app.py

# ===================================================================================
# 1) SETUP & IMPORTS
# ===================================================================================
from __future__ import annotations
import os
import sys
import base64
import struct
import textwrap
import requests
import atexit
from typing import List, Dict, Tuple, Generator

# --- Fast, safe defaults ---
os.environ.setdefault("HF_HUB_ENABLE_HF_TRANSFER", "1")
os.environ.setdefault("TOKENIZERS_PARALLELISM", "false")
os.environ.setdefault("COQUI_TOS_AGREED", "1")
os.environ.setdefault("GRADIO_ANALYTICS_ENABLED", "false")
os.environ.setdefault("TORCHAUDIO_USE_FFMPEG", "0")  # prevent torchaudio/ffmpeg (torio) path

# --- .env early (HF_TOKEN / SECRET_TOKEN) ---
from dotenv import load_dotenv
load_dotenv()

# --- NumPy sanity with torch 2.2.x ---
import numpy as _np
if int(_np.__version__.split(".", 1)[0]) >= 2:
    raise RuntimeError(
        f"Detected numpy=={_np.__version__}. Please pin numpy<2 (e.g., 1.26.4) for this Space."
    )

# --- Transformers sanity for TTS streaming ---
import transformers as _transformers
if _transformers.__version__ != "4.36.2":
    raise RuntimeError(
        f"Detected transformers=={_transformers.__version__}. "
        "Please pin transformers==4.36.2 for compatibility with Coqui TTS streaming."
    )


# --- Panda shim for Gradio on pandas<2.2 (no 'future.no_silent_downcasting') ---
try:
    import pandas as pd
    try:
        with pd.option_context("future.no_silent_downcasting", True):
            pass
    except Exception:
        from contextlib import contextmanager
        _orig_option_context = pd.option_context

        @contextmanager
        def _patched_option_context(*args, **kwargs):
            # filter out unsupported option pairs
            filtered = []
            i = 0
            while i < len(args):
                key = args[i]
                val = args[i + 1] if i + 1 < len(args) else None
                if key == "future.no_silent_downcasting":
                    i += 2
                    continue
                filtered.extend([key, val])
                i += 2
            with _orig_option_context(*filtered, **kwargs):
                yield

        pd.option_context = _patched_option_context  # type: ignore[attr-defined]
except Exception:
    pd = None  # noqa: N816

# --- Hugging Face Spaces & ZeroGPU (import BEFORE torch/diffusers) ---
try:
    import spaces  # Required for ZeroGPU on HF
except Exception:
    class _SpacesShim:
        def GPU(self, *args, **kwargs):
            def _wrap(fn): return fn
            return _wrap
    spaces = _SpacesShim()

import gradio as gr

# --- Core ML & Data Libraries (after spaces import) ---
import torch
import numpy as np
from huggingface_hub import HfApi, hf_hub_download
from llama_cpp import Llama

# --- Audio decode via ffmpeg-python (no torchaudio.load) ---
import ffmpeg

# --- TTS Libraries ---
from TTS.tts.configs.xtts_config import XttsConfig
from TTS.tts.models.xtts import Xtts
from TTS.utils.manage import ModelManager
import TTS.tts.models.xtts as xtts_module  # for monkey-patching load_audio

# --- Text & Audio Processing ---
import nltk
import langid
import emoji
import noisereduce as nr

# ===================================================================================
# 2) GLOBALS & HELPERS
# ===================================================================================

# Ensure NLTK resources (both 'punkt' and new 'punkt_tab' on newer NLTK)
def _ensure_nltk():
    try:
        nltk.data.find("tokenizers/punkt")
    except LookupError:
        nltk.download("punkt", quiet=True)
    try:
        nltk.data.find("tokenizers/punkt_tab/english")
    except LookupError:
        try:
            nltk.download("punkt_tab", quiet=True)
        except Exception:
            # fallback: downloading 'punkt' already satisfies older versions
            pass

