prompt (str): The input text provided by the user.
chatbot (List[gr.ChatMessage]): The existing conversation history.
Returns:
Tuple[str, List[gr.ChatMessage]]: A tuple of an empty string (clearing the prompt)
and the updated conversation history.
"""
if text_prompt:
chatbot.append(gr.ChatMessage(role="user", content=text_prompt))
return "", chatbot
def get_message_content(msg):
"""
Retrieve the content of a message that can be either a dictionary or a gr.ChatMessage.
Parameters:
msg (Union[dict, gr.ChatMessage]): The message object.
Returns:
str: The textual content of the message.
"""
if isinstance(msg, dict):
return msg.get("content", "")
return msg.content
def process_query(chatbot: List[Union[dict, gr.ChatMessage]]):
"""
Process the user's query using the RAG pipeline.
Parameters:
chatbot (List[Union[dict, gr.ChatMessage]]): The conversation history.
Returns:
List[Union[dict, gr.ChatMessage]]: The updated conversation history with the response.
"""
global RAG_CHAIN
if len(chatbot) == 0:
chatbot.append(
gr.ChatMessage(
role="assistant",
content="Please enter a question or upload documents to start the conversation.",
)
)
return chatbot
# Get the last user message as the prompt
user_messages = [
msg
for msg in chatbot
if (isinstance(msg, dict) and msg.get("role") == "user")
or (hasattr(msg, "role") and msg.role == "user")
]
if not user_messages:
chatbot.append(
gr.ChatMessage(
role="assistant",
content="Please enter a question to start the conversation.",
)
)
return chatbot
last_user_msg = user_messages[-1]
prompt = get_message_content(last_user_msg)
# Skip if the last message was about uploading a file
if (
"πŸ“¦ ZIP file uploaded:" in prompt
or "πŸ“„ File uploaded:" in prompt
or "πŸ“š Multiple files uploaded" in prompt
):
return chatbot
# Check if RAG chain is available
if RAG_CHAIN is None:
chatbot.append(
gr.ChatMessage(
role="assistant",
content="Please upload documents first to enable question answering.",
)
)
return chatbot
# Append a placeholder for the assistant's response
chatbot.append(gr.ChatMessage(role="assistant", content="Thinking..."))
try:
# Process the query through the RAG chain
response = RAG_CHAIN.invoke(prompt)
# Update the placeholder with the actual response
chatbot[-1].content = response
except Exception as e:
# Handle any errors
chatbot[-1].content = f"Error processing your query: {str(e)}"
return chatbot
def reset_app(chatbot):
"""
Reset the app by clearing the chat context and removing any uploaded files.
Parameters:
chatbot (List[Union[dict, gr.ChatMessage]]): The conversation history.
Returns:
List[Union[dict, gr.ChatMessage]]: A fresh conversation history.
"""
global EXTRACTED_FILES, VECTORSTORE, RAG_CHAIN
# Clear the global variables
EXTRACTED_FILES = {}
VECTORSTORE = None
RAG_CHAIN = None
# Reset the chatbot with a welcome message
return [
gr.ChatMessage(
role="assistant",
content="App has been reset. You can start a new conversation or upload new documents.",
)
]
# Define the Gradio UI components
chatbot_component = gr.Chatbot(
label="Llama 4 RAG",
type="messages",
bubble_full_width=False,
avatar_images=AVATAR_IMAGES,
scale=2,
height=350,
)
text_prompt_component = gr.Textbox(
placeholder="Ask a question about your documents...",
show_label=False,
autofocus=True,
scale=28,
)
upload_files_button_component = gr.UploadButton(
label="Upload",
file_count="multiple",
file_types=[".zip", ".docx"] + TEXT_EXTENSIONS,
scale=1,
min_width=80,
)
send_button_component = gr.Button(
value="Send", variant="primary", scale=1, min_width=80
)
reset_button_component = gr.Button(value="Reset", variant="stop", scale=1, min_width=80)
# Define input lists for button chaining
user_inputs = [text_prompt_component, chatbot_component]
with gr.Blocks(theme=gr.themes.Ocean()) as demo:
gr.HTML(TITLE)
with gr.Column():
chatbot_component.render()
with gr.Row(equal_height=True):
text_prompt_component.render()
send_button_component.render()
upload_files_button_component.render()
reset_button_component.render()
# When the Send button is clicked, first process the user text then process the query
send_button_component.click(
fn=user,
inputs=user_inputs,
outputs=[text_prompt_component, chatbot_component],
queue=False,
).then(
fn=process_query,
inputs=[chatbot_component],
outputs=[chatbot_component],
api_name="process_query",
)
# Allow submission using the Enter key
text_prompt_component.submit(
fn=user,
inputs=user_inputs,
outputs=[text_prompt_component, chatbot_component],
queue=False,
).then(
fn=process_query,
inputs=[chatbot_component],
outputs=[chatbot_component],
api_name="process_query_submit",
)
# Handle file uploads
upload_files_button_component.upload(
fn=upload_files,
inputs=[upload_files_button_component, chatbot_component],
outputs=[chatbot_component],
queue=False,
)
# Handle Reset button clicks
reset_button_component.click(
fn=reset_app,
inputs=[chatbot_component],
outputs=[chatbot_component],
queue=False,
)
# Launch the demo interface
demo.queue().launch()