app.py

import os
import re
import time
import json
import base64
import asyncio
import threading
import datetime
from io import BytesIO
from typing import AsyncGenerator, List, Tuple, Optional

import aiohttp
import gradio as gr
from PIL import Image
import warnings
import requests
import concurrent.futures

# 忽略asyncio警告
warnings.filterwarnings("ignore", category=DeprecationWarning)

# 设置事件循环策略
if hasattr(asyncio, 'WindowsProactorEventLoopPolicy') and isinstance(asyncio.get_event_loop_policy(), asyncio.WindowsProactorEventLoopPolicy):
    asyncio.set_event_loop_policy(asyncio.WindowsSelectorEventLoopPolicy())

# API配置
SAMBANOVA_API_KEY = os.getenv("SAMBANOVA_API_KEY", "").strip()
HF_TOKEN = os.getenv("HF_TOKEN", "").strip()
REVE_API_KEY = os.getenv("REVE_API_KEY", "").strip()

# 目录配置
OUTPUT_DIR = "generaciones_batuto"
REVE_OUTPUT_DIR = "generaciones_reve"
LOG_FILE = "neurocore_logs.jsonl"
os.makedirs(OUTPUT_DIR, exist_ok=True)
os.makedirs(REVE_OUTPUT_DIR, exist_ok=True)

# 模型列表
SAMBA_MODELS = [
    "DeepSeek-R1", "DeepSeek-V3.1", "DeepSeek-V3", "DeepSeek-V3-0324",
    "Meta-Llama-3.3-70B-Instruct", "Llama-4-Maverick-17B-128E-Instruct",
    "Meta-Llama-3.1-8B-Instruct", "Meta-Llama-3.2-11B-Vision-Instruct",
    "Qwen2.5-Coder-32B-Instruct", "Qwen2.5-72B-Instruct", "Qwen3-32B",
    "gpt-oss-120b", "ALLaM-7B-Instruct-preview", "CodeLlama-70b",
    "DeepSeek-Coder-V2", "DeepSeek-R1-0528", "DeepSeek-R1-Distill-Llama-70B",
    "Llama-3.3-Swallow-70B-Instruct-v0.4", "DeepSeek-V3.1-Terminus",
    "DeepSeek-V3.1-cb", "Qwen3-235B", "sambanovasystems/BLOOMChat-176B-v2"
]

HF_MODELS = [
    "mistralai/Codestral-22B-v0.1", "meta-llama/Llama-3.2-11B-Vision-Instruct",
    "JetBrains/Mellum-4b-sft-python", "WizardLM/WizardCoder-Python-34B-V1.0",
    "Qwen/Qwen2-Audio-7B-Instruct", "HuggingFaceTB/SmolLM2-1.7B-Instruct",
    "nvidia/nemotron-speech-streaming-en-0.6b", "openbmb/MiniCPM4.1-8B",
    "naver-hyperclovax/HyperCLOVAX-SEED-Text-Instruct-0.5B",
    "Qwen/Qwen3-Coder-Plus", "Qwen/Qwen3-Omni-30B-A3B-Instruct"
]

# 添加REVE作为特殊模型
ALL_MODELS = ["AUTO-SELECT", "REVE"] + SAMBA_MODELS + HF_MODELS

CSS = """
:root{--primary:#00C896;--secondary:#00FFE0;--bg:#000;--border:rgba(0,200,150,.35);}
body,.gradio-container{background:#000!important; color: #fff !important;}
.panel{border:1px solid var(--border);border-radius:16px;padding:12px}
.dark .gradio-container {background: #000 !important;}
.dark .gr-button-primary {background: linear-gradient(45deg, #00C896, #00FFE0) !important;}
.gr-button-primary {background: linear-gradient(45deg, #00C896, #00FFE0) !important;}
"""

def log_event(data: dict):
    """记录事件日志"""
    data["timestamp"] = datetime.datetime.now().isoformat()
    with open(LOG_FILE, "a", encoding="utf-8") as f:
        f.write(json.dumps(data, ensure_ascii=False) + "\n")

