keras-hub-nightly 0.16.1.dev202410150342__py3-none-any.whl → 0.16.1.dev202410170342__py3-none-any.whl

keras_hub/src/models/causal_lm.py

@@ -274,6 +274,7 @@ class CausalLM(Task):
         inputs,
         max_length=None,
         stop_token_ids="auto",
+        strip_prompt=False,
     ):
         """Generate text given prompt `inputs`.
 
@@ -309,6 +310,9 @@ class CausalLM(Task):
                 specify a list of token id's the model should stop on. Note that
                 sequences of tokens will each be interpreted as a stop token,
                 multi-token stop sequences are not supported.
+            strip_prompt: Optional. By default, generate() returns the full prompt
+                followed by its completion generated by the model. If this option
+                is set to True, only the newly generated text is returned.
         """
         # Setup our three main passes.
         # 1. Optionally preprocessing strings to dense integer tensors.
@@ -339,6 +343,33 @@ class CausalLM(Task):
         def generate(x):
             return generate_function(x, stop_token_ids=stop_token_ids)
 
+        def strip_prompt_function(x, prompt):
+            # This function removes the prompt from the generated
+            # response, in a batch-friendly fashion.
+            y = {}
+            prompt_mask = prompt["padding_mask"]
+            seq_len = prompt_mask.shape[1]
+
+            # We need to shift every output sequence by the size of the prompt.
+            shifts = -ops.sum(ops.cast(prompt_mask, "int"), axis=1) % seq_len
+            ix = ops.arange(seq_len, dtype="int")
+            ix = ops.expand_dims(ix, axis=0) - ops.expand_dims(shifts, axis=1)
+
+            # This produces the desired shift (in fact a rollover).
+            def roll_sequence(seq):
+                return ops.take_along_axis(seq, ix, axis=1)
+
+            # The shifting rolls the content over so the prompt is at the end of
+            # the sequence and the generated text is at the beginning. We mask
+            # it to retain the generated text only.
+            y["padding_mask"] = ops.logical_xor(
+                roll_sequence(prompt_mask), roll_sequence(x["padding_mask"])
+            )
+            # we assume the mask is enough and there is no need to zero-out the values
+            y["token_ids"] = roll_sequence(x["token_ids"])
+
+            return y
+
         def postprocess(x):
             return self.preprocessor.generate_postprocess(x)
 
@@ -347,7 +378,12 @@ class CausalLM(Task):
 
         if self.preprocessor is not None:
             inputs = [preprocess(x) for x in inputs]
-        outputs = [generate(x) for x in inputs]
+
+        if strip_prompt:
+            outputs = [strip_prompt_function(generate(x), x) for x in inputs]
+        else:
+            outputs = [generate(x) for x in inputs]
+
         if self.preprocessor is not None:
             outputs = [postprocess(x) for x in outputs]
 

keras_hub/src/models/deeplab_v3/deeplab_v3_presets.py

@@ -13,6 +13,6 @@ backbone_presets = {
             "path": "deeplabv3",
             "model_card": "https://arxiv.org/abs/1802.02611",
         },
-        "kaggle_handle": "kaggle://keras/deeplabv3/keras/deeplabv3_plus_resnet50_pascalvoc/3",
+        "kaggle_handle": "kaggle://keras/deeplabv3/keras/deeplab_v3_plus_resnet50_pascalvoc/3",
     },
 }

keras_hub/src/models/gemma/gemma_backbone.py

@@ -224,7 +224,7 @@ class GemmaBackbone(Backbone):
 
         Example:
         ```
-        # Feel free to change the mesh shape to balance data and model parallel
+        # Feel free to change the mesh shape to balance data and model parallelism
         mesh = keras.distribution.DeviceMesh(
             shape=(1, 8), axis_names=('batch', 'model'),
             devices=keras.distribution.list_devices())
@@ -232,11 +232,19 @@ class GemmaBackbone(Backbone):
             mesh, model_parallel_dim_name="model")
 
         distribution = keras.distribution.ModelParallel(
-            mesh, layout_map, batch_dim_name='batch')
+            layout_map=layout_map, batch_dim_name='batch')
         with distribution.scope():
            gemma_model = keras_hub.models.GemmaCausalLM.from_preset()
         ```
 
