Delete worker.py

worker.py

@@ -1,133 +0,0 @@
-# worker.py
-import os
-import time
-import uuid
-import requests
-from typing import Dict, Any, Optional
-from queue_manager import job_store, UPLOAD_DIR, RESULTS_DIR
-from models_job import JobStatus, JobResult, CharacterItem, Metrics
-
-HF_TOKEN = os.environ.get("HF_TOKEN")  # opcional
-TGI_BASE_URL = os.environ.get("TGI_BASE_URL")  # ej: https://org-tgi--space.hf.space
-INFERENCE_ENDPOINT_URL = os.environ.get("INFERENCE_ENDPOINT_URL")
-INFERENCE_MODEL_ID = os.environ.get("INFERENCE_MODEL_ID")  # p.ej. "meta-llama/Llama-3.1-8B-Instruct"
-
-def _auth_headers_json() -> Dict[str, str]:
-    headers = {"Content-Type": "application/json"}
-    if HF_TOKEN:
-        headers["Authorization"] = f"Bearer {HF_TOKEN}"
-    return headers
-
-def _call_tgi(prompt: str) -> str:
-    """
-    Ejemplo para TGI /v1/chat/completions (ajusta al formato de tu TGI).
-    """
-    if not TGI_BASE_URL:
-        # si no hay TGI configurado, devuelve texto de demo
-        return f"[DEMO] Respuesta generada para: {prompt[:60]}..."
-    url = f"{TGI_BASE_URL.rstrip('/')}/v1/chat/completions"
-    payload = {
-        "model": "tgi",  # no siempre necesario
-        "messages": [{"role": "user", "content": prompt}],
-        "max_tokens": 256
-    }
-    r = requests.post(url, headers=_auth_headers_json(), json=payload, timeout=120)
-    r.raise_for_status()
-    data = r.json()
-    # Ajusta según la respuesta de tu TGI
-    return data["choices"][0]["message"]["content"]
-
-def _call_inference_api(prompt: str) -> str:
-    """
-    Ejemplo para Inference API serverless.
-    """
-    if not INFERENCE_MODEL_ID:
-        return f"[DEMO] Inference API no configurado; prompt: {prompt[:60]}..."
-    url = f"https://api-inference.huggingface.co/models/{INFERENCE_MODEL_ID}"
-    r = requests.post(url, headers=_auth_headers_json(), json={"inputs": prompt, "parameters": {"max_new_tokens": 128}}, timeout=120)
-    r.raise_for_status()
-    out = r.json()
-    if isinstance(out, list) and out and "generated_text" in out[0]:
-        return out[0]["generated_text"]
-    return str(out)
-
-def _call_inference_endpoint(payload: Dict[str, Any]) -> Dict[str, Any]:
-    """
-    Ejemplo para Inference Endpoint dedicado.
-    """
-    if not INFERENCE_ENDPOINT_URL:
-        return {"text": "[DEMO] Endpoint no configurado"}
-    r = requests.post(INFERENCE_ENDPOINT_URL, headers=_auth_headers_json(), json=payload, timeout=120)
-    r.raise_for_status()
-    return r.json()
-
-def _fake_extract_characters() -> list[CharacterItem]:
-    return [
-        CharacterItem(name="Alice", screen_time_sec=312.5),
-        CharacterItem(name="Bob", screen_time_sec=288.0),
-    ]
-
-def process_job(job: Dict[str, Any]):
-    """
-    job = {
-      "job_id": str,
-      "mode": "book"|"une"|"both",
-      "local_path": "/app/data/uploads/xxx.mp4"  (si es subida),
-      "video_url": "https://..."                (si es por URL)
-    }
-    """
-    job_id = job["job_id"]
-    mode = job.get("mode", "both")
-    src_filename = os.path.basename(job.get("local_path") or job.get("video_url") or f"{job_id}.mp4")
-
-    # Marca a processing
-    st = JobStatus(job_id=job_id, status="processing", progress=5, message="Iniciando procesamiento…")
-    job_store.set_status(job_id, st)
-
-    # (1) Descarga si viene por URL (demo omite; implementa si lo necesitas)
-    local_path = job.get("local_path")
-    if not local_path and job.get("video_url"):
-        # Aquí descargarías el vídeo a local_path
-        # local_path = os.path.join(UPLOAD_DIR, f"{job_id}_{src_filename}")
-        # requests.get(... stream ...) -> write file
-        pass
-
-    # (2) ASR / Diarización / Preparaciones etc. (simulación)
-    time.sleep(1)
-    st.progress = 20; st.message = "Extrayendo transcripción/diálogos…"; job_store.set_status(job_id, st)
-    # Aquí llamarías a tus pipelines reales (Whisper, diarización, etc.)
-
-    # (3) Generación “libro” con LLM (demo)
-    book_text = None; book_mp3_url = None
-    if mode in ("book","both"):
-        prompt = "Genera una audiodescripción tipo libro con diálogos condensados del vídeo."
-        book_text = _call_tgi(prompt) if TGI_BASE_URL else _call_inference_api(prompt)
-        # Si sintetizas audio, guarda mp3 y pon su URL accesible (por simplicidad omitimos)
-        book_mp3_url = None
-        st.progress = 60; st.message = "Generando texto Libro…"; job_store.set_status(job_id, st)
-
-    # (4) Generación UNE (SRT + audio) (demo)
-    une_srt = None; une_mp3_url = None
-    if mode in ("une","both"):
-        # Genera un SRT mínimo de ejemplo
-        une_srt = "1\n00:00:00,000 --> 00:00:03,000\n[Audiodescripción UNE de ejemplo]\n"
-        une_mp3_url = None
-        st.progress = 80; st.message = "Generando SRT UNE…"; job_store.set_status(job_id, st)
-
-    # (5) Personajes, métricas (demo)
-    chars = _fake_extract_characters()
-    metrics = Metrics(wer=0.07, der=0.12, ux=4.3)
-
-    time.sleep(1)
-    st.progress = 100; st.message = "Completado"; st.status = "completed"; job_store.set_status(job_id, st)
-
-    result = JobResult(
-        job_id=job_id,
-        source_filename=src_filename,
-        duration_sec=None,
-        characters=chars,
-        book={"text": book_text, "mp3_url": book_mp3_url} if book_text or book_mp3_url else None,
-        une={"srt": une_srt, "mp3_url": une_mp3_url} if une_srt or une_mp3_url else None,
-        metrics=metrics
-    )
-    job_store.set_result(job_id, result)