def save_generation(content, model_name, type="text"):
    """保存生成的文本"""
    filename = f"{model_name.replace('/', '_')}_{int(time.time())}.txt"
    path = os.path.join(OUTPUT_DIR, filename)
    with open(path, "w", encoding="utf-8") as f:
        f.write(content)
    return path

def guardar_imagen_local(img, index):
    """保存图像到本地"""
    try:
        timestamp = int(time.time())
        nombre_archivo = f"reve_{timestamp}_{index}.png"
        ruta_completa = os.path.join(REVE_OUTPUT_DIR, nombre_archivo)
        img.save(ruta_completa)
        return ruta_completa
    except Exception as e:
        print(f"⚠️ Error guardando: {e}")
        return None

def llamar_api_reve(prompt, ratio, version, api_key, index):
    """调用REVE API生成单张图片"""
    API_URL = "https://api.reve.com/v1/image/create"
    
    payload = {
        "prompt": prompt,
        "aspect_ratio": ratio,
        "version": version
    }
    headers = {
        "Authorization": f"Bearer {api_key}",
        "Accept": "application/json",
        "Content-Type": "application/json"
    }

    try:
        response = requests.post(API_URL, headers=headers, json=payload, timeout=60)
        if response.status_code == 200:
            data = response.json()
            if "image" in data:
                img_bytes = base64.b64decode(data["image"])
                img = Image.open(BytesIO(img_bytes))
                guardar_imagen_local(img, index)
                return img, data.get('credits_used', 0), None
        return None, 0, f"Error {response.status_code}: {response.text}"
    except Exception as e:
        return None, 0, f"Excepción: {str(e)}"

def generar_imagenes_batch(prompt, api_key, ratio="9:16", version="latest", num_imagenes=1):
    """批量生成REVE图片"""
    if not api_key:
        return [], "❌ ¡Falta la API Key de REVE! Configúrala en las variables de entorno."

    imagenes_nuevas = []
    errores = []
    creditos_totales = 0

    with concurrent.futures.ThreadPoolExecutor(max_workers=num_imagenes) as executor:
        futuros = [
            executor.submit(llamar_api_reve, prompt, ratio, version, api_key, i) 
            for i in range(num_imagenes)
        ]
        
        for futuro in concurrent.futures.as_completed(futuros):
            img, creditos, error = futuro.result()
            if img:
                imagenes_nuevas.append(img)
                creditos_totales += creditos
            if error:
                errores.append(error)

    if imagenes_nuevas:
        return imagenes_nuevas, f"✅ {len(imagenes_nuevas)} imágenes generadas | Créditos usados: {creditos_totales}"
    else:
        return [], f"❌ Error: {'; '.join(errores[:2])}"

def smart_select(prompt: str) -> str:
    """智能选择模型"""
    p = prompt.lower()
    if any(x in p for x in ["código", "python", "script", "programa", "code"]): 
        return "DeepSeek-Coder-V2"
    if any(x in p for x in ["razona", "piensa", "matemáticas", "math", "logic", "resuelve"]): 
        return "DeepSeek-R1"
    if any(x in p for x in ["vision", "mira", "describe", "imagen", "image"]): 
        return "Meta-Llama-3.2-11B-Vision-Instruct"
    if any(x in p for x in ["audio", "sonido", "speech", "voz"]):
        return "Qwen/Qwen2-Audio-7B-Instruct"
    return "DeepSeek-V3.1"

async def stream_samba(model: str, prompt: str, temp: float, tokens: int) -> AsyncGenerator[str, None]:
    """从Sambanova API流式获取响应"""
    url = "https://api.sambanova.ai/v1/chat/completions"
    headers = {"Authorization": f"Bearer {SAMBANOVA_API_KEY}", "Content-Type": "application/json"}
    payload = {
        "model": model,
        "messages": [{"role": "user", "content": prompt}],
        "temperature": temp,
        "max_tokens": tokens,
        "stream": True
    }
    
    full_res = ""
    timeout = aiohttp.ClientTimeout(total=60.0)
    
