"""

  Qwen3-Embedding-0.6B local text embedding implementation.

  
  Internal model implementation used by the embedding service.
  """
  
  import threading
  from typing import List, Union
  
  import numpy as np
  from sentence_transformers import SentenceTransformer

  class Qwen3TextModel(object):

      """

      Thread-safe singleton text encoder using Qwen3-Embedding-0.6B (local inference).

      """
  
      _instance = None
      _lock = threading.Lock()
  

      def __new__(cls, model_id: str = "Qwen/Qwen3-Embedding-0.6B"):

          with cls._lock:
              if cls._instance is None:

                  cls._instance = super(Qwen3TextModel, cls).__new__(cls)
                  cls._instance.model = SentenceTransformer(model_id, trust_remote_code=True)

          return cls._instance
  
      def encode(
          self,
          sentences: Union[str, List[str]],
          normalize_embeddings: bool = True,
          device: str = "cuda",
          batch_size: int = 32,
      ) -> np.ndarray:
          if device == "gpu":
              device = "cuda"
  

          # Try requested device, fallback to CPU if CUDA is unavailable/insufficient.

          try:
              if device == "cuda":
                  import torch
  
                  if torch.cuda.is_available():
                      free_memory = (
                          torch.cuda.get_device_properties(0).total_memory
                          - torch.cuda.memory_allocated()
                      )
                      if free_memory < 1024 * 1024 * 1024:  # 1GB
                          device = "cpu"
                  else:
                      device = "cpu"
  
              self.model = self.model.to(device)
              embeddings = self.model.encode(
                  sentences,
                  normalize_embeddings=normalize_embeddings,
                  device=device,
                  show_progress_bar=False,
                  batch_size=batch_size,
              )
              return embeddings

          except Exception:
              if device != "cpu":
                  self.model = self.model.to("cpu")
                  embeddings = self.model.encode(
                      sentences,
                      normalize_embeddings=normalize_embeddings,
                      device="cpu",
                      show_progress_bar=False,
                      batch_size=batch_size,
                  )
                  return embeddings
              raise
  

      def encode_batch(
          self,
          texts: List[str],
          batch_size: int = 32,
          device: str = "cuda",
          normalize_embeddings: bool = True,
      ) -> np.ndarray:
          return self.encode(
              texts,
              batch_size=batch_size,
              device=device,
              normalize_embeddings=normalize_embeddings,
          )