Fix flakes in masked lm testing by removing any indeterminism (keras-…

…team#1171)
mattdangerw · Jul 20, 2023 · 85d7c53 · 85d7c53
1 parent 32ff68d
commit 85d7c53
Show file tree

Hide file tree

Showing 13 changed files with 34 additions and 14 deletions.
diff --git a/keras_nlp/models/albert/albert_masked_lm_preprocessor_test.py b/keras_nlp/models/albert/albert_masked_lm_preprocessor_test.py
@@ -57,7 +57,7 @@ def setUp(self):
 
         self.preprocessor = AlbertMaskedLMPreprocessor(
             tokenizer=tokenizer,
-            # Simplify out testing by masking every available token.
+            # Simplify our testing by masking every available token.
             mask_selection_rate=1.0,
             mask_token_rate=1.0,
             random_token_rate=0.0,

diff --git a/keras_nlp/models/albert/albert_masked_lm_test.py b/keras_nlp/models/albert/albert_masked_lm_test.py
@@ -60,7 +60,7 @@ def setUp(self):
 
         self.preprocessor = AlbertMaskedLMPreprocessor(
             tokenizer=tokenizer,
-            # Simplify out testing by masking every available token.
+            # Simplify our testing by masking every available token.
             mask_selection_rate=1.0,
             mask_token_rate=1.0,
             random_token_rate=0.0,

diff --git a/keras_nlp/models/bert/bert_masked_lm_preprocessor_test.py b/keras_nlp/models/bert/bert_masked_lm_preprocessor_test.py
@@ -36,7 +36,7 @@ def setUp(self):
 
         self.preprocessor = BertMaskedLMPreprocessor(
             tokenizer=tokenizer,
-            # Simplify out testing by masking every available token.
+            # Simplify our testing by masking every available token.
             mask_selection_rate=1.0,
             mask_token_rate=1.0,
             random_token_rate=0.0,

diff --git a/keras_nlp/models/bert/bert_masked_lm_test.py b/keras_nlp/models/bert/bert_masked_lm_test.py
@@ -35,11 +35,11 @@ def setUp(self):
         self.vocab += ["the", "quick", "brown", "fox", "."]
         self.preprocessor = BertMaskedLMPreprocessor(
             BertTokenizer(vocabulary=self.vocab),
-            # Simplify out testing by masking every available token.
+            # Simplify our testing by masking every available token.
             mask_selection_rate=1.0,
             mask_token_rate=1.0,
             random_token_rate=0.0,
-            mask_selection_length=2,
+            mask_selection_length=5,
             sequence_length=5,
         )
         self.backbone = BertBackbone(

diff --git a/keras_nlp/models/deberta_v3/deberta_v3_masked_lm_preprocessor_test.py b/keras_nlp/models/deberta_v3/deberta_v3_masked_lm_preprocessor_test.py
@@ -53,7 +53,7 @@ def setUp(self):
         self.tokenizer = DebertaV3Tokenizer(proto=self.proto)
         self.preprocessor = DebertaV3MaskedLMPreprocessor(
             tokenizer=self.tokenizer,
-            # Simplify out testing by masking every available token.
+            # Simplify our testing by masking every available token.
             mask_selection_rate=1.0,
             mask_token_rate=1.0,
             random_token_rate=0.0,

diff --git a/keras_nlp/models/deberta_v3/deberta_v3_masked_lm_test.py b/keras_nlp/models/deberta_v3/deberta_v3_masked_lm_test.py
@@ -54,7 +54,11 @@ def setUp(self):
         proto = bytes_io.getvalue()
         self.preprocessor = DebertaV3MaskedLMPreprocessor(
             DebertaV3Tokenizer(proto=proto),
-            mask_selection_length=2,
+            # Simplify our testing by masking every available token.
+            mask_selection_rate=1.0,
+            mask_token_rate=1.0,
+            random_token_rate=0.0,
+            mask_selection_length=5,
             sequence_length=5,
         )
         self.backbone = DebertaV3Backbone(

diff --git a/keras_nlp/models/distil_bert/distil_bert_masked_lm_preprocessor_test.py b/keras_nlp/models/distil_bert/distil_bert_masked_lm_preprocessor_test.py
@@ -38,7 +38,7 @@ def setUp(self):
             tokenizer=DistilBertTokenizer(
                 vocabulary=self.vocab,
             ),
-            # Simplify out testing by masking every available token.
+            # Simplify our testing by masking every available token.
             mask_selection_rate=1.0,
             mask_token_rate=1.0,
             random_token_rate=0.0,

diff --git a/keras_nlp/models/distil_bert/distil_bert_masked_lm_test.py b/keras_nlp/models/distil_bert/distil_bert_masked_lm_test.py
@@ -39,8 +39,12 @@ def setUp(self):
         self.vocab += ["the", "quick", "brown", "fox", "."]
         self.preprocessor = DistilBertMaskedLMPreprocessor(
             DistilBertTokenizer(vocabulary=self.vocab),
+            # Simplify our testing by masking every available token.
+            mask_selection_rate=1.0,
+            mask_token_rate=1.0,
+            random_token_rate=0.0,
+            mask_selection_length=5,
             sequence_length=5,
-            mask_selection_length=2,
         )
         self.backbone = DistilBertBackbone(
             vocabulary_size=self.preprocessor.tokenizer.vocabulary_size(),

diff --git a/keras_nlp/models/f_net/f_net_masked_lm_test.py b/keras_nlp/models/f_net/f_net_masked_lm_test.py
@@ -54,8 +54,12 @@ def setUp(self):
         self.proto = bytes_io.getvalue()
         self.preprocessor = FNetMaskedLMPreprocessor(
             FNetTokenizer(proto=self.proto),
+            # Simplify our testing by masking every available token.
+            mask_selection_rate=1.0,
+            mask_token_rate=1.0,
+            random_token_rate=0.0,
+            mask_selection_length=5,
             sequence_length=5,
-            mask_selection_length=2,
         )
         self.backbone = FNetBackbone(
             vocabulary_size=self.preprocessor.tokenizer.vocabulary_size(),

diff --git a/keras_nlp/models/roberta/roberta_masked_lm_preprocessor_test.py b/keras_nlp/models/roberta/roberta_masked_lm_preprocessor_test.py
@@ -54,7 +54,7 @@ def setUp(self):
                 vocabulary=vocab,
                 merges=merges,
             ),
-            # Simplify out testing by masking every available token.
+            # Simplify our testing by masking every available token.
             mask_selection_rate=1.0,
             mask_token_rate=1.0,
             random_token_rate=0.0,

diff --git a/keras_nlp/models/roberta/roberta_masked_lm_test.py b/keras_nlp/models/roberta/roberta_masked_lm_test.py
@@ -53,8 +53,12 @@ def setUp(self):
         self.merges = merges
         self.preprocessor = RobertaMaskedLMPreprocessor(
             RobertaTokenizer(vocabulary=self.vocab, merges=self.merges),
+            # Simplify our testing by masking every available token.
+            mask_selection_rate=1.0,
+            mask_token_rate=1.0,
+            random_token_rate=0.0,
+            mask_selection_length=5,
             sequence_length=5,
-            mask_selection_length=2,
         )
         self.backbone = RobertaBackbone(
             vocabulary_size=self.preprocessor.tokenizer.vocabulary_size(),

diff --git a/keras_nlp/models/xlm_roberta/xlm_roberta_masked_lm_preprocessor_test.py b/keras_nlp/models/xlm_roberta/xlm_roberta_masked_lm_preprocessor_test.py
@@ -56,7 +56,7 @@ def setUp(self):
         self.tokenizer = XLMRobertaTokenizer(proto=self.proto)
         self.preprocessor = XLMRobertaMaskedLMPreprocessor(
             tokenizer=self.tokenizer,
-            # Simplify out testing by masking every available token.
+            # Simplify our testing by masking every available token.
             mask_selection_rate=1.0,
             mask_token_rate=1.0,
             random_token_rate=0.0,

diff --git a/keras_nlp/models/xlm_roberta/xlm_roberta_masked_lm_test.py b/keras_nlp/models/xlm_roberta/xlm_roberta_masked_lm_test.py
@@ -59,8 +59,12 @@ def setUp(self):
 
         self.preprocessor = XLMRobertaMaskedLMPreprocessor(
             XLMRobertaTokenizer(proto=self.proto),
+            # Simplify our testing by masking every available token.
+            mask_selection_rate=1.0,
+            mask_token_rate=1.0,
+            random_token_rate=0.0,
+            mask_selection_length=5,
             sequence_length=5,
-            mask_selection_length=2,
         )
 
         self.backbone = XLMRobertaBackbone(