    try:
        async with aiohttp.ClientSession(timeout=timeout) as session:
            async with session.post(url, headers=headers, json=payload) as resp:
                if resp.status != 200:
                    error_text = await resp.text()
                    yield f"Error {resp.status}: {error_text}"
                    return
                
                async for line in resp.content:
                    if line:
                        line = line.decode("utf-8").strip()
                        if line.startswith("data: "):
                            data_str = line[6:]
                            if data_str == "[DONE]":
                                break
                            try:
                                data = json.loads(data_str)
                                if "choices" in data and data["choices"]:
                                    delta = data["choices"][0]["delta"].get("content", "")
                                    full_res += delta
                                    yield full_res
                            except json.JSONDecodeError:
                                continue
    except Exception as e:
        yield f"Error de conexión: {str(e)}"
    
    if full_res:
        save_generation(full_res, model)
        log_event({"model": model, "prompt": prompt[:100], "response_length": len(full_res)})

def handle_execution(model: str, prompt: str, temp: float, tokens: int, n: int, ratio: str, version: str):
    """处理执行请求"""
    if not prompt.strip():
        return "Por favor ingresa un comando.", []
    
    active_model = smart_select(prompt) if model == "AUTO-SELECT" else model
    
    # 处理REVE图像生成
    if active_model == "REVE":
        if not REVE_API_KEY:
            return "❌ Error: Falta REVE_API_KEY en las variables de entorno.", []
        
        images, message = generar_imagenes_batch(
            prompt, 
            REVE_API_KEY, 
            ratio, 
            version, 
            n
        )
        return message, images
    
    # 处理文本模型
    if active_model in SAMBA_MODELS and not SAMBANOVA_API_KEY:
        return "❌ Error: Falta SAMBANOVA_API_KEY en las variables de entorno.", []
    
    # 返回流式生成器
    return stream_samba(active_model, prompt, temp, tokens)

def create_interface():
    """创建Gradio界面"""
    with gr.Blocks(title="BATUTO X • Neurocore") as demo:
        gr.HTML("""
        <div style="text-align: center; padding: 20px; background: linear-gradient(45deg, #000, #001a14); border-radius: 16px; margin-bottom: 20px;">
            <h1 style="color: #00C896; margin: 0; font-size: 2.5em;">⚡ BATUTO X • NEUROCORE PRO</h1>
            <p style="color: #00FFE0; margin-top: 10px;">Interfaz de Generación Multimodal Avanzada</p>
        </div>
        """)
        
        # 用于存储额外控件的状态
        extra_controls_state = gr.State({"show_image_controls": False})
        
        with gr.Row():
            with gr.Column(scale=1):
                with gr.Group():
                    model_opt = gr.Dropdown(
                        ALL_MODELS, 
                        value="AUTO-SELECT", 
                        label="🧠 Modelo",
                        info="Selecciona un modelo o usa AUTO-SELECT para detección inteligente"
                    )
                    
                    temp_opt = gr.Slider(
                        0, 1.5, 0.7, 
                        label="🌡️ Temperature",
                        info="Controla la aleatoriedad (0 = determinístico, 1.5 = muy creativo)"
                    )
                    
                    tokens_opt = gr.Slider(
                        128, 8192, 2048, 
                        step=128,
                        label="📏 Máximo Tokens",
                        info="Longitud máxima de la respuesta"
                    )
                    
                    # REVE特定的控件（初始隐藏）
                    with gr.Group(visible=False) as image_controls:
                        num_opt = gr.Slider(
                            1, 4, 1, step=1, 
                            label="🖼️ Cantidad de Imágenes",
                            info="Número de imágenes a generar"
                        )
                        
                        ratio_opt = gr.Dropdown(
                            ["16:9", "9:16", "3:2", "2:3", "4:3", "3:4", "1:1"], 
                            value="9:16",
                            label="📐 Aspect Ratio",
                            info="Proporción de la imagen"
                        )
                        
