Upload 3 files

Dockerfile

@@ -0,0 +1,13 @@
+FROM python:3.10
+
+RUN apt-get update && apt-get install -y curl zstd lshw
+RUN curl -fsSL https://ollama.com/install.sh | sh
+
+WORKDIR /app
+COPY requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
+
+COPY app.py .
+
+EXPOSE 7860
+CMD ["python", "app.py"]

app.py

@@ -0,0 +1,229 @@
+import os
+import requests
+import subprocess
+import time
+import threading
+import gradio as gr
+
+from langchain_text_splitters import RecursiveCharacterTextSplitter
+from langchain_community.vectorstores import Chroma
+from langchain_huggingface import HuggingFaceEmbeddings
+from langchain_community.document_loaders import PyPDFLoader
+from langchain_ollama import ChatOllama
+from langchain import hub
+from langchain_core.output_parsers import StrOutputParser
+from rerankers import Reranker
+
+# ──────────────────────────────────────────────
+# 1. Arrancar Ollama en background
+# ──────────────────────────────────────────────
+def start_ollama():
+    subprocess.Popen(["ollama", "serve"], stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL)
+    time.sleep(5)  # Esperar a que el servidor esté listo
+    subprocess.run(["ollama", "pull", "gemma3:4b"], check=True)
+
+print("Iniciando Ollama...")
+ollama_thread = threading.Thread(target=start_ollama)
+ollama_thread.start()
+ollama_thread.join()
+print("Ollama listo.")
+
+# ──────────────────────────────────────────────
+# 2. Descargar y procesar el PDF
+# ──────────────────────────────────────────────
+PDF_URL = "https://escueladepacientes.es/images/Pdfs/Guia_Informativa_Diabetes_1.pdf"
+PDF_PATH = "Guia_Informativa_Diabetes_1.pdf"
+
+if not os.path.exists(PDF_PATH):
+    print("Descargando PDF...")
+    response = requests.get(PDF_URL)
+    with open(PDF_PATH, "wb") as f:
+        f.write(response.content)
+
+print("Cargando y procesando documento...")
+loader = PyPDFLoader(PDF_PATH)
+documents = loader.load()
+
+text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=20)
+all_splits = text_splitter.split_documents(documents)
+
+# ──────────────────────────────────────────────
+# 3. Embeddings y base de datos vectorial
+# ──────────────────────────────────────────────
+print("Creando embeddings...")
+model_name = "sentence-transformers/paraphrase-multilingual-mpnet-base-v2"
+model_kwargs = {"device": "cuda" if os.environ.get("CUDA_VISIBLE_DEVICES") else "cpu"}
+
+embeddings = HuggingFaceEmbeddings(model_name=model_name, model_kwargs=model_kwargs)
+vectordb = Chroma.from_documents(
+    documents=all_splits,
+    embedding=embeddings,
+    persist_directory="chroma_db"
+)
+print("Base de datos vectorial lista.")
+
+# ──────────────────────────────────────────────
+# 4. LLM y reranker
+# ──────────────────────────────────────────────
+llm = ChatOllama(model="gemma3:4b", temperature=0, top_k=50, top_p=0.95)
+ranker = Reranker("answerdotai/answerai-colbert-small-v1", model_type="colbert")
+
+# ──────────────────────────────────────────────
+# 5. Funciones RAG
+# ──────────────────────────────────────────────
+def rag_sin_reranking(query: str) -> tuple[str, str]:
+    docs = vectordb.similarity_search_with_score(query)
+    prompt = hub.pull("rlm/rag-prompt")
+    rag_chain = prompt | llm | StrOutputParser()
+
+    context = []
+    for doc, score in docs:
+        if score < 7:
+            context.append(doc.to_json()["kwargs"]["page_content"])
+
+    if context:
+        answer = rag_chain.invoke({"context": "\n\n".join(context), "question": query})
+        sources = "\n\n---\n\n".join(
+            f"📄 Página {doc.to_json()['kwargs']['metadata'].get('page', '?')} "
+            f"(score: {score:.2f})\n{doc.to_json()['kwargs']['page_content'][:300]}..."
+            for doc, score in docs if score < 7
+        )
+        return answer, sources
+    return "No tengo información para responder a esta pregunta.", ""
+
+
+def rag_con_reranking(query: str) -> tuple[str, str]:
+    docs = vectordb.similarity_search_with_score(query)
+    prompt = hub.pull("rlm/rag-prompt")
+    rag_chain = prompt | llm | StrOutputParser()
+
+    context = []
+    for doc, score in docs:
+        if score < 7:
+            context.append(doc.to_json()["kwargs"]["page_content"])
+
+    if context:
+        ranking = ranker.rank(query=query, docs=context)
+        best_context = ranking[0].text
+        answer = rag_chain.invoke({"context": best_context, "question": query})
+        return answer, f"📄 Contexto seleccionado por reranking:\n\n{best_context}"