+        To see how the layout map was applied, load the model then run (for one decoder block):
+        ```
+        embedding_layer = gemma_model.backbone.get_layer("token_embedding")
+        decoder_block_1 = gemma_model.backbone.get_layer('decoder_block_1')
+        for variable in embedding_layer.weights + decoder_block_1.weights:
+            print(f'{variable.path:<58}  {str(variable.shape):<16}  {str(variable.value.sharding.spec)}')
+        ```
+
         Args:
             device_mesh: The `keras.distribution.DeviceMesh` instance for
                 distribution.
@@ -246,7 +254,7 @@ class GemmaBackbone(Backbone):
                 the data should be partition on.
         Return:
             `keras.distribution.LayoutMap` that contains the sharding spec
-            of all the model weights.
+            for all the model weights.
         """
         # The weight path and shape of the Gemma backbone is like below (for 2G)
         # token_embedding/embeddings,  (256128, 2048), 524550144

keras_hub/src/models/llama/llama_backbone.py

@@ -175,3 +175,121 @@ class LlamaBackbone(Backbone):
             }
         )
         return config
+
+    @staticmethod
+    def get_layout_map(
+        device_mesh,
+        model_parallel_dim_name="model",
+        data_parallel_dim_name="batch",
+    ):
+        """Get a `keras.distribution.LayoutMap` for model parallel distribution.
+
+        The returned `LayoutMap` contains the sharding spec for the Llama
+        backbone weights, so that you can use it to distribute weights across
+        the accelerators.
+
+        Example:
+        ```
+        # Feel free to change the mesh shape to balance data and model parallelism
+        mesh = keras.distribution.DeviceMesh(
+            shape=(1, 8),
+            axis_names=('batch', 'model'),
+            devices=keras.distribution.list_devices(),
+        )
+        layout_map = LlamaBackbone.get_layout_map(
+            mesh,
+            model_parallel_dim_name="model",
+        )
+
+        distribution = keras.distribution.ModelParallel(
+            layout_map=layout_map,
+            batch_dim_name='batch',
+        )
+
+        with distribution.scope():
+           llama_model = keras_hub.models.LlamaCausalLM.from_preset()
+        ```
+
+        To see how the layout map was applied, load the model then run (for one decoder block):
+        ```
+        embedding_layer = llama_model.backbone.get_layer("token_embedding")
+        decoder_block_1 = llama_model.backbone.get_layer('transformer_layer_0')
+        for variable in embedding_layer.weights + decoder_block_1.weights:
+            print(f'{variable.path:<58}  {str(variable.shape):<16}  {str(variable.value.sharding.spec)}')
+        ```
+
+        Args:
+            device_mesh: The `keras.distribution.DeviceMesh` instance for
+                distribution.
+            model_parallel_dim_name: The axis name of the device mesh, where
+                the weights should be partition on.
+            data_parallel_dim_name: The axis name of the device mesh, where
+                the data should be partition on.
+        Return:
+            `keras.distribution.LayoutMap` that contains the sharding spec
+            for all the model weights.
+        """
+        # The weight path and shape of the Llama backbone is like below
+        # token_embedding/embeddings                                  (128256, 2048)
+        # repeat block for decoder
+        # transformer_layer_0/self_attention/query/kernel             (2048, 32, 64)
+        # transformer_layer_0/self_attention/key/kernel               (2048, 8, 64)
+        # transformer_layer_0/self_attention/value/kernel             (2048, 8, 64)
+        # transformer_layer_0/self_attention/attention_output/kernel  (32, 64, 2048)
+        # transformer_layer_0/self_attention_layernorm/scale          (2048,)
+        # transformer_layer_0/feedforward_intermediate_dense/kernel   (2048, 8192)
+        # transformer_layer_0/feedforward_gate_dense/kernel           (2048, 8192)
+        # transformer_layer_0/feedforward_output_dense/kernel         (8192, 2048)
+        # transformer_layer_0/feedforward_layernorm/scale             (2048,)
+
+        if not isinstance(device_mesh, keras.distribution.DeviceMesh):
+            raise ValueError(
+                "Invalid device_mesh type. Expected `keras.distribution.Device`,"
+                f" got {type(device_mesh)}"
+            )
+        if model_parallel_dim_name not in device_mesh.axis_names:
+            raise ValueError(
+                f"{model_parallel_dim_name} is not found in the "
+                f"device_mesh.axis_names. {device_mesh.axis_name=}"
+            )
+        if data_parallel_dim_name not in device_mesh.axis_names:
+            raise ValueError(
+                f"{data_parallel_dim_name} is not found in the "
+                f"device_mesh.axis_names. {device_mesh.axis_name=}"
+            )
+        # Note that it is possible to further config the mesh to be 3D, eg
+        # (data, seq, model). We leave it as 2D for now for simplicity.
+        data_dim = data_parallel_dim_name
+        model_dim = model_parallel_dim_name
+        # The sharding config is based on the Gemma team training config.
+        # See https://arxiv.org/abs/2403.08295
+        layout_map = keras.distribution.LayoutMap(device_mesh)
+        layout_map["token_embedding/embeddings"] = (model_dim, data_dim)
+        layout_map[
+            "transformer_layer.*self_attention.*(query|key|value).kernel"
+        ] = (
+            model_dim,
+            data_dim,
+            None,
+        )
+        layout_map["transformer_layer.*attention_output.kernel"] = (
+            model_dim,
+            None,
+            data_dim,
+        )
+        layout_map[
+            "transformer_layer.*feedforward_intermediate_dense.kernel"
+        ] = (
+            data_dim,
+            model_dim,
+        )
+        layout_map["transformer_layer.*feedforward_gate_dense.kernel"] = (
+            data_dim,
+            model_dim,
+        )
+        layout_map["transformer_layer.*feedforward_output_dense.kernel"] = (
+            model_dim,
+            data_dim,
+        )
+
+        return layout_map