                        version_opt = gr.Dropdown(
                            ["latest", "reve-create@20250915"], 
                            value="latest",
                            label="🔧 Versión",
                            info="Versión del modelo REVE"
                        )
            
            with gr.Column(scale=2):
                with gr.Group():
                    prompt_input = gr.Textbox(
                        lines=5, 
                        label="💬 Entrada",
                        placeholder="Escribe tu comando aquí...\nEjemplo: 'Genera un código Python para ordenar una lista' o 'Crea una imagen de un dragón cibernético'",
                        elem_classes=["prompt-box"]
                    )
        
        send_btn = gr.Button(
            "🚀 EJECUTAR COMANDO", 
            variant="primary", 
            size="lg"
        )
        
        with gr.Group():
            canvas = gr.Textbox(
                lines=12, 
                label="📤 Salida",
                interactive=False
            )
        
        gallery = gr.Gallery(
            label="🎨 Galería de Imágenes",
            columns=2,
            height=400,
            visible=False
        )
        
        # 根据模型选择显示/隐藏控件
        def toggle_controls(model):
            if model == "REVE":
                return [
                    gr.Group(visible=True),  # image_controls
                    gr.Gallery(visible=True),  # gallery
                    gr.Textbox(visible=True),  # canvas
                    {"show_image_controls": True}
                ]
            else:
                return [
                    gr.Group(visible=False),  # image_controls
                    gr.Gallery(visible=False),  # gallery
                    gr.Textbox(visible=True),  # canvas
                    {"show_image_controls": False}
                ]
        
        model_opt.change(
            fn=toggle_controls,
            inputs=model_opt,
            outputs=[image_controls, gallery, canvas, extra_controls_state]
        )
        
        # 连接执行按钮
        send_btn.click(
            fn=handle_execution,
            inputs=[model_opt, prompt_input, temp_opt, tokens_opt, num_opt, ratio_opt, version_opt],
            outputs=[canvas, gallery]
        )
        
        # 示例
        with gr.Accordion("📚 Ejemplos de Uso", open=False):
            gr.Examples(
                examples=[
                    ["Escribe un programa en Python que implemente el algoritmo de ordenamiento quicksort", "AUTO-SELECT"],
                    ["Explica la teoría de la relatividad de Einstein en términos simples", "AUTO-SELECT"],
                    ["Genera una imagen de un dragón cibernético en una ciudad futurista", "REVE"],
                    ["Analiza este código y sugiere mejoras: def factorial(n): return 1 if n==0 else n*factorial(n-1)", "AUTO-SELECT"],
                    ["Resuelve esta ecuación: x² + 5x + 6 = 0", "AUTO-SELECT"]
                ],
                inputs=[prompt_input, model_opt],
                label="Haz clic en un ejemplo para cargarlo"
            )
    
    return demo

def main():
    """主函数"""
    print("🚀 Iniciando BATUTO X Neurocore")
    print(f"📁 Directorio de salida: {os.path.abspath(OUTPUT_DIR)}")
    print(f"🎨 Directorio de imágenes REVE: {os.path.abspath(REVE_OUTPUT_DIR)}")
    print(f"📝 Archivo de logs: {os.path.abspath(LOG_FILE)}")
    
    if SAMBANOVA_API_KEY:
        print("✅ SAMBANOVA_API_KEY configurada")
    else:
        print("⚠️  SAMBANOVA_API_KEY no encontrada - modelos Sambanova no funcionarán")
    
    if REVE_API_KEY:
        print("✅ REVE_API_KEY configurada")
    else:
        print("⚠️  REVE_API_KEY no encontrada - generación de imágenes REVE no disponible")
    
    # 创建界面
    demo = create_interface()
    
    # 启动应用
    demo.launch(
        server_name="0.0.0.0",
        server_port=int(os.getenv("PORT", "7860")),
        share=False,
        debug=False,
        show_error=True,
        css=CSS,
        theme=gr.themes.Default(
            primary_hue="emerald",
            neutral_hue="zinc",
            font=[gr.themes.GoogleFont("Inter"), "Arial", "sans-serif"]
        )
    )

if __name__ == "__main__":
    main()