diff options
Diffstat (limited to 'bicleaner_ai/models.py')
| -rw-r--r-- | bicleaner_ai/models.py | 14 |
1 files changed, 7 insertions, 7 deletions
diff --git a/bicleaner_ai/models.py b/bicleaner_ai/models.py index cdb403e..ab087c0 100644 --- a/bicleaner_ai/models.py +++ b/bicleaner_ai/models.py @@ -439,8 +439,8 @@ class Transformer(BaseModel): decay_steps=self.settings["steps_per_epoch"]//4, decay_rate=0.2) self.settings["scheduler"] = scheduler - self.settings["optimizer"] = Adam(learning_rate=settings["scheduler"], - clipnorm=settings["clipnorm"]) + self.settings["optimizer"] = Adam(learning_rate=self.settings["scheduler"], + clipnorm=self.settings["clipnorm"]) def get_generator(self, batch_size, shuffle): return ConcatSentenceGenerator( @@ -452,8 +452,10 @@ class Transformer(BaseModel): def build_model(self, compile=True): settings = self.settings inputs = layers.Input(shape=(settings["maxlen"],), dtype='int32') - embedding = TokenAndPositionEmbedding(self.wv, + embedding = TokenAndPositionEmbedding(settings['vocab_size'], + settings['emb_dim'], settings["maxlen"], + self.wv, trainable=True) transformer_block = TransformerBlock( settings["emb_dim"], @@ -631,10 +633,8 @@ class BCXLMRoberta(BaseModel): batch_size=self.settings["batch_size"], callbacks=[earlystop], verbose=verbose) - self.model.save_pretrained(self.dir + '/' - + self.settings["model_file"]) - self.tokenizer.save_pretrained(self.dir + '/' - + self.settings["vocab_file"]) + self.model.save_pretrained(self.dir) + self.tokenizer.save_pretrained(self.dir) y_true = dev_generator.y with redirect_stdout(sys.stderr): |