refactor: Introduce accumulate_usage function for token aggregation, enhance alias map expansion in BatchProcessor, and streamline error handling in PreclassificationService. Update PDF processing utilities for improved image conversion and error logging.

2026-03-01 20:55:52 -05:00
parent 3c258ebf12
commit e7fde25227
29 changed files with 321 additions and 1206 deletions
@@ -27,177 +27,90 @@ class InvoiceExtractorStrategy(ABC):
    raise NotImplementedError


+_LOG_EXTRA_BASE = {"dpi": '-', "pdf": '-', "radicado": '-'}
+
+
 class OllamaExtractor(InvoiceExtractorStrategy):
  """Extractor contra un servidor Ollama/OpenAI-compatible.

  Parámetros configurables: `api_key`, `base_url` y `model`.
  """
  def __init__(self, api_key: Optional[str] = None, base_url: Optional[str] = None, model: Optional[str] = None, llm_client: Optional[LLMClient] = None) -> None:
-    # Defaults desde settings si no se proveen
    self.api_key = api_key or settings.api.api_key
    self.base_url = base_url or settings.api.base_url
    self.model = model or settings.api.model
    self.llm = llm_client or LLMClient(api_key=self.api_key, base_url=self.base_url, default_model=self.model, request_timeout_s=settings.api.request_timeout_s)
    self._logger = logging.getLogger(__name__)

+  def _build_messages(self, images_b64: List[str], instruction: str, history: Optional[List[Dict[str, Any]]]) -> List[Dict[str, Any]]:
+    content: List[Dict[str, Any]] = [
+      {"type": "image_url", "image_url": {"url": f"data:image/png;base64,{img}"}}
+      for img in images_b64
+    ]
+    content.append({"type": "text", "text": instruction})
+    messages = list(history) if history else []
+    messages.append({"role": "user", "content": content})
+    return messages
+
+  def _log_call_start(self, images_b64: List[str], instruction: str, history: Optional[List[Dict[str, Any]]]) -> None:
+    prompt_len = len(instruction) if isinstance(instruction, str) else 0
+    self._logger.info(
+      "backend_call_start",
+      extra={**_LOG_EXTRA_BASE, "base_url": self.base_url, "model": self.model,
+             "timeout_s": settings.api.request_timeout_s, "images": len(images_b64),
+             "prompt_len": prompt_len, "history_len": len(history) if history else 0},
+    )
+    self._logger.debug(
+      f"Llamada backend: images={len(images_b64)}, prompt_len={prompt_len}, history_len={len(history) if history else 0}",
+      extra=_LOG_EXTRA_BASE,
+    )
+
+  def _log_call_ok(self, elapsed: float) -> None:
+    self._logger.info(
+      "Extracción completada",
+      extra={**_LOG_EXTRA_BASE, "base_url": self.base_url, "model": self.model,
+             "timeout_s": settings.api.request_timeout_s, "elapsed_ms": int(elapsed * 1000)},
+    )
+    self._logger.debug(f"Backend OK (elapsed_ms={int(elapsed*1000)})", extra=_LOG_EXTRA_BASE)
+
+  def _log_call_error(self, elapsed: float) -> None:
+    self._logger.exception(
+      "Error en llamada al backend",
+      extra={**_LOG_EXTRA_BASE, "base_url": self.base_url, "model": self.model,
+             "timeout_s": settings.api.request_timeout_s, "elapsed_ms": int(elapsed * 1000)},
+    )
+    self._logger.debug(f"Backend error tras {int(elapsed*1000)} ms", extra=_LOG_EXTRA_BASE)
+
  def extract(self, images_b64: List[str], instruction: str, history: Optional[List[Dict[str, Any]]] = None) -> Dict[str, Any]:
    """Construye el prompt multimodal y devuelve el texto crudo del modelo."""
-    content: List[Dict[str, Any]] = []
-    for image_b64 in images_b64:
-      content.append({
-        "type": "image_url",
-        "image_url": {"url": f"data:image/png;base64,{image_b64}"},
-      })
-    content.append({"type": "text", "text": instruction})
-    
-    messages = []
-    if history:
-        messages.extend(history)
-    messages.append({"role": "user", "content": content})
-
+    messages = self._build_messages(images_b64, instruction, history)
+    self._log_call_start(images_b64, instruction, history)
    start = time.time()
    try:
-      # Trazas DEBUG del request (sin exponer imágenes ni texto completo)
-      prompt_len = len(instruction) if isinstance(instruction, str) else 0
-      # Log INFO con metadatos efectivos de la llamada
-      self._logger.info(
-        "backend_call_start",
-        extra={
-          "base_url": self.base_url,
-          "model": self.model,
-          "timeout_s": settings.api.request_timeout_s,
-          "images": len(images_b64),
-          "prompt_len": prompt_len,
-          "history_len": len(history) if history else 0,
-          "dpi": '-', "pdf": '-', "radicado": '-',
-        },
-      )
-      self._logger.debug(
-        f"Llamada backend: images={len(images_b64)}, prompt_len={prompt_len}, history_len={len(history) if history else 0}",
-        extra={"dpi": '-', "pdf": '-', "radicado": '-'},
-      )
      text, usage_dict = self.llm.chat_text(
-        messages=messages,
-        model=self.model,
-        timeout_s=settings.api.request_timeout_s,
-        component="extractor",
+        messages=messages, model=self.model,
+        timeout_s=settings.api.request_timeout_s, component="extractor",
      )
-      elapsed = time.time() - start
-      self._logger.info(
-        "Extracción completada",
-        extra={
-          "dpi": '-', "pdf": '-', "radicado": '-',
-          "base_url": self.base_url, "model": self.model,
-          "timeout_s": settings.api.request_timeout_s,
-          "elapsed_ms": int(elapsed * 1000),
-        },
-      )
-      self._logger.debug(
-        f"Backend OK (elapsed_ms={int(elapsed*1000)})",
-        extra={"dpi": '-', "pdf": '-', "radicado": '-'},
-      )
-      return {
-        "raw": text,
-        "usage": usage_dict,
-      }
+      self._log_call_ok(time.time() - start)
+      return {"raw": text, "usage": usage_dict}
    except Exception:
-      elapsed = time.time() - start
-      self._logger.exception(
-        "Error en llamada al backend",
-        extra={
-          "dpi": '-', "pdf": '-', "radicado": '-',
-          "base_url": self.base_url, "model": self.model,
-          "timeout_s": settings.api.request_timeout_s,
-          "elapsed_ms": int(elapsed * 1000),
-        },
-      )
-      try:
-        self._logger.debug(
-          f"Backend error tras {int(elapsed*1000)} ms",
-          extra={"dpi": '-', "pdf": '-', "radicado": '-'},
-        )
-      except Exception:
-        pass
+      self._log_call_error(time.time() - start)
      raise