_ensure_nltk()

# Cached models & latents
tts_model: Xtts | None = None
llm_model: Llama | None = None
# store as torch.Tensors (CPU at startup)
voice_latents: Dict[str, Tuple[torch.Tensor, torch.Tensor]] = {}

# Config
HF_TOKEN = os.environ.get("HF_TOKEN")
api = HfApi(token=HF_TOKEN) if HF_TOKEN else None
repo_id = "ruslanmv/ai-story-server"
SECRET_TOKEN = os.getenv("SECRET_TOKEN", "secret")
SENTENCE_SPLIT_LENGTH = 250
LLM_STOP_WORDS = ["</s>", "<|user|>", "/s>"]

# IMPORTANT: With ZeroGPU, DO NOT use CUDA at startup even if torch sees it.
USE_STARTUP_CUDA = os.getenv("USE_STARTUP_CUDA", "false").lower() == "true"

# Roles & prompts
default_system_message = (
    "You're a storyteller crafting a short tale for young listeners. Keep sentences short and simple. "
    "Use narrative style only, without lists or complex words. Type numbers as words (e.g., 'ten')."
)
system_message = os.environ.get("SYSTEM_MESSAGE", default_system_message)
ROLES = ["Cloée", "Julian", "Pirate", "Thera"]
ROLE_PROMPTS = {role: system_message for role in ROLES}
ROLE_PROMPTS["Pirate"] = (
    "You are AI Beard, a pirate. Craft your response from his first-person perspective. "
    "Keep answers short, as if in a real conversation. Only provide the words AI Beard would speak."
)

# ---------- small utils ----------
def pcm_to_wav(pcm_data: bytes, sample_rate: int = 24000, channels: int = 1, bit_depth: int = 16) -> bytes:
    if pcm_data.startswith(b"RIFF"):
        return pcm_data
    byte_rate = sample_rate * channels * bit_depth // 8
    block_align = channels * bit_depth // 8
    chunk_size = 36 + len(pcm_data)
    header = struct.pack(
        "<4sI4s4sIHHIIHH4sI",
        b"RIFF", chunk_size, b"WAVE", b"fmt ",
        16, 1, channels, sample_rate, byte_rate, block_align, bit_depth,
        b"data", len(pcm_data)
    )
    return header + pcm_data

def split_sentences(text: str, max_len: int) -> List[str]:
    sentences = nltk.sent_tokenize(text)
    out: List[str] = []
    for s in sentences:
        if len(s) > max_len:
            out.extend(textwrap.wrap(s, max_len, break_long_words=True))
        else:
            out.append(s)
    return out

def format_prompt_zephyr(message: str, history: List[Tuple[str, str | None]], system_message: str) -> str:
    prompt = f"<|system|>\n{system_message}</s>"
    for user_prompt, bot_response in history:
        if bot_response:
            prompt += f"<|user|>\n{user_prompt}</s><|assistant|>\n{bot_response}</s>"
    prompt += f"<|user|>\n{message}</s><|assistant|>"
    return prompt

# ---------- robust audio decode (mono via ffmpeg) ----------
def _decode_audio_ffmpeg_to_mono(path: str, target_sr: int) -> np.ndarray:
    """
    Return float32 waveform in [-1, 1], mono, resampled to target_sr.
    Shape: (samples,)
    """
    try:
        out, _ = (
            ffmpeg
            .input(path)
            .output("pipe:", format="s16le", acodec="pcm_s16le", ac=1, ar=target_sr)
            .run(capture_stdout=True, capture_stderr=True, cmd="ffmpeg")
        )
        pcm = np.frombuffer(out, dtype=np.int16)
        if pcm.size == 0:
            raise RuntimeError("ffmpeg produced empty audio.")
        return (pcm.astype(np.float32) / 32767.0)
    except ffmpeg.Error as e:
        raise RuntimeError(f"ffmpeg decode failed: {e.stderr.decode(errors='ignore') if e.stderr else e}") from e