keras_hub/src/models/llama/llama_causal_lm.py

@@ -42,7 +42,9 @@ class LlamaCausalLM(CausalLM):
         self.preprocessor = preprocessor
 
         # === Functional Model ===
-        inputs = backbone.inputs
+        # This must be "backbone.input" i.e. the full input structure,
+        # rather than "backbone.inputs" which is the flattened list of inputs.
+        inputs = backbone.input
         hidden_states = backbone(inputs)
         outputs = backbone.token_embedding(hidden_states, reverse=True)
         super().__init__(

keras_hub/src/models/mistral/mistral_causal_lm.py

@@ -42,7 +42,9 @@ class MistralCausalLM(CausalLM):
         self.preprocessor = preprocessor
 
         # === Functional Model ===
-        inputs = backbone.inputs
+        # This must be "backbone.input" i.e. the full input structure,
+        # rather than "backbone.inputs" which is the flattened list of inputs.
+        inputs = backbone.input
         hidden_states = backbone(inputs)
         outputs = backbone.token_embedding(hidden_states, reverse=True)
         super().__init__(

keras_hub/src/models/pali_gemma/pali_gemma_causal_lm.py

@@ -110,7 +110,9 @@ class PaliGemmaCausalLM(CausalLM):
         self.backbone = backbone
 
         # === Functional Model ===
-        inputs = backbone.inputs
+        # This must be "backbone.input" i.e. the full input structure,
+        # rather than "backbone.inputs" which is the flattened list of inputs.
+        inputs = backbone.input
         hidden_state = backbone(inputs=inputs)
         outputs = backbone.token_embedding(hidden_state, reverse=True)
         outputs = outputs[:, backbone.image_sequence_length :, :]

keras_hub/src/models/phi3/phi3_causal_lm.py

@@ -41,7 +41,9 @@ class Phi3CausalLM(CausalLM):
         self.preprocessor = preprocessor
 
         # === Functional Model ===
-        inputs = backbone.inputs
+        # This must be "backbone.input" i.e. the full input structure,
+        # rather than "backbone.inputs" which is the flattened list of inputs.
+        inputs = backbone.input
         hidden_states = backbone(inputs)
         outputs = backbone.token_embedding(hidden_states, reverse=True)
         super().__init__(

keras_hub/src/tests/test_case.py

@@ -569,6 +569,15 @@ class TestCase(tf.test.TestCase, parameterized.TestCase):
         ds = tf.data.Dataset.from_tensor_slices(train_data).batch(batch_size)
         x, y, sw = keras.utils.unpack_x_y_sample_weight(train_data)
 
