Merge branch 'master' into filter_relations

flair/embeddings/document.py

@@ -210,7 +210,7 @@ def embed(self, sentences: Union[list[Sentence], Sentence]):
             sentences = [sentences]
 
         raw_sentences = [s.to_original_text() for s in sentences]
-        tfidf_vectors = torch.from_numpy(self.vectorizer.transform(raw_sentences).A)
+        tfidf_vectors = torch.from_numpy(self.vectorizer.transform(raw_sentences).toarray())
 
         for sentence_id, sentence in enumerate(sentences):
             sentence.set_embedding(self.name, tfidf_vectors[sentence_id])
@@ -691,10 +691,9 @@ def _add_embeddings_internal(self, sentences: list[Sentence]):
 
         lengths: list[int] = [len(sentence.tokens) for sentence in sentences]
         padding_length: int = max(max(lengths), self.min_sequence_length)
-
         pre_allocated_zero_tensor = torch.zeros(
             self.embeddings.embedding_length * padding_length,
-            dtype=self.convs[0].weight.dtype,
+            dtype=cast(torch.nn.Conv1d, self.convs[0]).weight.dtype,
             device=flair.device,
         )
 

flair/embeddings/token.py

@@ -1466,7 +1466,7 @@ def _add_embeddings_internal(self, sentences: list[Sentence]) -> list[Sentence]:
             word = token.text if self.field is None else token.get_label(self.field).value
 
             if word.strip() == "":
-                ids = [self.spm.vocab_size(), self.embedder.spm.vocab_size()]
+                ids = [self.spm.vocab_size(), self.spm.vocab_size()]
             else:
                 if self.do_preproc:
                     word = self._preprocess(word)

flair/file_utils.py

@@ -382,4 +382,4 @@ def load_torch_state(model_file: str) -> dict[str, typing.Any]:
         # to load models on some Mac/Windows setups
         # see https://github.com/zalandoresearch/flair/issues/351
         f = load_big_file(model_file)
-        return torch.load(f, map_location="cpu")
+        return torch.load(f, map_location="cpu", weights_only=False)

flair/models/pairwise_regression_model.py

@@ -345,7 +345,7 @@ def evaluate(
                 f"spearman: {metric.spearmanr():.4f}"
             )
 
-            scores = {
+            eval_metrics = {
                 "loss": eval_loss.item(),
                 "mse": metric.mean_squared_error(),
                 "mae": metric.mean_absolute_error(),
@@ -354,12 +354,12 @@ def evaluate(
             }
 
             if main_evaluation_metric[0] in ("correlation", "other"):
-                main_score = scores[main_evaluation_metric[1]]
+                main_score = eval_metrics[main_evaluation_metric[1]]
             else:
-                main_score = scores["spearman"]
+                main_score = eval_metrics["spearman"]
 
             return Result(
                 main_score=main_score,
                 detailed_results=detailed_result,
-                scores=scores,
+                scores=eval_metrics,
             )

flair/models/sequence_tagger_model.py

@@ -862,8 +862,8 @@ def push_to_hub(
             self.save(local_model_path)
 
             # Determine if model card already exists
-            info = model_info(repo_id, use_auth_token=token)
-            write_readme = all(f.rfilename != "README.md" for f in info.siblings)
+            info = model_info(repo_id, token=token)
+            write_readme = info.siblings is None or all(f.rfilename != "README.md" for f in info.siblings)
 
             # Generate and save model card
             if write_readme:

flair/models/text_regression_model.py

@@ -137,7 +137,7 @@ def evaluate(
         out_path: Optional[Union[str, Path]] = None,
         embedding_storage_mode: EmbeddingStorageMode = "none",
         mini_batch_size: int = 32,
-        main_evaluation_metric: tuple[str, str] = ("micro avg", "f1-score"),
+        main_evaluation_metric: tuple[str, str] = ("correlation", "pearson"),
         exclude_labels: Optional[list[str]] = None,
         gold_label_dictionary: Optional[Dictionary] = None,
         return_loss: bool = True,
@@ -195,16 +195,23 @@ def evaluate(
                 f"spearman: {metric.spearmanr():.4f}"
             )
 
-            result: Result = Result(
-                main_score=metric.pearsonr(),
+            eval_metrics = {
+                "loss": eval_loss.item(),
+                "mse": metric.mean_squared_error(),
+                "mae": metric.mean_absolute_error(),
+                "pearson": metric.pearsonr(),
+                "spearman": metric.spearmanr(),
+            }
+
+            if main_evaluation_metric[0] in ("correlation", "other"):
+                main_score = eval_metrics[main_evaluation_metric[1]]
+            else:
+                main_score = eval_metrics["spearman"]
+
+            result = Result(
+                main_score=main_score,
                 detailed_results=detailed_result,
-                scores={
-                    "loss": eval_loss.item(),
-                    "mse": metric.mean_squared_error(),
-                    "mae": metric.mean_absolute_error(),
-                    "pearson": metric.pearsonr(),
-                    "spearman": metric.spearmanr(),
-                },
+                scores=eval_metrics,
             )
 
             return result

flair/nn/model.py

@@ -899,7 +899,9 @@ def predict(
 
                         if has_unknown_label:
                             has_any_unknown_label = True
-                            scores = torch.index_select(scores, 0, torch.tensor(filtered_indices, device=flair.device))
+                            scores = torch.index_select(
+                                scores, 0, torch.tensor(filtered_indices, device=flair.device, dtype=torch.int32)
+                            )
 
                         gold_labels = self._prepare_label_tensor([data_points[index] for index in filtered_indices])
                         overall_loss += self._calculate_loss(scores, gold_labels)[0]

requirements.txt

@@ -17,7 +17,7 @@ scikit-learn>=1.0.2
 segtok>=1.5.11
 sqlitedict>=2.0.0
 tabulate>=0.8.10
-torch>=1.5.0,!=1.8
+torch>=1.13.1
 tqdm>=4.63.0
 transformer-smaller-training-vocab>=0.2.3
 transformers[sentencepiece]>=4.25.0,<5.0.0