# ---------- monkey-patch XTTS internal loader to avoid torchaudio.load() ----------
def _patched_load_audio(audiopath: str, load_sr: int):
    """
    Expected by XTTS: return torch.FloatTensor [1, samples] normalized to [-1, 1], resampled to load_sr.
    """
    wav = _decode_audio_ffmpeg_to_mono(audiopath, target_sr=load_sr)
    audio = torch.from_numpy(wav).float().unsqueeze(0)  # [1, N] on CPU
    return audio

xtts_module.load_audio = _patched_load_audio
try:
    import TTS.utils.audio as _tts_audio_mod
    _tts_audio_mod.load_audio = _patched_load_audio
except Exception:
    pass

def _coqui_cache_dir() -> str:
    # Coqui cache default on Linux
    return os.path.join(os.path.expanduser("~"), ".local", "share", "tts")

# ===================================================================================
# 3) PRECACHE & MODEL LOADERS (RUN BEFORE FIRST INFERENCE)
# ===================================================================================

def precache_assets() -> None:
    """Download voice WAVs, XTTS weights, and Zephyr GGUF to local cache before any inference."""
    print("Pre-caching voice files...")
    file_names = ["cloee-1.wav", "julian-bedtime-style-1.wav", "pirate_by_coqui.wav", "thera-1.wav"]
    base_url = "https://raw.githubusercontent.com/ruslanmv/ai-story-server/main/voices/"
    os.makedirs("voices", exist_ok=True)
    for name in file_names:
        dst = os.path.join("voices", name)
        if not os.path.exists(dst):
            try:
                resp = requests.get(base_url + name, timeout=30)
                resp.raise_for_status()
                with open(dst, "wb") as f:
                    f.write(resp.content)
            except Exception as e:
                print(f"Failed to download {name}: {e}")

    print("Pre-caching XTTS v2 model files...")
    ModelManager().download_model("tts_models/multilingual/multi-dataset/xtts_v2")

    print("Pre-caching Zephyr GGUF...")
    try:
        hf_hub_download(
            repo_id="TheBloke/zephyr-7B-beta-GGUF",
            filename="zephyr-7b-beta.Q5_K_M.gguf",
            force_download=False
        )
    except Exception as e:
        print(f"Warning: GGUF pre-cache error: {e}")

def _load_xtts(device: str) -> Xtts:
    """Load XTTS from the local cache. Always CPU at startup for ZeroGPU compatibility."""
    print(f"Loading Coqui XTTS V2 model on {device.upper()}...")
    model_name = "tts_models/multilingual/multi-dataset/xtts_v2"
    ModelManager().download_model(model_name)  # idempotent
    model_dir = os.path.join(_coqui_cache_dir(), model_name.replace("/", "--"))

    cfg = XttsConfig()
    cfg.load_json(os.path.join(model_dir, "config.json"))
    model = Xtts.init_from_config(cfg)
    model.load_checkpoint(
        cfg,
        checkpoint_dir=model_dir,
        eval=True,
        use_deepspeed=False,
    )
    model.to(device)
    print("XTTS model loaded.")
    return model

def _load_llama_cpu_only() -> Llama:
    """Load Llama (Zephyr GGUF) on CPU only (ZeroGPU friendly)."""
    print("Loading LLM (Zephyr GGUF) on CPU...")
    zephyr_model_path = hf_hub_download(
        repo_id="TheBloke/zephyr-7B-beta-GGUF",
        filename="zephyr-7b-beta.Q5_K_M.gguf"
    )
    llm = Llama(
        model_path=zephyr_model_path,
        n_gpu_layers=0,   # never touch CUDA at startup
        n_ctx=4096,
        n_batch=512,
        verbose=False
    )
    print("LLM loaded (CPU).")
    return llm

def init_models_and_latents() -> None:
    """
    Preload TTS and LLM on CPU and compute voice latents on CPU.
    This avoids any CUDA tensors outside the @spaces.GPU window.
    """
    global tts_model, llm_model, voice_latents

    target_device = "cpu"  # FORCE CPU at startup for ZeroGPU compatibility

    if tts_model is None:
        tts_model = _load_xtts(device=target_device)
    else:
        tts_model.to("cpu")

    if llm_model is None:
        llm_model = _load_llama_cpu_only()