+        # Test: the tree struct output by the
+        # preprocessor must match what model expects.
+        preprocessed_data = preprocessor(*train_data)[0]
+        tree.assert_same_structure(
+            preprocessed_data,
+            task._inputs_struct,
+            check_types=False,
+        )
+
         # Test predict.
         output = task.predict(x)
         if expected_output_shape is not None:

keras_hub/src/version_utils.py

@@ -1,7 +1,7 @@
 from keras_hub.src.api_export import keras_hub_export
 
 # Unique source of truth for the version number.
-__version__ = "0.16.1.dev202410150342"
+__version__ = "0.16.1.dev202410170342"
 
 
 @keras_hub_export("keras_hub.version")

{keras_hub_nightly-0.16.1.dev202410150342.dist-info → keras_hub_nightly-0.16.1.dev202410170342.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: keras-hub-nightly
-Version: 0.16.1.dev202410150342
+Version: 0.16.1.dev202410170342
 Summary: Industry-strength Natural Language Processing extensions for Keras.
 Home-page: https://github.com/keras-team/keras-hub
 Author: Keras team

{keras_hub_nightly-0.16.1.dev202410150342.dist-info → keras_hub_nightly-0.16.1.dev202410170342.dist-info}/RECORD

@@ -9,7 +9,7 @@ keras_hub/api/tokenizers/__init__.py,sha256=_f-r_cyUM2fjBB7iO84ThOdqqsAxHNIewJ2E
 keras_hub/api/utils/__init__.py,sha256=Gp1E6gG-RtKQS3PBEQEOz9PQvXkXaJ0ySGMqZ7myN7A,215
 keras_hub/src/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 keras_hub/src/api_export.py,sha256=9pQZK27JObxWZ96QPLBp1OBsjWigh1iuV6RglPGMRk0,1499
-keras_hub/src/version_utils.py,sha256=rY0OztXmBo2_0LAdO10JpHYw9H8oOtH9eTkj7k4SVno,222
+keras_hub/src/version_utils.py,sha256=JKhHcqjvch67-7KPLpPGS3nhs1bP6bpaXaMsSpxp0p4,222
 keras_hub/src/bounding_box/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 keras_hub/src/bounding_box/converters.py,sha256=a5po8DBm87oz2EXfi-0uEZHCMlCJPIb4-MaZIdYx3Dg,17865
 keras_hub/src/bounding_box/formats.py,sha256=YmskOz2BOSat7NaE__J9VfpSNGPJJR0znSzA4lp8MMI,3868
@@ -50,7 +50,7 @@ keras_hub/src/metrics/rouge_l.py,sha256=JlZhMBV6wS_6zMd57pkTc6yxHkEJT9fVQMlPZKek
 keras_hub/src/metrics/rouge_n.py,sha256=JoFtmgjF4Ic263ny6bfD6vMHKreH9le3HnOOxemupRc,3620
 keras_hub/src/models/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 keras_hub/src/models/backbone.py,sha256=2OZx6WAx2q9JK2yue5BoUUipIBjpOJRVNnMjXLVDLRk,11185
-keras_hub/src/models/causal_lm.py,sha256=zGUamLuL2HlTgummUhfnA8Uoe4QMsGGLD4uJazxJe-Y,15079
+keras_hub/src/models/causal_lm.py,sha256=0Asl5v86jm4PnqCoQv7u4Sy8NfMoUvrQxnxQSTQLKog,16721
 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
 keras_hub/src/models/image_classifier.py,sha256=yt6cjhPfqs8A_eWXBsXdXFzn-aRgH2rVHUq7Zu7CyK8,7804
