Update app.py

app.py

@@ -1,216 +1,3 @@
-# import os
-# import numpy as np
-# import gradio as gr
-# from typing import List, Tuple
-# from pypdf import PdfReader
-# from sentence_transformers import SentenceTransformer
-# from huggingface_hub import InferenceClient
-
-# # -------------------------------------------------
-# # Config
-# # -------------------------------------------------
-# # You can swap to another chat model if needed, e.g.:
-# # "mistralai/Mistral-Nemo-Instruct-2407" or "meta-llama/Llama-3.1-8B-Instruct"
-# GEN_MODEL = os.getenv("GEN_MODEL", "mistralai/Mistral-7B-Instruct-v0.2")
-# HF_TOKEN = os.getenv("HUGGINGFACEHUB_API_TOKEN")  # set in Space Secrets
-# EMB_MODEL_NAME = "sentence-transformers/all-MiniLM-L6-v2"
-# CHUNK_SIZE = 900
-# CHUNK_OVERLAP = 150
-# TOP_K = 4
-
-# # -------------------------------------------------
-# # Try FAISS; fallback to pure NumPy search
-# # -------------------------------------------------
-# USE_FAISS = True
-# try:
-#     import faiss  # type: ignore
-# except Exception:
-#     USE_FAISS = False
-
-# # -------------------------------------------------
-# # Globals
-# # -------------------------------------------------
-# emb = SentenceTransformer(EMB_MODEL_NAME)
-# index = None              # FAISS index (if available)
-# matrix = None             # fallback: stacked embeddings
-# doc_chunks: List[str] = []
-# doc_meta: List[dict] = []
-# client = InferenceClient(model=GEN_MODEL, token=HF_TOKEN)
-
-# SYSTEM_PROMPT = (
-#     "You are a helpful assistant. Use the given CONTEXT to answer the QUESTION.\n"
-#     "If the answer is not in the context, say you don't know.\n"
-#     "Be concise and list source filenames as [source: file.pdf] at the end."
-# )
-
-# # -------------------------------------------------
-# # Helpers
-# # -------------------------------------------------
-# def _extract_text_from_pdf(path: str) -> str:
-#     r = PdfReader(path)
-#     pages = [(p.extract_text() or "") for p in r.pages]
-#     return "\n".join(pages)
-
-# def _chunk_text(text: str, size: int, overlap: int) -> List[str]:
-#     chunks, step = [], size - overlap
-#     i, n = 0, len(text)
-#     while i < n:
-#         chunk = text[i:i+size].strip()
-#         if chunk: chunks.append(chunk)
-#         i += step
-#     return chunks
-
-# def _embed(texts: List[str]) -> np.ndarray:
-#     X = emb.encode(texts, convert_to_numpy=True, normalize_embeddings=True)
-#     return np.asarray(X, dtype=np.float32)
-
-# def _ensure_index(dim: int):
-#     global index, matrix
-#     if USE_FAISS:
-#         index = faiss.IndexFlatIP(dim)
-#     else:
-#         index = None
-#         matrix = None
-
-# def _add_embeddings(E: np.ndarray):
-#     global matrix
-#     if USE_FAISS:
-#         index.add(E)
-#     else:
-#         matrix = E if matrix is None else np.vstack([matrix, E])
-
-# def _search(qv: np.ndarray, k: int):
-#     if USE_FAISS:
-#         return index.search(qv, k)  # returns (D, I)
-#     sims = matrix @ qv[0]  # IP because vectors are normalized
-#     I = np.argsort(-sims)[:k]
-#     D = sims[I]
-#     return D[None, :], I[None, :]
-
-# # -------------------------------------------------
-# # Build index
-# # -------------------------------------------------
-# def build_from_pdfs(files) -> str:
-#     global doc_chunks, doc_meta
-#     doc_chunks, doc_meta = [], []
-
-#     for f in files:
-#         text = _extract_text_from_pdf(f.name)
-#         chunks = _chunk_text(text, CHUNK_SIZE, CHUNK_OVERLAP)
-#         for c in chunks:
-#             doc_chunks.append(c)
-#             doc_meta.append({"file": os.path.basename(f.name)})
-
-#     if not doc_chunks:
-#         return "No text extracted. Check your PDFs."
-
-#     E = _embed(doc_chunks)
-#     _ensure_index(E.shape[1])
-#     _add_embeddings(E)
-#     return f"Indexed {len(doc_chunks)} chunks from {len(files)} file(s)."
-
-# # -------------------------------------------------
-# # Retrieval + Generation
-# # -------------------------------------------------
-# def _retrieve(query: str, k: int = TOP_K) -> Tuple[List[int], List[str]]:
-#     qv = _embed([query])
-#     _, idxs = _search(qv, k)
-#     ids = [i for i in idxs[0].tolist() if i >= 0]
-#     return ids, [doc_chunks[i] for i in ids]
-
-# def _call_chat(messages):
-#     """
-#     Try several Hugging Face client paths for max compatibility.
-#     Returns generated string or raises last exception.
-#     """
-#     # 1) Newer helper
-#     try:
-#         resp = client.chat_completion(messages=messages, max_tokens=512, temperature=0.2, top_p=0.95)
-#         # resp.choices[0].message.content (object or dict)
-#         choice = resp.choices[0]
