app.py

import os
import chromadb
import gradio as gr
from transformers import AutoModelForSeq2SeqLM, AutoTokenizer, pipeline
from langchain_chroma import Chroma
from langchain_huggingface import HuggingFaceEmbeddings, HuggingFacePipeline
from langchain.document_loaders import PyPDFLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.chains import create_retrieval_chain, LLMChain
from langchain.prompts import PromptTemplate
from collections import OrderedDict

# Load embeddings model
embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")

# Load Chroma database (Avoid reprocessing documents)
CHROMA_PATH = "./chroma_db"
if not os.path.exists(CHROMA_PATH):
    raise FileNotFoundError("ChromaDB folder not found. Make sure it's uploaded to the repo.")

chroma_client = chromadb.PersistentClient(path=CHROMA_PATH)
db = Chroma(embedding_function=embeddings, client=chroma_client)

# Load the model
model_name = "google/flan-t5-large"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSeq2SeqLM.from_pretrained(model_name)

# Create pipeline
qa_pipeline = pipeline(
    "text2text-generation",
    model=model,
    tokenizer=tokenizer,
    device=0,
    max_length=512,
    min_length=50,
    do_sample=False,
    repetition_penalty=1.2
)

# Wrap pipeline in LangChain
llm = HuggingFacePipeline(pipeline=qa_pipeline)
retriever = db.as_retriever(search_kwargs={"k": 3})


def clean_context(context_list, max_tokens=350, min_length=50):
    """
    Cleans retrieved document context:
    - Removes duplicates while preserving order
    - Limits total token count
    - Ensures useful, non-repetitive context
    """

    # Preserve order while removing duplicates
    unique_texts = list(OrderedDict.fromkeys([doc.page_content.strip() for doc in context_list]))

    # Remove very short texts (e.g., headers)
    filtered_texts = [text for text in unique_texts if len(text.split()) > min_length]

    # Avoid near-duplicate entries
    deduplicated_texts = []
    seen_texts = set()
    for text in filtered_texts:
        if not any(text in s for s in seen_texts):  # Avoid near-duplicates
            deduplicated_texts.append(text)
            seen_texts.add(text)

    # Limit context based on token count
    trimmed_context = []
    total_tokens = 0
    for text in deduplicated_texts:
        tokenized_text = tokenizer.encode(text, add_special_tokens=False)
        token_count = len(tokenized_text)

        if total_tokens + token_count > max_tokens:
            remaining_tokens = max_tokens - total_tokens
            if remaining_tokens > 20:
                trimmed_context.append(tokenizer.decode(tokenized_text[:remaining_tokens]))
            break

        trimmed_context.append(text)
        total_tokens += token_count

    return "\n\n".join(trimmed_context) if trimmed_context else "No relevant context found."

# Define prompt
prompt_template = PromptTemplate(
    template="""
    You are a Kubernetes instructor. Answer the question based on the provided context.
    If the context does not provide an answer, say "I don't have enough information."

    Context:
    {context}

    Question:
    {input}

    Answer:
    """,
    input_variables=["context", "input"]
)

llm_chain = LLMChain(llm=llm, prompt=prompt_template)
qa_chain = create_retrieval_chain(retriever, llm_chain)

# Query function
def get_k8s_answer(query):
    retrieved_context = retriever.get_relevant_documents(query)
    cleaned_context = clean_context(retrieved_context, max_tokens=350)  # Ensure context size is within limits

    # Ensure total input tokens < 512 before passing to model
    input_text = f"Context:\n{cleaned_context}\n\nQuestion: {query}\nAnswer:"
    total_tokens = len(tokenizer.encode(input_text, add_special_tokens=True))

    if total_tokens > 512:
      # Trim context further to fit within the limit
      allowed_tokens = 512 - len(tokenizer.encode(query, add_special_tokens=True)) - 50  # 50 tokens for the model's response
      cleaned_context = clean_context(retrieved_context, max_tokens=allowed_tokens)

      # Recalculate total tokens
      input_text = f"Context:\n{cleaned_context}\n\nQuestion: {query}\nAnswer:"
      total_tokens = len(tokenizer.encode(input_text, add_special_tokens=True))

      if total_tokens > 512:
          return "Error: Even after trimming, input is too large."

    response = qa_chain.invoke({"input": query, "context": cleaned_context})
    return response

def get_k8s_answer_text(query):
  model_full_answer = get_k8s_answer(query)
  if 'answer' in model_full_answer.keys():
    if 'text' in model_full_answer['answer'].keys():
      return model_full_answer['answer']['text']
  return "Error"

# Gradio Interface
demo = gr.Interface(
    fn=get_k8s_answer_text,
    inputs=gr.Textbox(label="Ask a Kubernetes Question"),
    outputs=gr.Textbox(label="Answer"),
    title="Kubernetes RAG Assistant",
    description="Ask any Kubernetes-related question and get a step-by-step answer based on documentation."
)

if __name__ == "__main__":
    demo.launch()