@@ -122,7 +122,7 @@ keras_hub/src/models/deeplab_v3/deeplab_v3_backbone.py,sha256=WyFhuLcjFPFVuNL09b
 keras_hub/src/models/deeplab_v3/deeplab_v3_image_converter.py,sha256=mRkH3HdhpV0fCcQcVXEvIX7SNk-bAMb3SAHzgK-FD5c,371
 keras_hub/src/models/deeplab_v3/deeplab_v3_image_segmeter_preprocessor.py,sha256=hR9S6lNYamY0EBDBo3e1qTCiwtftmLXrN-UYuzfw5Io,581
 keras_hub/src/models/deeplab_v3/deeplab_v3_layers.py,sha256=qmEiolOOriLAojXB67xXW9IOo717kaCGeDVZJLaGY98,7834
-keras_hub/src/models/deeplab_v3/deeplab_v3_presets.py,sha256=tfTOz0H_XX1o-0oM7O3j7OyKxPDIesrV8FMO4IfbbBk,702
+keras_hub/src/models/deeplab_v3/deeplab_v3_presets.py,sha256=lbkP16g-w2-4RKSnISwW-RfvI6qqbE8yZzjRwgiXUIU,703
 keras_hub/src/models/deeplab_v3/deeplab_v3_segmenter.py,sha256=tiMDcCFANHMUx3IVtW3r1P_JTazgPPsbW4IktIytKEU,3650
 keras_hub/src/models/densenet/__init__.py,sha256=r7StyamnWeeZxOk9r4ZYNbS_YVhu9YGPyXhNxljvdPg,269
 keras_hub/src/models/densenet/densenet_backbone.py,sha256=dN9lUwKzO3E2HthNV2x54ozeBEQ0ilNs5uYHshFQpT0,6723
@@ -164,7 +164,7 @@ keras_hub/src/models/falcon/falcon_tokenizer.py,sha256=2B5vmpakj_iVB7T95_8OVreJb
 keras_hub/src/models/falcon/falcon_transformer_decoder.py,sha256=QqIK6v97uBXZFBG3qS6O8HrP9_93uOFzvHQgOiMO2eY,8125
 keras_hub/src/models/gemma/__init__.py,sha256=rVzOJMJ39bgVlT8UdC0t8PlN2c237GKTBmfHIsbPuOQ,251
 keras_hub/src/models/gemma/gemma_attention.py,sha256=1CVN5z9GKoU8TuNMih2_MweDkpd98xSqdic9F8xIBE8,8317
-keras_hub/src/models/gemma/gemma_backbone.py,sha256=EttzmZHWXWl95__09reVFJxT__MtUSUtJAw15-Dao28,12914
+keras_hub/src/models/gemma/gemma_backbone.py,sha256=P5srrrqIrFIBF84KCKKl9vKyYiq0CxjhdcVk76PKVTQ,13377
 keras_hub/src/models/gemma/gemma_causal_lm.py,sha256=BNBoQIf0HoqCooalYsWE-28v5BGUNvL9YdUB8_NSkBU,16770
 keras_hub/src/models/gemma/gemma_causal_lm_preprocessor.py,sha256=bpKkEurWIfa6Kp9s4pz84-sBDSA6ZFNHP8nXG1fFQrg,2912
 keras_hub/src/models/gemma/gemma_decoder_block.py,sha256=tpBfH86Q48EvV0COkd1g2FJg9zHp7ktZBjegs3ehOYo,7588
@@ -187,8 +187,8 @@ keras_hub/src/models/gpt_neo_x/gpt_neo_x_decoder.py,sha256=xSLDgavOhhm3SZc18VN60
 keras_hub/src/models/gpt_neo_x/gpt_neo_x_tokenizer.py,sha256=aKso-8yGrynn3tZ5xm2egcXIBQo3__sWZDBtjmS3ZgU,1991
 keras_hub/src/models/llama/__init__.py,sha256=svVZjGi71R3lVbq0AdbqlXj909mr3Rp9EPXdiO0w0G0,251
 keras_hub/src/models/llama/llama_attention.py,sha256=HzTWtvTjfN_j0vA9-ComstHpI81tzUrJU3RSSvSCaI4,7194
-keras_hub/src/models/llama/llama_backbone.py,sha256=6tkTvAwhFZjnHFIzQbUYlgByMt2qQE2F3sfBluVhON0,6703
-keras_hub/src/models/llama/llama_causal_lm.py,sha256=JyTiCt1mxvf6QNxhjCjAW-aopTL4teS1EHTb_K-RGrs,13109
