app.py

import os
import sys
import torch
import pickle
import time
from fastapi import FastAPI
from fastapi.responses import HTMLResponse, StreamingResponse
from pydantic import BaseModel
from huggingface_hub import snapshot_download
import uvicorn

# ======================
# OPTIMIZACIÓN CPU
# ======================
torch.set_num_threads(max(1, os.cpu_count() // 2))
torch.set_grad_enabled(False)

# ======================
# DISPOSITIVO
# ======================
if torch.cuda.is_available():
    DEVICE = "cuda"
    print("✅ GPU detectada. Usando CUDA.")
else:
    DEVICE = "cpu"
    print("⚠️ GPU no detectada. Usando CPU.")

MODEL_REPO = "teszenofficial/mtp1"

# ======================
# DESCARGA MODELO
# ======================
print("--- SISTEMA MTP 1.1 ---")
repo_path = snapshot_download(
    repo_id=MODEL_REPO,
    repo_type="model",
    local_dir="mtptz_repo"
)

sys.path.insert(0, repo_path)

from model import MTPMiniModel
from tokenizer import MTPTokenizer

# ======================
# CARGA MODELO
# ======================
print("Cargando modelo...")

pkl_file = next(f for f in os.listdir(repo_path) if f.endswith(".pkl"))
with open(os.path.join(repo_path, pkl_file), "rb") as f:
    model_data = pickle.load(f)

tokenizer = MTPTokenizer(os.path.join(repo_path, "mtp_tokenizer.model"))

config = model_data["config"]

model = MTPMiniModel(
    vocab_size=model_data["vocab_size"],
    d_model=config["model"]["d_model"],
    n_layers=config["model"]["n_layers"],
    n_heads=config["model"]["n_heads"],
    d_ff=config["model"]["d_ff"],
    max_seq_len=config["model"]["max_seq_len"],
    dropout=0.0
)

model.load_state_dict(model_data["model_state_dict"])
model.to(DEVICE)
model.eval()

VOCAB_SIZE = tokenizer.sp.get_piece_size()
model.vocab_size = VOCAB_SIZE

print(f"🚀 MTP 1.1 listo en {DEVICE.upper()}")

# ======================
# FASTAPI
# ======================
app = FastAPI(title="MTP 1.1")

class Prompt(BaseModel):
    text: str

# ======================
# GENERACIÓN NORMAL (NO STREAM)
# ======================
@app.post("/generate")
def generate(prompt: Prompt):
    try:
        text = prompt.text.strip()
        if not text:
            return {"reply": ""}

        full_prompt = f"### Instrucción:\n{text}\n\n### Respuesta:\n"
        tokens = [tokenizer.bos_id()] + tokenizer.encode(full_prompt)
        input_ids = torch.tensor([tokens], device=DEVICE)

        with torch.no_grad():
            output = model.generate(
                input_ids,
                max_new_tokens=80,
                temperature=0.7,
                top_k=50,
                top_p=0.9
            )

        gen = output[0, len(tokens):].tolist()
        safe = [t for t in gen if 0 <= t < VOCAB_SIZE and t != tokenizer.eos_id()]
        reply = tokenizer.decode(safe).strip()

        return {"reply": reply}

    except Exception as e:
        print("❌ ERROR:", e)
        return {"reply": "Error interno."}

# ======================
# GENERACIÓN STREAMING (TIPO CHATGPT)
# ======================
@app.post("/generate_stream")
def generate_stream(prompt: Prompt):
    def stream():
        try:
            text = prompt.text.strip()
            full_prompt = f"### Instrucción:\n{text}\n\n### Respuesta:\n"

            tokens = [tokenizer.bos_id()] + tokenizer.encode(full_prompt)
            input_ids = torch.tensor([tokens], device=DEVICE)

            for _ in range(80):
                with torch.no_grad():
                    logits = model(input_ids)[:, -1, :]
                    logits = logits[:, :VOCAB_SIZE]
                    probs = torch.softmax(logits / 0.7, dim=-1)
                    next_id = torch.argmax(probs, dim=-1).item()

                if next_id == tokenizer.eos_id():
                    break

                if 0 <= next_id < VOCAB_SIZE:
                    token_text = tokenizer.decode([next_id])
                    yield token_text
                    input_ids = torch.cat(
                        [input_ids, torch.tensor([[next_id]], device=DEVICE)],
                        dim=1
                    )
                time.sleep(0.015)

        except Exception as e:
            print("❌ STREAM ERROR:", e)
            yield "\n[error]"

    return StreamingResponse(stream(), media_type="text/plain")

# ======================
# FRONTEND HTML COMPLETO
# ======================
@app.get("/", response_class=HTMLResponse)
def ui():
    return """
<!DOCTYPE html>
<html lang="es">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width,initial-scale=1">
<title>MTP 1.1</title>
<style>
body{margin:0;background:#131314;color:#e3e3e3;font-family:Inter,system-ui}
#chat{max-width:900px;margin:auto;padding:20px}
.msg{margin:12px 0;white-space:pre-wrap}
.user{color:#8ab4f8}
.bot{color:#e3e3e3}
input{width:100%;padding:12px;border-radius:10px;border:none;background:#1e1f20;color:white}
button{margin-top:10px;padding:10px;border-radius:10px;border:none;background:#4a9eff;color:black;font-weight:bold}
</style>
</head>
<body>

<div id="chat">
<div class="msg bot">Hola, soy MTP 1.1.</div>
</div>

<input id="inp" placeholder="Escribe algo…" />
<button onclick="send()">Enviar</button>

<script>
async function send(){
 const input=document.getElementById('inp');
 const text=input.value.trim();
 if(!text)return;
 input.value="";
 const chat=document.getElementById('chat');
 chat.innerHTML+=`<div class="msg user">${text}</div>`;
 const bot=document.createElement('div');
 bot.className="msg bot";
 chat.appendChild(bot);

 const res=await fetch('/generate_stream',{
   method:'POST',
   headers:{'Content-Type':'application/json'},
   body:JSON.stringify({text})
 });

 const reader=res.body.getReader();
 const decoder=new TextDecoder();
 while(true){
   const {value,done}=await reader.read();
   if(done)break;
   bot.textContent+=decoder.decode(value);
   window.scrollTo(0,document.body.scrollHeight);
 }
}
</script>

</body>
</html>
"""

# ======================
# ENTRYPOINT
# ======================
if __name__ == "__main__":
    uvicorn.run(app, host="0.0.0.0", port=7860)