+    return "No tengo información para responder a esta pregunta.", ""
+
+
+# ──────────────────────────────────────────────
+# 6. Lógica del chat
+# ──────────────────────────────────────────────
+def chat(message: str, history: list, mode: str, temperature: float, top_k: int, top_p: float):
+    # Actualizar parámetros del LLM si han cambiado
+    global llm
+    llm = ChatOllama(model="gemma3:4b", temperature=temperature, top_k=top_k, top_p=top_p)
+
+    if mode == "LLM base (sin RAG)":
+        chain = llm | StrOutputParser()
+        answer = chain.invoke(message)
+        sources = "_Sin recuperación de documentos._"
+
+    elif mode == "RAG sin reranking":
+        answer, sources = rag_sin_reranking(message)
+
+    else:  # RAG con reranking
+        answer, sources = rag_con_reranking(message)
+
+    # Añadir fuentes al final de la respuesta
+    full_response = answer
+    if sources:
+        full_response += f"\n\n---\n**📚 Fuentes utilizadas:**\n{sources}"
+
+    history.append((message, full_response))
+    return history, history, ""  # history, state, limpiar input
+
+
+# ──────────────────────────────────────────────
+# 7. Interfaz Gradio
+# ──────────────────────────────────────────────
+with gr.Blocks(title="RAG - Guía de Diabetes", theme=gr.themes.Soft()) as demo:
+
+    gr.Markdown("""
+    # 🩺 Sistema de Question Answering sobre Diabetes
+    Basado en la [Guía Informativa de Diabetes](https://escueladepacientes.es/mi-enfermedad/diabetes)
+    de la **Escuela de Pacientes**.
+    Puedes elegir entre tres modos de respuesta y ajustar los parámetros de generación.
+    """)
+
+    with gr.Row():
+        with gr.Column(scale=3):
+            chatbot = gr.Chatbot(
+                label="Conversación",
+                height=500,
+                bubble_full_width=False,
+            )
+            with gr.Row():
+                msg_input = gr.Textbox(
+                    placeholder="Escribe tu pregunta aquí...",
+                    label="Pregunta",
+                    scale=4,
+                    autofocus=True,
+                )
+                send_btn = gr.Button("Enviar", variant="primary", scale=1)
+
+            clear_btn = gr.Button("🗑️ Limpiar conversación", variant="secondary")
+
+        with gr.Column(scale=1):
+            gr.Markdown("### ⚙️ Configuración")
+
+            mode = gr.Radio(
+                choices=["LLM base (sin RAG)", "RAG sin reranking", "RAG con reranking"],
+                value="RAG con reranking",
+                label="Modo de respuesta",
+            )
+
+            gr.Markdown("### 🎛️ Parámetros de generación")
+
+            temperature = gr.Slider(
+                minimum=0.0, maximum=2.0, value=0.0, step=0.1,
+                label="Temperature",
+                info="0 = determinista, 2 = muy aleatorio"
+            )
+            top_k = gr.Slider(
+                minimum=1, maximum=100, value=50, step=1,
+                label="Top-k",
+                info="Número de tokens candidatos"
+            )
+            top_p = gr.Slider(
+                minimum=0.1, maximum=1.0, value=0.95, step=0.05,
+                label="Top-p",
+                info="Nucleus sampling threshold"
+            )
+
+            gr.Markdown("### 💡 Preguntas de ejemplo")
+            examples = gr.Examples(
+                examples=[
+                    ["¿Qué es la glucosa?"],
+                    ["¿Qué tratamiento tiene la diabetes tipo 1?"],
+                    ["¿Cuáles son los síntomas de la hipoglucemia?"],
+                    ["¿Qué diferencia hay entre diabetes tipo 1 y tipo 2?"],
+                    ["¿Cuál es la receta de la tarta de queso?"],
+                ],
+                inputs=msg_input,
+            )
+
+    # Estado para mantener el historial
+    state = gr.State([])
+
+    # Eventos
+    send_btn.click(
+        fn=chat,
+        inputs=[msg_input, state, mode, temperature, top_k, top_p],
+        outputs=[chatbot, state, msg_input],
+    )
+    msg_input.submit(
+        fn=chat,
+        inputs=[msg_input, state, mode, temperature, top_k, top_p],
+        outputs=[chatbot, state, msg_input],
+    )
+    clear_btn.click(
+        fn=lambda: ([], [], ""),
+        outputs=[chatbot, state, msg_input],
+    )
+
+demo.launch()

requirements.txt

@@ -0,0 +1,12 @@
+langchain_community
+langchain
+langchain-huggingface
+langchain-text-splitters
+langchain_ollama
+chromadb
+pypdf
+rerankers[transformers]
+requests
+gradio
+sentence-transformers
+huggingface_hub