+keras_hub/src/models/llama/llama_backbone.py,sha256=ElMjhfyTwXcChQPcrKo3bZozeRhzGyCXqOWA_siQFj8,11687
+keras_hub/src/models/llama/llama_causal_lm.py,sha256=9bP4-XDCMgsZuH1ILIMzmwq2Fyy6vkk1Vsht-lMGCNo,13258
 keras_hub/src/models/llama/llama_causal_lm_preprocessor.py,sha256=VTboOMiRBoxHrwP343upLUTsv3AG65r2H8h_PNPVphE,3047
 keras_hub/src/models/llama/llama_decoder.py,sha256=6iERIblED0ZB5w_EUlHks4UvMnsrWONdO_Xdz2OzhWM,8623
 keras_hub/src/models/llama/llama_layernorm.py,sha256=LfRbePHUJs00Ptf7dvNaw3Aj9n1xBMBpE_rS5zzsYMo,1050
@@ -203,7 +203,7 @@ keras_hub/src/models/llama3/llama3_tokenizer.py,sha256=J-KxRc08vGs4olFw_4mtJs0W_
 keras_hub/src/models/mistral/__init__.py,sha256=vjBlzcrIsFSwJKnfwfTNMKstIEKGFTE3kVcdAdfwlnE,263
 keras_hub/src/models/mistral/mistral_attention.py,sha256=HCkUIc2DVIlYC5hhwomENlqLOsKTvbCKF0lx0_OBAyA,7862
 keras_hub/src/models/mistral/mistral_backbone.py,sha256=x4BfyfWTCUXcjPSxdPSl8QITXgzUg1oJlAQt2acZfv4,7245
-keras_hub/src/models/mistral/mistral_causal_lm.py,sha256=OQ3IbdkVlNIXsByZ5ClJoCs0PA86AbMeG6UESbnlfE8,13085
+keras_hub/src/models/mistral/mistral_causal_lm.py,sha256=gEGUnB6yOib9G71n5Em5X8TPOllJW53UXlUCNJkL_ZU,13234
 keras_hub/src/models/mistral/mistral_causal_lm_preprocessor.py,sha256=_4qq-uKktfIg_i081ZWjZGEIYZpedBwtBGpchQQ-qEk,3079
 keras_hub/src/models/mistral/mistral_layer_norm.py,sha256=nimMZ5CTPK8v9eflfrGuzqmv-2vd2rGlPvcHOMwYZyg,1063
 keras_hub/src/models/mistral/mistral_presets.py,sha256=gucgdaFAiU-vRDS1g9zWGHjbDF_jaCiljPibCF4yVqY,1329
@@ -227,7 +227,7 @@ keras_hub/src/models/opt/opt_presets.py,sha256=J1IJ5VRcZZ6UZJSLrxpbWXw39YmbRd_WQ
 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=srZyBsA5tulO_Fb03g9FE-vaw2j9ftfxnAy4P8cYB6o,10916
-keras_hub/src/models/pali_gemma/pali_gemma_causal_lm.py,sha256=qlcBnFtPgKIRtdHgA4rrhiktBJq4h_uV-HriuuRBVwc,11196
+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=Q_sPAULiSo_ZJeXklZjCLhvOMXk8MrPZhEXtL5yNOiI,5175
 keras_hub/src/models/pali_gemma/pali_gemma_image_converter.py,sha256=5yM_jUtrFsWIieiwfFBoP7mtPmQAwywkeLKbd7fhmzk,371
@@ -237,7 +237,7 @@ keras_hub/src/models/pali_gemma/pali_gemma_vit.py,sha256=wP1UtW0WnlRmga-JQRxWTfA
 keras_hub/src/models/phi3/__init__.py,sha256=zIbf1MU-ks91mEkjTRJAsk51N3BBnXDF2JM1vO-13PQ,245
 keras_hub/src/models/phi3/phi3_attention.py,sha256=dN8QwwTP9TxPBDv0MCvObLF3nHm1H6xbYr3T1K0nmg8,9243
 keras_hub/src/models/phi3/phi3_backbone.py,sha256=fY-OY2ZrqxDHglYjTM0OCacBdEQHwj-XNmU0MnXL7iU,8885
-keras_hub/src/models/phi3/phi3_causal_lm.py,sha256=a1TVyDgEv3Sd66Cf7xfa28dESGrUX1bM7pHAw1QfTaw,8240
+keras_hub/src/models/phi3/phi3_causal_lm.py,sha256=kMMq7fQ8hlb_mLO_nU1lGVqILayulVvzzZgl2EvY9_k,8389
