Your commit message

app.py

@@ -0,0 +1,111 @@
+# Install required packages (for Colab/CLI)
+# !pip install -U bitsandbytes langchain pypdf peft transformers accelerate datasets langchain-community faiss-cpu gradio
+!pip install gradio langchain transformers peft accelerate bitsandbytes faiss-cpu pypdf langchain-community
+
+import gradio as gr
+import torch
+from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline as hf_pipeline
+from peft import PeftModel
+
+from langchain_community.document_loaders import PyPDFLoader
+from langchain.text_splitter import RecursiveCharacterTextSplitter
+from langchain.embeddings import HuggingFaceEmbeddings
+from langchain.vectorstores import FAISS
+from langchain.chains import RetrievalQA
+from langchain_community.llms import HuggingFacePipeline
+
+import tempfile
+import os
+
+# Load model and tokenizer
+def load_llm():
+    base_model = "HuggingFaceTB/SmolLM2-360M"
+    finetuned_dir = "./smollm2-finetuned-lora1"
+
+    model = AutoModelForCausalLM.from_pretrained(
+        base_model,
+        device_map="cpu",  # Use "auto" if GPU is available
+        torch_dtype=torch.float32
+    )
+    model = PeftModel.from_pretrained(model, finetuned_dir)
+    model.eval()
+
+    tokenizer = AutoTokenizer.from_pretrained(finetuned_dir, use_fast=False)
+    tokenizer.pad_token = tokenizer.eos_token
+
+    hf_pipe = hf_pipeline(
+        task="text-generation",
+        model=model,
+        tokenizer=tokenizer,
+        max_new_tokens=200,
+        do_sample=True,
+        temperature=0.7,
+        top_p=0.9,
+        pad_token_id=tokenizer.eos_token_id
+    )
+
+    return HuggingFacePipeline(pipeline=hf_pipe)
+
+# PDF → Chunks → Vectorstore → RAG
+def process_pdf(pdf_file):
+    with tempfile.NamedTemporaryFile(delete=False, suffix=".pdf") as tmp:
+        tmp.write(pdf_file.read())
+        tmp_path = tmp.name
+
+    loader = PyPDFLoader(tmp_path)
+    documents = loader.load()
+
+    splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=200)
+    chunks = splitter.split_documents(documents)
+
+    embeddings = HuggingFaceEmbeddings(model_name="all-MiniLM-L6-v2")
+    vectordb = FAISS.from_documents(chunks, embeddings)
+
+    retriever = vectordb.as_retriever(search_kwargs={"k": 4})
+    chain = RetrievalQA.from_chain_type(
+        llm=llm,
+        retriever=retriever,
+        return_source_documents=True
+    )
+
+    # Clean up temp file
+    os.unlink(tmp_path)
+
+    return chain
+
+# Gradio states
+llm = load_llm()
+qa_chain = None
+
+def upload_and_prepare(pdf_file):
+    global qa_chain
+    qa_chain = process_pdf(pdf_file)
+    return "✅ PDF processed. You can now chat."
+
+def ask_question(user_query):
+    if qa_chain is None:
+        return "⚠️ Please upload and process a PDF first."
+    result = qa_chain({"query": user_query})
+    return result["result"]
+
+# Gradio UI
+with gr.Blocks() as demo:
+    gr.Markdown("## 📄 Chat with your PDF (SmolLM2 + LangChain + LoRA)")
+
+    with gr.Row():
+        pdf_input = gr.File(label="Upload a PDF", file_types=[".pdf"])
+        process_button = gr.Button("Process PDF")
+
+    status_output = gr.Textbox(label="Status")
+
+    with gr.Row():
+        user_input = gr.Textbox(label="Ask a question")
+        ask_button = gr.Button("Ask")
+    
+    answer_output = gr.Textbox(label="Answer")
+
+    process_button.click(fn=upload_and_prepare, inputs=pdf_input, outputs=status_output)
+    ask_button.click(fn=ask_question, inputs=user_input, outputs=answer_output)
+
+# Launch the app
+demo.launch()