Fix np.load allow_pickle=False to work correctly with .npz format (#1329)

* Initial plan

* Fix np.load allow_pickle=False to work with .npz NpzFile format

- Replace .item().get(key) with [key] dict-style access on NpzFile
- Remove variables instance attribute; use local variable instead
- Add type annotation for variables local var as np.lib.npyio.NpzFile
- Add allow_pickle=False to embeddings.npy load in words_spelling_correction.py

Co-authored-by: wannaphong <8536487+wannaphong@users.noreply.github.com>

---------

Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com>
Co-authored-by: wannaphong <8536487+wannaphong@users.noreply.github.com>

Author: Copilot

pythainlp/spell/words_spelling_correction.py

@@ -84,7 +84,9 @@ def _load_embeddings(self) -> tuple[list[str], NDArray[np.float32]]:
         """Loads embeddings matrix and vocabulary list."""
         import numpy as np
 
-        input_matrix = np.load(os.path.join(self.model_dir, "embeddings.npy"))
+        input_matrix = np.load(
+            os.path.join(self.model_dir, "embeddings.npy"), allow_pickle=False
+        )
         words = []
         vocab_path = os.path.join(self.model_dir, "vocabulary.txt")
         with open(vocab_path, encoding="utf-8") as f:

pythainlp/transliterate/w2p.py

@@ -60,7 +60,6 @@ class Thai_W2P:
     p2idx: dict[str, int]
     idx2p: dict[int, str]
     checkpoint: Optional[str]
-    variables: "NDArray"
     enc_emb: "NDArray"
     enc_w_ih: "NDArray"
     enc_w_hh: "NDArray"
@@ -101,52 +100,33 @@ def _load_variables(self) -> None:
 
         if self.checkpoint is None:
             raise RuntimeError("checkpoint path is not set")
-        self.variables: "NDArray" = np.load(self.checkpoint, allow_pickle=False)
-        # (29, 64). (len(graphemes), emb)
-        self.enc_emb: "NDArray" = self.variables.item().get(
-            "encoder.emb.weight"
+        variables: "np.lib.npyio.NpzFile" = np.load(
+            self.checkpoint, allow_pickle=False
         )
+        # (29, 64). (len(graphemes), emb)
+        self.enc_emb: "NDArray" = variables["encoder.emb.weight"]
         # (3*128, 64)
-        self.enc_w_ih: "NDArray" = self.variables.item().get(
-            "encoder.rnn.weight_ih_l0"
-        )
+        self.enc_w_ih: "NDArray" = variables["encoder.rnn.weight_ih_l0"]
         # (3*128, 128)
-        self.enc_w_hh: "NDArray" = self.variables.item().get(
-            "encoder.rnn.weight_hh_l0"
-        )
+        self.enc_w_hh: "NDArray" = variables["encoder.rnn.weight_hh_l0"]
         # (3*128,)
-        self.enc_b_ih: "NDArray" = self.variables.item().get(
-            "encoder.rnn.bias_ih_l0"
-        )
+        self.enc_b_ih: "NDArray" = variables["encoder.rnn.bias_ih_l0"]
         # (3*128,)
-        self.enc_b_hh: "NDArray" = self.variables.item().get(
-            "encoder.rnn.bias_hh_l0"
-        )
-
+        self.enc_b_hh: "NDArray" = variables["encoder.rnn.bias_hh_l0"]
         # (74, 64). (len(phonemes), emb)
-        self.dec_emb: "NDArray" = self.variables.item().get(
-            "decoder.emb.weight"
-        )
+        self.dec_emb: "NDArray" = variables["decoder.emb.weight"]
         # (3*128, 64)
-        self.dec_w_ih: "NDArray" = self.variables.item().get(
-            "decoder.rnn.weight_ih_l0"
-        )
+        self.dec_w_ih: "NDArray" = variables["decoder.rnn.weight_ih_l0"]
         # (3*128, 128)
-        self.dec_w_hh: "NDArray" = self.variables.item().get(
-            "decoder.rnn.weight_hh_l0"
-        )
+        self.dec_w_hh: "NDArray" = variables["decoder.rnn.weight_hh_l0"]
         # (3*128,)
-        self.dec_b_ih: "NDArray" = self.variables.item().get(
-            "decoder.rnn.bias_ih_l0"
-        )
+        self.dec_b_ih: "NDArray" = variables["decoder.rnn.bias_ih_l0"]
         # (3*128,)
-        self.dec_b_hh: "NDArray" = self.variables.item().get(
-            "decoder.rnn.bias_hh_l0"
-        )
+        self.dec_b_hh: "NDArray" = variables["decoder.rnn.bias_hh_l0"]
         # (74, 128)
-        self.fc_w: "NDArray" = self.variables.item().get("decoder.fc.weight")
+        self.fc_w: "NDArray" = variables["decoder.fc.weight"]
         # (74,)
-        self.fc_b: "NDArray" = self.variables.item().get("decoder.fc.bias")
+        self.fc_b: "NDArray" = variables["decoder.fc.bias"]
 
     def _sigmoid(self, x: "np.ndarray") -> "np.ndarray":
         import numpy as np