  async def extract_async(self, images_b64: List[str], instruction: str, history: Optional[List[Dict[str, Any]]] = None) -> Dict[str, Any]:
-    """Construye el prompt multimodal y devuelve el texto crudo del modelo de forma asíncrona."""
-    content: List[Dict[str, Any]] = []
-    for image_b64 in images_b64:
-      content.append({
-        "type": "image_url",
-        "image_url": {"url": f"data:image/png;base64,{image_b64}"},
-      })
-    content.append({"type": "text", "text": instruction})
-    
-    messages = []
-    if history:
-        messages.extend(history)
-    messages.append({"role": "user", "content": content})
-
+    """Versión asíncrona de extract."""
+    messages = self._build_messages(images_b64, instruction, history)
+    self._log_call_start(images_b64, instruction, history)
    start = time.time()
    try:
-      # Trazas DEBUG del request (sin exponer imágenes ni texto completo)
-      prompt_len = len(instruction) if isinstance(instruction, str) else 0
-      # Log INFO con metadatos efectivos de la llamada
-      self._logger.info(
-        "backend_call_start",
-        extra={
-          "base_url": self.base_url,
-          "model": self.model,
-          "timeout_s": settings.api.request_timeout_s,
-          "images": len(images_b64),
-          "prompt_len": prompt_len,
-          "history_len": len(history) if history else 0,
-          "dpi": '-', "pdf": '-', "radicado": '-',
-        },
-      )
-      self._logger.debug(
-        f"Llamada backend: images={len(images_b64)}, prompt_len={prompt_len}, history_len={len(history) if history else 0}",
-        extra={"dpi": '-', "pdf": '-', "radicado": '-'},
-      )
      text, usage_dict = await self.llm.chat_text_async(
-        messages=messages,
-        model=self.model,
-        timeout_s=settings.api.request_timeout_s,
-        component="extractor",
+        messages=messages, model=self.model,
+        timeout_s=settings.api.request_timeout_s, component="extractor",
      )
-      elapsed = time.time() - start
-      self._logger.info(
-        "Extracción completada",
-        extra={
-          "dpi": '-', "pdf": '-', "radicado": '-',
-          "base_url": self.base_url, "model": self.model,
-          "timeout_s": settings.api.request_timeout_s,
-          "elapsed_ms": int(elapsed * 1000),
-        },
-      )
-      self._logger.debug(
-        f"Backend OK (elapsed_ms={int(elapsed*1000)})",
-        extra={"dpi": '-', "pdf": '-', "radicado": '-'},
-      )
-      return {
-        "raw": text,
-        "usage": usage_dict,
-      }
+      self._log_call_ok(time.time() - start)
+      return {"raw": text, "usage": usage_dict}
    except Exception:
-      elapsed = time.time() - start
-      self._logger.exception(
-        "Error en llamada al backend",
-        extra={
-          "dpi": '-', "pdf": '-', "radicado": '-',
-          "base_url": self.base_url, "model": self.model,
-          "timeout_s": settings.api.request_timeout_s,
-          "elapsed_ms": int(elapsed * 1000),
-        },
-      )
-      try:
-        self._logger.debug(
-          f"Backend error tras {int(elapsed*1000)} ms",
-          extra={"dpi": '-', "pdf": '-', "radicado": '-'},
-        )
-      except Exception:
-        pass
+      self._log_call_error(time.time() - start)
      raise

  def update_config(self, *, api_key: str | None = None, base_url: str | None = None, model: str | None = None) -> None: