keras-hub-nightly 0.21.0.dev202505270408__py3-none-any.whl → 0.21.0.dev202505280410__py3-none-any.whl

keras_hub/src/layers/preprocessing/multi_segment_packer.py

@@ -3,6 +3,7 @@ from keras_hub.src.layers.preprocessing.preprocessing_layer import (
     PreprocessingLayer,
 )
 from keras_hub.src.utils.tensor_utils import convert_to_ragged_batch
+from keras_hub.src.utils.tensor_utils import pad
 from keras_hub.src.utils.tensor_utils import preprocessing_function
 
 try:
@@ -66,6 +67,8 @@ class MultiSegmentPacker(PreprocessingLayer):
                 "waterfall" algorithm that allocates quota in a
                 left-to-right manner and fills up the buckets until we run
                 out of budget. It support arbitrary number of segments.
+        padding_side: str. Whether to pad the input on the "left" or "right".
+            Defaults to "right".
 
     Returns:
         A tuple with two elements. The first is the dense, packed token
@@ -124,6 +127,7 @@ class MultiSegmentPacker(PreprocessingLayer):
         sep_value=None,
         pad_value=None,
         truncate="round_robin",
+        padding_side="right",
         **kwargs,
     ):
         super().__init__(**kwargs)
@@ -162,6 +166,7 @@ class MultiSegmentPacker(PreprocessingLayer):
         self.end_value = end_value
 
         self.pad_value = pad_value
+        self.padding_side = padding_side
 
     def get_config(self):
         config = super().get_config()
@@ -173,6 +178,7 @@ class MultiSegmentPacker(PreprocessingLayer):
                 "sep_value": self._sep_value,
                 "pad_value": self.pad_value,
                 "truncate": self.truncate,
+                "padding_side": self.padding_side,
             }
         )
         return config
@@ -287,10 +293,18 @@ class MultiSegmentPacker(PreprocessingLayer):
         # Pad to dense tensor output.
         sequence_length = sequence_length or self.sequence_length
         shape = tf.cast([-1, sequence_length], "int64")
-        token_ids = token_ids.to_tensor(
-            shape=shape, default_value=self.pad_value
+        token_ids = pad(
+            token_ids,
+            shape=shape,
+            padding_side=self.padding_side,
+            pad_value=self.pad_value,
+        )
+        segment_ids = pad(
+            segment_ids,
+            shape=shape,
+            padding_side=self.padding_side,
+            pad_value=0,
         )
-        segment_ids = segment_ids.to_tensor(shape=shape)
         # Remove the batch dim if added.
         if unbatched:
             token_ids = tf.squeeze(token_ids, 0)

keras_hub/src/layers/preprocessing/start_end_packer.py

@@ -3,6 +3,7 @@ from keras_hub.src.layers.preprocessing.preprocessing_layer import (
     PreprocessingLayer,
 )
 from keras_hub.src.utils.tensor_utils import convert_to_ragged_batch
+from keras_hub.src.utils.tensor_utils import pad
 from keras_hub.src.utils.tensor_utils import preprocessing_function
 
 try:
@@ -39,6 +40,8 @@ class StartEndPacker(PreprocessingLayer):
             0 or "" will be added depending on the dtype of the input tensor.
         return_padding_mask: bool. Whether to return a boolean padding mask of
             all locations that are filled in with the `pad_value`.
+        padding_side: str. Whether to pad the input on the "left" or "right".
+            Defaults to "right".
 
     Call arguments:
         inputs: A `tf.Tensor`, `tf.RaggedTensor`, or list of python strings.
@@ -111,6 +114,7 @@ class StartEndPacker(PreprocessingLayer):
         pad_value=None,
         return_padding_mask=False,
         name=None,
+        padding_side="right",
         **kwargs,
     ):
         super().__init__(name=name, **kwargs)
@@ -139,6 +143,7 @@ class StartEndPacker(PreprocessingLayer):
 
         self.pad_value = pad_value
         self.return_padding_mask = return_padding_mask
+        self.padding_side = padding_side
 
     @preprocessing_function
     def call(
@@ -154,6 +159,13 @@ class StartEndPacker(PreprocessingLayer):
         batch_size = tf.shape(x)[0]
         sequence_length = sequence_length or self.sequence_length
         dtype = inputs.dtype
+        # Truncate.
+        truncation_length = sequence_length
+        if add_start_value and self.start_value is not None:
+            truncation_length -= len(self.start_value)
+        if add_end_value and self.end_value is not None:
+            truncation_length -= len(self.end_value)
+        x = x[..., :truncation_length]
 
         # Concatenate start and end tokens.
         if add_start_value and self.start_value is not None:
@@ -167,23 +179,28 @@ class StartEndPacker(PreprocessingLayer):
             end_token_id_tensor = tf.repeat(
                 end_value[tf.newaxis, :], repeats=batch_size, axis=0
             )
-            # Trim to leave room for end token.
-            x = x[..., : sequence_length - len(self.end_value)]
             x = tf.concat([x, end_token_id_tensor], axis=-1)
 
         # Pad to desired length.
-        outputs = x.to_tensor(
-            default_value=self.pad_value,
+        outputs = pad(
+            x,
+            pad_value=self.pad_value,
+            padding_side=self.padding_side,
             shape=(batch_size, sequence_length),
         )
         outputs = tf.squeeze(outputs, axis=0) if unbatched else outputs
 
         if self.return_padding_mask:
             mask = tf.ones_like(x, dtype="bool")
-            mask = mask.to_tensor(shape=(batch_size, sequence_length))
+
+            mask = pad(
+                mask,
+                pad_value=False,
+                padding_side=self.padding_side,
+                shape=(batch_size, sequence_length),
+            )
             mask = tf.squeeze(mask, axis=0) if unbatched else mask
             return outputs, mask
-
         return outputs
 
     def get_config(self):
@@ -195,6 +212,7 @@ class StartEndPacker(PreprocessingLayer):
                 "end_value": self._end_value,
                 "pad_value": self.pad_value,
                 "return_padding_mask": self.return_padding_mask,
+                "padding_side": self.padding_side,
             }
         )
         return config

keras_hub/src/models/backbone.py

@@ -189,23 +189,26 @@ class Backbone(keras.Model):
         saver = get_preset_saver(preset_dir)
         saver.save_backbone(self, max_shard_size=max_shard_size)
 
-    def get_lora_target_names(self):
-        """Returns list of layer names which are to be LoRA-fied.
-
-        Subclasses can override this method if the names of layers to be
-        LoRa-fied are different.
-        """
+    def default_lora_layer_names(self):
+        """Returns list of layer names which are to be LoRA-fied."""
         return ["query_dense", "value_dense", "query", "value"]
 
-    def enable_lora(self, rank, target_names=None):
+    def enable_lora(self, rank, target_layer_names=None):
         """Enable Lora on the backbone.
 
         Calling this method will freeze all weights on the backbone,
         while enabling Lora on the query & value `EinsumDense` layers
         of the attention layers.
+
+        Args:
+            rank: The rank of the LoRA factorization.
+            target_layer_names: A list of strings, the names of the layers to
+                apply LoRA to. If `None`, this will be populated with the
+                default LoRA layer names as returned by
+                `backbone.default_lora_layer_names()`.
         """
-        if target_names is None:
-            target_names = self.get_lora_target_names()
+        if target_layer_names is None:
+            target_layer_names = self.default_lora_layer_names()
         self.trainable = True
         self._lora_enabled_layers = []
         self._lora_rank = rank
@@ -214,7 +217,7 @@ class Backbone(keras.Model):
         all_layers = self._flatten_layers(include_self=False)
         all_layers = [lyr for lyr in all_layers if lyr.weights]
         for i, layer in enumerate(all_layers):
-            for name in target_names:
+            for name in target_layer_names:
                 if layer.name == name:
                     if hasattr(layer, "enable_lora"):
                         layer.trainable = True

keras_hub/src/models/gemma3/gemma3_backbone.py

@@ -402,8 +402,8 @@ class Gemma3Backbone(Backbone):
         )
         return config
 
-    def get_lora_target_names(self):
-        target_names = super().get_lora_target_names()
+    def default_lora_layer_names(self):
+        target_names = super().default_lora_layer_names()
 
         # Add these for `Gemma3VITAttention`.
         if not self.text_only_model:

keras_hub/src/models/pali_gemma/pali_gemma_backbone.py

@@ -274,8 +274,8 @@ class PaliGemmaBackbone(Backbone):
         # Keep the image_sequence_length as a backbone property for easy access.
         self.image_sequence_length = self.vit_encoder.image_sequence_length
 
-    def get_lora_target_names(self):
-        target_names = super().get_lora_target_names()
+    def default_lora_layer_names(self):
+        target_names = super().default_lora_layer_names()
 
         # Add these for `PaliGemmaVITAttention`.
         target_names += ["query_proj", "value_proj"]

keras_hub/src/utils/tensor_utils.py

@@ -21,6 +21,20 @@ except ImportError:
 NO_CONVERT_COUNTER = threading.local()
 
 
+def pad(x, shape, padding_side, pad_value):
+    if padding_side == "left":
+        x = x[..., ::-1]
+
+    outputs = x.to_tensor(
+        default_value=pad_value,
+        shape=shape,
+    )
+
+    if padding_side == "left":
+        outputs = outputs[..., ::-1]
+    return outputs
+
+
 @contextlib.contextmanager
 def no_convert_scope():
     try:

keras_hub/src/version.py

@@ -1,7 +1,7 @@
 from keras_hub.src.api_export import keras_hub_export
 
 # Unique source of truth for the version number.
-__version__ = "0.21.0.dev202505270408"
+__version__ = "0.21.0.dev202505280410"
 
 
 @keras_hub_export("keras_hub.version")

{keras_hub_nightly-0.21.0.dev202505270408.dist-info → keras_hub_nightly-0.21.0.dev202505280410.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: keras-hub-nightly
-Version: 0.21.0.dev202505270408
+Version: 0.21.0.dev202505280410
 Summary: Pretrained models for Keras.
 Author-email: Keras team <keras-users@googlegroups.com>
 License-Expression: Apache-2.0

{keras_hub_nightly-0.21.0.dev202505270408.dist-info → keras_hub_nightly-0.21.0.dev202505280410.dist-info}/RECORD

@@ -5,7 +5,7 @@ keras_hub/models/__init__.py,sha256=itSzodVUeuX6HQnmsSXY0Wv-5Htbu397410R-SFW_4I,
 keras_hub/samplers/__init__.py,sha256=aFQIkiqbZpi8vjrPp2MVII4QUfE-eQjra5fMeHsoy7k,886
 keras_hub/src/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 keras_hub/src/api_export.py,sha256=9pQZK27JObxWZ96QPLBp1OBsjWigh1iuV6RglPGMRk0,1499
-keras_hub/src/version.py,sha256=wYddPGeUnPWM-nXzfJ5EfT4qBySxC20hFfv4RXAnu-o,222
+keras_hub/src/version.py,sha256=LkNchDcdwovYSb5xEjf09V8MHim3X9wy6VrAyA_7afw,222
 keras_hub/src/layers/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 keras_hub/src/layers/modeling/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 keras_hub/src/layers/modeling/alibi_bias.py,sha256=1XBTHI52L_iJDhN_w5ydu_iMhCuTgQAxEPwcLA6BPuk,4411
@@ -28,11 +28,11 @@ keras_hub/src/layers/preprocessing/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQe
 keras_hub/src/layers/preprocessing/audio_converter.py,sha256=YGh_kQw65a1Z6S5zzSNVP-ChyLYHq3-eOYpOS53xIN8,4156
 keras_hub/src/layers/preprocessing/image_converter.py,sha256=p2CoSV_zfHIVZqLo1hQk2BdOL_RtBlr5wUtgpAmtwwY,15926
 keras_hub/src/layers/preprocessing/masked_lm_mask_generator.py,sha256=itxWq3FHYlR0I7jKarQlSKbSmRLl9ut_UTSP3ZDwP0A,8162
-keras_hub/src/layers/preprocessing/multi_segment_packer.py,sha256=ZNqnUFnc9Af122Q7T6YyUoXgIdU9AgIJfsvR1UrCjFU,12068
+keras_hub/src/layers/preprocessing/multi_segment_packer.py,sha256=APP62tF9Tw4zah7oL5maSYRXMwcR4RwicZMhQq2wRxY,12509
 keras_hub/src/layers/preprocessing/preprocessing_layer.py,sha256=WyX41b9Ev_YJ5uVQVOAqD0PQasMOPDoyDjl_PkzkAkE,687
 keras_hub/src/layers/preprocessing/random_deletion.py,sha256=_EmBt4d8TTPLF3OQhA8HoBmej-BX_BocbjeW6jzi6Wo,9768
 keras_hub/src/layers/preprocessing/random_swap.py,sha256=cV7HqMwu_JHTbhe9UMVAsZdOTLsukyZDteEBYp0idiM,9509
-keras_hub/src/layers/preprocessing/start_end_packer.py,sha256=lY2K937z6JucxNe7VknynhhjrcUfFigU6mqIdv2gS-Y,7973
+keras_hub/src/layers/preprocessing/start_end_packer.py,sha256=F_yCyI6yyxAfunb37C0AzFX3lKjaZg08HMjUXOpjgwc,8642
 keras_hub/src/metrics/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 keras_hub/src/metrics/bleu.py,sha256=pnid5azpAxO6vKEfUtAby3nH29OGbwYKgVGOGeoaA3I,13694
 keras_hub/src/metrics/edit_distance.py,sha256=kjhe8uNjvv8aN49RyrKAbNi7a8_OlB8fMza0J_CfNQg,6353
@@ -43,7 +43,7 @@ keras_hub/src/metrics/rouge_n.py,sha256=JoFtmgjF4Ic263ny6bfD6vMHKreH9le3HnOOxemu
 keras_hub/src/models/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 keras_hub/src/models/audio_to_text.py,sha256=XoOjXtKBX6K1fz-zOXcdVo3FpjuxCMnJZh2LQcYXb_0,2726
 keras_hub/src/models/audio_to_text_preprocessor.py,sha256=GS-WWyJ6aSsPRxi_0bxvxA00h2mT2FEwSdAoQXAUYVI,3249
-keras_hub/src/models/backbone.py,sha256=KS2x3HFWKhEYhroUFT3uZgSkeW_48zPGqUNvxCDDIQQ,11534
+keras_hub/src/models/backbone.py,sha256=utZP09_u5FpMGiq8jl3W98TCW8CysndwLw2VCs3BHz8,11780
 keras_hub/src/models/causal_lm.py,sha256=ReaF-i3SHsCkHh4c28jM72QjMQ8x7yiCwG39FRb-7KE,16786
 keras_hub/src/models/causal_lm_preprocessor.py,sha256=YY7VJZicdmnjDSWi9g4_pEpd5bdJK166GlWcapvokF0,6663
 keras_hub/src/models/feature_pyramid_backbone.py,sha256=clEW-TTQSVJ_5qFNdDF0iABkin1p_xlBUFjJrC7T0IA,2247
@@ -196,7 +196,7 @@ keras_hub/src/models/gemma/gemma_tokenizer.py,sha256=FhcyNL4lo63MqOhTQPFr07-u3Bd
 keras_hub/src/models/gemma/rms_normalization.py,sha256=fku-JEo2sNy-ytX7ySD1sRzdhRAPmYex_z8oFk1NiG8,833
 keras_hub/src/models/gemma3/__init__.py,sha256=oPFadkdK5DRLD6sYx83iTetY5daWuSzmJilLjokHcbU,257
 keras_hub/src/models/gemma3/gemma3_attention.py,sha256=VstFCTVsplcDNSgnyBcSpLgKn-pktJ39D5Ri-Bb7BQA,13628
-keras_hub/src/models/gemma3/gemma3_backbone.py,sha256=xw6gbFZWZuREcN1iyPj-1Hm-3EmRglgFD5fQSzDp3zA,16439
+keras_hub/src/models/gemma3/gemma3_backbone.py,sha256=CaVUQAKrBd1b_7gF7dyTWLjJebzzMd24_3oUipVu5gE,16445
 keras_hub/src/models/gemma3/gemma3_causal_lm.py,sha256=U3C9TWlIz8VefAxQ0wJ6bDz18wqHBie8B26Ub_nFZs4,13843
 keras_hub/src/models/gemma3/gemma3_causal_lm_preprocessor.py,sha256=vjt4N-zr0Eb5kvkOR-WUgskDTNe64L_6tYnhyNb6xaE,29601
 keras_hub/src/models/gemma3/gemma3_decoder_block.py,sha256=6PLlpDxxF67stDv74fw9nNgUHBWmTLx6qGygJwyu5FY,10819
@@ -286,7 +286,7 @@ keras_hub/src/models/opt/opt_causal_lm_preprocessor.py,sha256=xHfslVMOZlAIj2V2jI
 keras_hub/src/models/opt/opt_presets.py,sha256=LrjgI5gbq4Cvfl_pmeCnKn4hS_V_0GYTeJaDc9tbeZM,1745
 keras_hub/src/models/opt/opt_tokenizer.py,sha256=oDHeed4xf07tm14hj_C78BkzMuuRwRP2cRHmqYnObrs,2557
 keras_hub/src/models/pali_gemma/__init__.py,sha256=uODWTlttOOchcTLpiYHCEWMXnDxIz8ZVIeYFQN2bd8o,288
-keras_hub/src/models/pali_gemma/pali_gemma_backbone.py,sha256=_Sa22j4jk_7400h33S22w0S8Dh8Lzzl6A5WeEp55zSk,13637
+keras_hub/src/models/pali_gemma/pali_gemma_backbone.py,sha256=e1KAg4bmK1PrmYW-Ewx3vD7S2DlX9K8LmbRwv30VEkA,13643
 keras_hub/src/models/pali_gemma/pali_gemma_causal_lm.py,sha256=AViEs6YltUqWnIVo7J02JkXcanBgLSdwZwF56TVr8gc,11345
 keras_hub/src/models/pali_gemma/pali_gemma_causal_lm_preprocessor.py,sha256=F57y0fZ0wYYxfGIjfrJc1W9uQpViYFx5bvFjj5CqUbI,4814
 keras_hub/src/models/pali_gemma/pali_gemma_decoder_block.py,sha256=24ABQ1vGlppV-KfWh0YqJjzM_Lu2GIwvyJ4X2XXie_A,5616
@@ -471,7 +471,7 @@ keras_hub/src/utils/keras_utils.py,sha256=2qrh4F-rqceVFSx0-cbsFBfWae5hBXFb_sEtPP
 keras_hub/src/utils/pipeline_model.py,sha256=jgzB6NQPSl0KOu08N-TazfOnXnUJbZjH2EXXhx25Ftg,9084
 keras_hub/src/utils/preset_utils.py,sha256=fx0gNqOTdvW-ZdP0Y3ZaCGE7frYBhwi3lG_GO0swG4w,34602
 keras_hub/src/utils/python_utils.py,sha256=N8nWeO3san4YnGkffRXG3Ix7VEIMTKSN21FX5TuL7G8,202
-keras_hub/src/utils/tensor_utils.py,sha256=vRbvvnFwA6FutJ7InC1w60HDTVNi87CniDGOLQ3hKPA,15855
+keras_hub/src/utils/tensor_utils.py,sha256=WrohV6-hvxtLE6rRRhtN4hy8GkHikV-NrRnVEYUwJQo,16133
 keras_hub/src/utils/coco/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 keras_hub/src/utils/coco/coco_utils.py,sha256=x_QnUUvZ92zoFzMJugiInHORc4NrMdWVBkpp8BAYF6s,2586
 keras_hub/src/utils/imagenet/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
@@ -502,7 +502,7 @@ keras_hub/src/utils/transformers/preset_loader.py,sha256=1nfS5xVsl-JROGXJXltTqV1
 keras_hub/src/utils/transformers/safetensor_utils.py,sha256=CYUHyA4y-B61r7NDnCsFb4t_UmSwZ1k9L-8gzEd6KRg,3339
 keras_hub/tokenizers/__init__.py,sha256=uMjjm0mzUkRb0e4Ac_JK8aJ9cKGUi5UqmzWoWAFJprE,4164
 keras_hub/utils/__init__.py,sha256=jXPqVGBpJr_PpYmqD8aDG-fRMlxH-ulqCR2SZMn288Y,646
-keras_hub_nightly-0.21.0.dev202505270408.dist-info/METADATA,sha256=qziMUgq8DFxeCK5FC1L2PpjieCgcwivAs4cN68I1I8s,7393
-keras_hub_nightly-0.21.0.dev202505270408.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-keras_hub_nightly-0.21.0.dev202505270408.dist-info/top_level.txt,sha256=N4J6piIWBKa38A4uV-CnIopnOEf8mHAbkNXafXm_CuA,10
-keras_hub_nightly-0.21.0.dev202505270408.dist-info/RECORD,,
+keras_hub_nightly-0.21.0.dev202505280410.dist-info/METADATA,sha256=DW6jOe7Tbk32rdB5bnZHYlyBZYuzTYIui1EoKkhPMpY,7393
+keras_hub_nightly-0.21.0.dev202505280410.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+keras_hub_nightly-0.21.0.dev202505280410.dist-info/top_level.txt,sha256=N4J6piIWBKa38A4uV-CnIopnOEf8mHAbkNXafXm_CuA,10
+keras_hub_nightly-0.21.0.dev202505280410.dist-info/RECORD,,