-#         msg = getattr(choice, "message", None) or choice["message"]
-#         return getattr(msg, "content", None) or msg["content"]
-#     except Exception as e1:
-#         last = e1
-#     # 2) OpenAI-style
-#     try:
-#         resp = client.chat.completions.create(model=GEN_MODEL, messages=messages, max_tokens=512, temperature=0.2, top_p=0.95)
-#         choice = resp.choices[0]
-#         msg = getattr(choice, "message", None) or choice["message"]
-#         return getattr(msg, "content", None) or msg["content"]
-#     except Exception as e2:
-#         last = e2
-#     # 3) Text generation with a single composed prompt
-#     try:
-#         prompt = f"[INST] {SYSTEM_PROMPT}\n\n{messages[-1]['content']} [/INST]"
-#         return client.text_generation(prompt, max_new_tokens=512, temperature=0.2, top_p=0.95,
-#                                       repetition_penalty=1.05, do_sample=True, return_full_text=False).strip()
-#     except Exception as e3:
-#         last = e3
-#     # 4) Old conversational task
-#     try:
-#         conv = client.conversational(
-#             past_user_inputs=[],
-#             generated_responses=[],
-#             text=messages[-1]["content"],
-#             parameters={"temperature": 0.2, "max_new_tokens": 512},
-#         )
-#         return conv["generated_text"] if isinstance(conv, dict) else conv.generated_text
-#     except Exception as e4:
-#         last = e4
-#     raise last
-
-# def answer(question: str) -> str:
-#     if not question.strip():
-#         return "Ask a question."
-#     if (USE_FAISS and index is None) or (not USE_FAISS and matrix is None) or not doc_chunks:
-#         return "Upload PDFs and click **Build Index** first."
-
-#     ids, ctx_chunks = _retrieve(question, TOP_K)
-#     contexts, files = [], []
-#     for rank, i in enumerate(ids, start=1):
-#         chunk = doc_chunks[i][:1000]
-#         fname = doc_meta[i]["file"]
-#         contexts.append(f"[{rank}] {fname}\n{chunk}")
-#         files.append(fname)
-
-#     context_str = "\n\n---\n".join(contexts)
-#     messages = [
-#         {"role": "system", "content": SYSTEM_PROMPT},
-#         {"role": "user", "content": f"QUESTION: {question}\n\nCONTEXT:\n{context_str}"},
-#     ]
-
-#     try:
-#         out = _call_chat(messages)
-#         unique_files = ", ".join(sorted(set(files))) if files else "N/A"
-#         return f"{out.strip()}\n\nSources: {unique_files}"
-#     except Exception as e:
-#         return f"Generation error: {e}\n(Verify your HUGGINGFACEHUB_API_TOKEN and model availability.)"
-
-# # -------------------------------------------------
-# # UI
-# # -------------------------------------------------
-# with gr.Blocks(title="Mistral 7B PDF-RAG") as demo:
-#     gr.Markdown("# 📚 PDF-RAG (Mistral-7B-Instruct)\nUpload PDFs → Build Index → Ask questions. Answers cite sources.")
-
-#     with gr.Row():
-#         with gr.Column(scale=1):
-#             files = gr.File(file_count="multiple", file_types=[".pdf"], label="Upload PDF books")
-#             build_btn = gr.Button("Build Index", variant="primary")
-#             status = gr.Markdown()
-#         with gr.Column(scale=2):
-#             q = gr.Textbox(label="Ask a question", placeholder="What does the book say about ...?")
-#             ask_btn = gr.Button("Ask ➜")
-#             a = gr.Markdown()
-
-#     build_btn.click(build_from_pdfs, inputs=[files], outputs=[status])
-#     ask_btn.click(answer, inputs=[q], outputs=[a])
-#     q.submit(answer, inputs=[q], outputs=[a])
-
-# if __name__ == "__main__":
-#     demo.launch()
-
-
 
 """
 BOOK BUDDY — Ask questions about your PDFs
@@ -260,7 +47,7 @@ client = InferenceClient(model=ROBOT_MODEL, token=HF_TOKEN)
 # A friendly rule for the robot
 ROBOT_RULES = (
     "You are a helpful assistant. Use the given CONTEXT to answer the QUESTION.\n"
-    "If the answer is not in the context, say you don't know.\n"
+    "!!IMPORTANT!! - If the answer is not in the context, Strictly say 'I don't know.', Do not respond any other answers!!!\n"
     "Be short and add source filenames at the end like [source: file.pdf]."
 )
 
@@ -420,10 +207,8 @@ with gr.Blocks(title="📚 Book Buddy — Ask your PDFs") as demo:
             q = gr.Textbox(label="Ask a question", placeholder="Example: Give me 3 key points from this book.")
             examples = gr.Examples(
                 examples=[
-                    ["What is a molecule?"],
                     ["Summarize the main idea in 2 sentences."],
                     ["List 3 important facts from this book."],
-                    ["What is the name of the book?"],
                 ],
                 inputs=q,
             )