Update app.py

app.py

@@ -1,8 +1,11 @@
+# app.py para Hugging Face Spaces
+# Usa CPU con optimizaciones máximas
+
 from fastapi import FastAPI, HTTPException
 from fastapi.middleware.cors import CORSMiddleware
 from pydantic import BaseModel
 from typing import Optional
-from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig
+from transformers import AutoTokenizer, AutoModelForCausalLM
 from peft import PeftModel
 import torch
 import os
@@ -10,19 +13,16 @@ import os
 # =========================
 # CONFIG
 # =========================
-
 BASE_MODEL = "mistralai/Mistral-7B-Instruct-v0.3"
 LORA_MODEL = "Delta0723/techmind-pro-v9"
+OFFLOAD_DIR = "./offload_folder"
 
-# Crear carpeta para offload si no existe
-os.makedirs("offload", exist_ok=True)
+os.makedirs(OFFLOAD_DIR, exist_ok=True)
 
 # =========================
 # FastAPI Setup
 # =========================
-
-app = FastAPI(title="TechMind Pro API")
-
+app = FastAPI(title="TechMind Pro v9")
 app.add_middleware(
     CORSMiddleware,
     allow_origins=["*"],
@@ -30,52 +30,62 @@ app.add_middleware(
     allow_headers=["*"]
 )
 
+# Variable global para modelo
+model = None
+tokenizer = None
+
 # =========================
-# Load Model
+# Load Model (lazy loading)
 # =========================
-
-print("🚀 Cargando modelo y tokenizer...")
-
-try:
+def load_model():
+    global model, tokenizer
+    
+    if model is not None:
+        return
+    
+    print("🚀 Cargando modelo...")
+    
     tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL, use_fast=False)
     tokenizer.pad_token = tokenizer.eos_token
-
-    quant_config = BitsAndBytesConfig(load_in_4bit=True)
-
+    
+    # Cargar en CPU con int8 (más ligero que 4bit para CPU)
     base_model = AutoModelForCausalLM.from_pretrained(
         BASE_MODEL,
-        device_map="auto",
-        trust_remote_code=True,
-        offload_folder="offload",
-        quantization_config=quant_config
+        device_map={"": "cpu"},
+        torch_dtype=torch.float16,
+        low_cpu_mem_usage=True,
+        offload_folder=OFFLOAD_DIR,
+        offload_state_dict=True
     )
-
-    model = PeftModel.from_pretrained(base_model, LORA_MODEL)
+    
+    # Cargar LoRA
+    model = PeftModel.from_pretrained(
+        base_model,
+        LORA_MODEL,
+        device_map={"": "cpu"},
+        offload_folder=OFFLOAD_DIR
+    )
+    
     model.eval()
-
-    print("✅ Modelo listo para usar")
-
-except Exception as e:
-    print("❌ Error al cargar el modelo:", e)
-    raise e
+    print("✅ Modelo cargado")
 
 # =========================
 # Data Models
 # =========================
-
 class Query(BaseModel):
     question: str
-    max_tokens: Optional[int] = 300
+    max_tokens: Optional[int] = 200
     temperature: Optional[float] = 0.7
 
 # =========================
 # Utilidades
 # =========================
-
-def generate_answer(question: str, max_tokens=300, temperature=0.7) -> str:
+def generate_answer(question: str, max_tokens=200, temperature=0.7) -> str:
+    load_model()  # Carga lazy
+    
     prompt = f"<s>[INST] {question} [/INST]"
-    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
-
+    inputs = tokenizer(prompt, return_tensors="pt", truncation=True, max_length=512)
+    
     with torch.no_grad():
         outputs = model.generate(
             **inputs,
@@ -83,19 +93,24 @@ def generate_answer(question: str, max_tokens=300, temperature=0.7) -> str:
             temperature=temperature,
             top_p=0.95,
             do_sample=True,
-            pad_token_id=tokenizer.eos_token_id
+            pad_token_id=tokenizer.eos_token_id,
+            num_beams=1  # Velocidad
         )
-
+    
     decoded = tokenizer.decode(outputs[0], skip_special_tokens=True)
     return decoded.split("[/INST]")[-1].strip() if "[/INST]" in decoded else decoded
 
 # =========================
 # Endpoints
 # =========================
-
 @app.get("/")
 def root():
-    return {"TechMind": "Mistral-7B Instruct + LoRA v9", "status": "online"}
+    return {
+        "model": "TechMind Pro v9",
+        "base": BASE_MODEL,
+        "lora": LORA_MODEL,
+        "status": "online"
+    }
 
 @app.post("/ask")
 def ask_q(req: Query):
@@ -104,3 +119,63 @@ def ask_q(req: Query):
         return {"response": result}
     except Exception as e:
         raise HTTPException(status_code=500, detail=str(e))
+
+# =========================
+# README.md para el Space
+# =========================
+"""
+---
+title: TechMind Pro v9
+emoji: 🤖
+colorFrom: blue
+colorTo: purple
+sdk: docker
+pinned: false
+---
+
+# TechMind Pro v9
+
+API para el modelo TechMind Pro v9 (Mistral-7B + LoRA fine-tuned)
+
+## Uso
+
+```bash
+curl -X POST "https://YOUR-SPACE.hf.space/ask" \
+  -H "Content-Type: application/json" \
+  -d '{"question": "¿Qué es Python?"}'
+```
+"""
+
+# =========================
+# Dockerfile para el Space
+# =========================
+"""
+FROM python:3.10-slim
+
+WORKDIR /app
+
+RUN apt-get update && apt-get install -y \
+    git \
+    && rm -rf /var/lib/apt/lists/*
+
+COPY requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
+
+COPY . .
+
+CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "7860"]
+"""
+
+# =========================
+# requirements.txt
+# =========================
+"""
+fastapi
+uvicorn[standard]
+transformers>=4.35.0
+peft
+torch
+accelerate
+sentencepiece
+protobuf
+"""