    # Pre-compute latents once on CPU (uses our ffmpeg loader)
    if not voice_latents:
        print("Computing voice conditioning latents (CPU)...")
        with torch.no_grad():
            for role, filename in [
                ("Cloée", "cloee-1.wav"),
                ("Julian", "julian-bedtime-style-1.wav"),
                ("Pirate", "pirate_by_coqui.wav"),
                ("Thera", "thera-1.wav"),
            ]:
                path = os.path.join("voices", filename)
                voice_latents[role] = tts_model.get_conditioning_latents(
                    audio_path=path, gpt_cond_len=30, max_ref_length=60
                )
        print("Voice latents ready (CPU).")

# Ensure we close Llama cleanly to avoid __del__ issues at interpreter shutdown
def _close_llm():
    global llm_model
    try:
        if llm_model is not None:
            llm_model.close()
    except Exception:
        pass
atexit.register(_close_llm)

# ===================================================================================
# 4) INFERENCE HELPERS
# ===================================================================================

def generate_text_stream(llm_instance: Llama, prompt: str,
                         history: List[Tuple[str, str | None]],
                         system_message_text: str) -> Generator[str, None, None]:
    formatted = format_prompt_zephyr(prompt, history, system_message_text)
    stream = llm_instance(
        formatted,
        temperature=0.7,
        max_tokens=512,
        top_p=0.95,
        stop=LLM_STOP_WORDS,
        stream=True
    )
    for resp in stream:
        ch = resp["choices"][0]["text"]
        try:
            is_single_emoji = (len(ch) == 1 and emoji.is_emoji(ch))
        except Exception:
            is_single_emoji = False
        if "<|user|>" in ch or is_single_emoji:
            continue
        yield ch

def _latents_to_device(latents: Tuple[torch.Tensor, torch.Tensor], device: torch.device) -> Tuple[torch.Tensor, torch.Tensor]:
    g, s = latents
    if isinstance(g, torch.Tensor):
        g = g.to(device)
    if isinstance(s, torch.Tensor):
        s = s.to(device)
    return g, s

def generate_audio_stream(tts_instance: Xtts, text: str, language: str,
                          latents: Tuple[torch.Tensor, torch.Tensor]) -> Generator[bytes, None, None]:
    gpt_cond_latent, speaker_embedding = latents
    try:
        for chunk in tts_instance.inference_stream(
            text=text,
            language=language,
            gpt_cond_latent=gpt_cond_latent,
            speaker_embedding=speaker_embedding,
            temperature=0.85,
        ):
            if chunk is None:
                continue
            f32 = chunk.detach().cpu().numpy().squeeze().astype(np.float32)
            f32 = np.clip(f32, -1.0, 1.0)
            s16 = (f32 * 32767.0).astype(np.int16)
            yield s16.tobytes()
    except RuntimeError as e:
        print(f"Error during TTS inference: {e}")
        if "device-side assert" in str(e) and api:
            try:
                gr.Warning("Critical GPU error. Attempting to restart the Space...")
                api.restart_space(repo_id=repo_id)
            except Exception:
                pass

# ===================================================================================
# 5) ZERO-GPU ENTRYPOINT (also works on native GPU)
# ===================================================================================

