fix embedding size for xlm roberta models

flairNLP · Oct 27, 2023 · 032260c · 032260c
1 parent c375fba
commit 032260c
Showing 1 changed file with 2 additions and 2 deletions.
diff --git a/flair/embeddings/transformer.py b/flair/embeddings/transformer.py
@@ -1139,8 +1139,8 @@ def is_supported_t5_model(config: PretrainedConfig) -> bool:
         # If we use a context separator, add a new special token
         self.use_context_separator = use_context_separator
         if use_context_separator:
-            self.tokenizer.add_special_tokens({"additional_special_tokens": [SENTENCE_BOUNDARY_TAG]})
-            transformer_model.resize_token_embeddings(len(self.tokenizer))
+            added = self.tokenizer.add_special_tokens({"additional_special_tokens": [SENTENCE_BOUNDARY_TAG]})
+            transformer_model.resize_token_embeddings(transformer_model.config.vocab_size + added)
 
         super().__init__(**self.to_args())