Fix loading issue of fine-tuned NER models

hankcs · hankcs · commit 11977415095b · 2024-12-01T19:38:50.000-08:00
diff --git a/hanlp/components/taggers/transformers/transformer_tagger.py b/hanlp/components/taggers/transformers/transformer_tagger.py
@@ -151,7 +151,7 @@ def build_model(self, training=True, extra_embeddings: Embedding = None, finetun
             self.config.get('secondary_encoder', None),
             extra_embeddings=extra_embeddings.module(self.vocabs) if extra_embeddings else None,
         )
-        if finetune:
+        if finetune and self.model:
             model_state = model.state_dict()
             load_state = self.model.state_dict()
             safe_state = filter_state_dict_safely(model_state, load_state)
diff --git a/hanlp/version.py b/hanlp/version.py
@@ -2,7 +2,7 @@
 # Author: hankcs
 # Date: 2019-12-28 19:26
 
-__version__ = '2.1.0-beta.62'
+__version__ = '2.1.0-beta.63'
 """HanLP version"""
 
 
diff --git a/setup.py b/setup.py
@@ -64,7 +64,7 @@
         'transformers>=4.1.1',
         'sentencepiece>=0.1.91',  # Essential for tokenization_bert_japanese
         'torch>=1.6.0',
-        'hanlp-common>=0.0.20',
+        'hanlp-common>=0.0.21',
         'hanlp-trie>=0.0.4',
         'hanlp-downloader',
         *TOKENIZERS,

Original file line number	Diff line number	Diff line change
`@@ -151,7 +151,7 @@ def build_model(self, training=True, extra_embeddings: Embedding = None, finetun`
`151`	`151`	`self.config.get('secondary_encoder', None),`
`152`	`152`	`extra_embeddings=extra_embeddings.module(self.vocabs) if extra_embeddings else None,`
`153`	`153`	`)`
`154`		`- if finetune:`
	`154`	`+ if finetune and self.model:`
`155`	`155`	`model_state = model.state_dict()`
`156`	`156`	`load_state = self.model.state_dict()`
`157`	`157`	`safe_state = filter_state_dict_safely(model_state, load_state)`