 keras_hub/src/models/phi3/phi3_causal_lm_preprocessor.py,sha256=gNx1k-n7d0XDwpNbcZiO9yLkwdXYCvwGyA3b0QCnPAE,3043
 keras_hub/src/models/phi3/phi3_decoder.py,sha256=1raVexz1TkpqvMwW1Zs08KSxTs9gDc6VWUKJ9sM1VFY,9587
 keras_hub/src/models/phi3/phi3_layernorm.py,sha256=Oqu81tGd97Lzx3kG1QEtZ0S6gbfn3GLgRzY8UWGJRBo,1049
@@ -333,7 +333,7 @@ keras_hub/src/samplers/serialization.py,sha256=K6FC4AY1sfOLLIk2k4G783XWnQ_Rk3z1Q
 keras_hub/src/samplers/top_k_sampler.py,sha256=WSyrhmOCan55X2JYAnNWE88rkx66sXqdoerl87nOrDQ,2250
 keras_hub/src/samplers/top_p_sampler.py,sha256=9r29WdqBlrW_2TBma6QqkRps2Uit4a6iZPmq1Gsiuko,3400
 keras_hub/src/tests/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-keras_hub/src/tests/test_case.py,sha256=9-yV87k508ciVOJUBiypd8HmnDTHVtWU2m_RbOHMv5Q,26005
+keras_hub/src/tests/test_case.py,sha256=KMFdQoTqAGotj8Pt8AxXjTJ_f0qwavIGUh-iqN1nQvA,26304
 keras_hub/src/tokenizers/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 keras_hub/src/tokenizers/byte_pair_tokenizer.py,sha256=fGFp3WgPNYGTztpSGMl0kKFjn1bCeZB71lSJfT1eqEE,24052
 keras_hub/src/tokenizers/byte_tokenizer.py,sha256=vjgrTT8FdtZVAlr0mU13alzADcUhtMrzgOs4lYeHvAQ,10648
@@ -368,7 +368,7 @@ keras_hub/src/utils/transformers/convert_mistral.py,sha256=kVhN9h1ZFVhwkNW8p3wnS
 keras_hub/src/utils/transformers/convert_pali_gemma.py,sha256=B1leeDw96Yvu81hYumf66hIid07k5NLqoeWAJgPnaLs,10649
 keras_hub/src/utils/transformers/preset_loader.py,sha256=GS44hZUuGQCtzsyn8z44ZpHdftd3DFemwV2hx2bQa-U,2738
 keras_hub/src/utils/transformers/safetensor_utils.py,sha256=rPK-Uw1CG0DX0d_UAD-r2cG9fw8GI8bvAlrcXfQ9g4c,3323
-keras_hub_nightly-0.16.1.dev202410150342.dist-info/METADATA,sha256=Tj8fIeiKR1xN6oFPr7bWgB_jGSpMm8ZiyE5baY9IC6U,7458
-keras_hub_nightly-0.16.1.dev202410150342.dist-info/WHEEL,sha256=GV9aMThwP_4oNCtvEC2ec3qUYutgWeAzklro_0m4WJQ,91
-keras_hub_nightly-0.16.1.dev202410150342.dist-info/top_level.txt,sha256=N4J6piIWBKa38A4uV-CnIopnOEf8mHAbkNXafXm_CuA,10
-keras_hub_nightly-0.16.1.dev202410150342.dist-info/RECORD,,
+keras_hub_nightly-0.16.1.dev202410170342.dist-info/METADATA,sha256=SEFjDxUxSNwbNsZtCnanXay7BQFJrNDM1zgriF55ATQ,7458
+keras_hub_nightly-0.16.1.dev202410170342.dist-info/WHEEL,sha256=OVMc5UfuAQiSplgO0_WdW7vXVGAt9Hdd6qtN4HotdyA,91
+keras_hub_nightly-0.16.1.dev202410170342.dist-info/top_level.txt,sha256=N4J6piIWBKa38A4uV-CnIopnOEf8mHAbkNXafXm_CuA,10
+keras_hub_nightly-0.16.1.dev202410170342.dist-info/RECORD,,

{keras_hub_nightly-0.16.1.dev202410150342.dist-info → keras_hub_nightly-0.16.1.dev202410170342.dist-info}/WHEEL

@@ -1,5 +1,5 @@
 Wheel-Version: 1.0
-Generator: setuptools (75.1.0)
+Generator: setuptools (75.2.0)
 Root-Is-Purelib: true
 Tag: py3-none-any