@spaces.GPU(duration=120)  # ZeroGPU allocates a GPU only for this function call
def generate_story_and_speech(secret_token_input: str, input_text: str, chatbot_role: str) -> List[Dict[str, str]]:
    if secret_token_input != SECRET_TOKEN:
        raise gr.Error("Invalid secret token provided.")
    if not input_text:
        return []

    # Ensure models/latents exist (CPU)
    if tts_model is None or llm_model is None or not voice_latents:
        init_models_and_latents()

    # If ZeroGPU granted CUDA for this call, move XTTS to CUDA; keep LLM on CPU.
    try:
        if torch.cuda.is_available():
            tts_model.to("cuda")
            device = torch.device("cuda")
        else:
            tts_model.to("cpu")
            device = torch.device("cpu")
    except Exception:
        tts_model.to("cpu")
        device = torch.device("cpu")

    # Generate story text (LLM on CPU)
    history: List[Tuple[str, str | None]] = [(input_text, None)]
    full_story_text = "".join(
        generate_text_stream(llm_model, history[-1][0], history[:-1], system_message_text=ROLE_PROMPTS[chatbot_role])
    ).strip()
    if not full_story_text:
        return []

    # Split into TTS-friendly sentences
    sentences = split_sentences(full_story_text, SENTENCE_SPLIT_LENGTH)
    lang = langid.classify(sentences[0])[0] if sentences else "en"

    results: List[Dict[str, str]] = []
    for sentence in sentences:
        if not any(c.isalnum() for c in sentence):
            continue

        # Move cached latents to the same device as the model for this call
        lat_dev = _latents_to_device(voice_latents[chatbot_role], device)

        audio_chunks = generate_audio_stream(tts_model, sentence, lang, lat_dev)
        pcm_data = b"".join(chunk for chunk in audio_chunks if chunk)

        # Optional noise reduction (best-effort, CPU)
        try:
            data_s16 = np.frombuffer(pcm_data, dtype=np.int16)
            if data_s16.size > 0:
                float_data = (data_s16.astype(np.float32) / 32767.0)
                reduced = nr.reduce_noise(y=float_data, sr=24000)
                final_pcm = np.clip(reduced * 32767.0, -32768, 32767).astype(np.int16).tobytes()
            else:
                final_pcm = pcm_data
        except Exception:
            final_pcm = pcm_data

        b64_wav = base64.b64encode(pcm_to_wav(final_pcm, sample_rate=24000, channels=1, bit_depth=16)).decode("utf-8")
        results.append({"text": sentence, "audio": b64_wav})

    # Return XTTS to CPU to release GPU instantly
    try:
        tts_model.to("cpu")
    except Exception:
        pass

    return results

# ===================================================================================
# 6) STARTUP: PRECACHE & UI
# ===================================================================================

def build_ui() -> gr.Interface:
    return gr.Interface(
        fn=generate_story_and_speech,
        inputs=[
            gr.Textbox(label="Secret Token", type="password", value=SECRET_TOKEN),
            gr.Textbox(placeholder="What should the story be about?", label="Story Prompt"),
            gr.Dropdown(choices=ROLES, label="Select a Storyteller", value="Cloée"),
        ],
        outputs=gr.JSON(label="Story and Audio Output"),
        title="AI Storyteller with ZeroGPU",
        description="Enter a prompt to generate a short story with voice narration using on-demand GPU.",
        flagging_mode="never",      # avoid deprecated allow_flagging path
        analytics_enabled=False,
    )

if __name__ == "__main__":
    print("===== Startup: pre-cache assets and preload models =====")
    print(f"Python: {sys.version.split()[0]} | Torch CUDA visible: {torch.cuda.is_available()} (will not use at startup)")
    precache_assets()        # 1) download everything to disk
    init_models_and_latents() # 2) load on CPU + compute voice latents on CPU
    print("Models and assets ready. Launching UI...")

    demo = build_ui()
    demo.queue().launch(server_name="0.0.0.0", server_port=int(os.getenv("PORT", "7860")))