keras-hub-nightly 0.21.0.dev202505220409__tar.gz → 0.21.0.dev202505240409__tar.gz

{keras_hub_nightly-0.21.0.dev202505220409 → keras_hub_nightly-0.21.0.dev202505240409}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: keras-hub-nightly
-Version: 0.21.0.dev202505220409
+Version: 0.21.0.dev202505240409
 Summary: Pretrained models for Keras.
 Author-email: Keras team <keras-users@googlegroups.com>
 License-Expression: Apache-2.0

keras_hub_nightly-0.21.0.dev202505240409/keras_hub/src/models/audio_to_text.py

@@ -0,0 +1,66 @@
+from keras_hub.src.models.seq_2_seq_lm import Seq2SeqLM
+
+
+class AudioToText(Seq2SeqLM):
+    """Base class for audio-to-text models.
+
+    `AudioToText` tasks wrap a `keras_hub.models.Backbone` (capable of
+    processing audio and text features) and a
+    `keras_hub.models.AudioToTextPreprocessor` to create a model for
+    audio-to-text tasks like speech recognition or audio transcription.
+
+    These models typically consist of an encoder that processes audio input
+    and a decoder that generates a textual representation.
+
+    `AudioToText` tasks provide a high-level `generate()` method for
+    auto-regressively generating text from audio input. An optional text
+    prompt can also be provided to the decoder to guide generation. The
+    sampling strategy for generation (e.g., greedy, top-k, top-p) can be
+    controlled via the `sampler` argument in the `compile()` method.
+
+    When calling `fit()`, inputs should consist of audio data and corresponding
+    target text transcriptions. The model is trained to predict the target text
+    token-by-token.
+
+    All `AudioToText` tasks include a `from_preset()` constructor which
+    can be used to load pre-trained configurations and weights for specific
+    audio-to-text models.
+    This constructor can also be called on the base `AudioToText` class,
+    which will automatically select the correct subclass based on the preset.
+
+    Examples:
+    ```python
+    # Load a Moonshine backbone with pre-trained weights.
+    # AudioToText is a base class. You will typically work with a specific
+    # implementation, such as `keras_hub.models.MoonshineAudioToText`.
+    # The following examples demonstrate common usage patterns.
+
+    # Initialize a model from a preset using the specific subclass.
+    audio_to_text = keras_hub.models.MoonshineAudioToText.from_preset(
+        "moonshine_base_en"
+    )
+
+    # Initialize a model from a preset using the base class.
+    audio_to_text_model_base = keras_hub.models.AudioToText.from_preset(
+        "moonshine_base_en"
+    )
+
+    # Generate text from an audio input.
+    audio_input_tensor = keras.random.normal((1, 16000, 1))
+    generated_output = audio_to_text_model.generate(
+        {"audio": audio_input_tensor}
+    )
+
+    # Generate conditioned on the `"The quick brown fox."` as an input sequence.
+    prompted_output = audio_to_text_model.generate(
+        {"audio": audio_input_tensor, "text": "The quick brown fox."}
+    )
+
+    # Use a different sampling strategy for generation.
+    audio_to_text_model.compile(sampler="greedy")
+    greedy_output = audio_to_text_model.generate(
+        {"audio": audio_input_tensor}
+    )
+    """
+
+    # TODO: Fill in once audio to text task model requirements are clearer.

keras_hub_nightly-0.21.0.dev202505240409/keras_hub/src/models/audio_to_text_preprocessor.py

@@ -0,0 +1,80 @@
+from keras_hub.src.models.seq_2_seq_lm_preprocessor import Seq2SeqLMPreprocessor
+
+
+class AudioToTextPreprocessor(Seq2SeqLMPreprocessor):
+    """Base class for audio-to-text preprocessing layers.
+
+    `AudioToTextPreprocessor` layers wrap an audio feature extractor (specific
+    to the subclass) and a `keras_hub.tokenizer.Tokenizer` to create a
+    preprocessing layer for audio-to-text tasks. It is intended to be
+    paired with a `keras_hub.models.AudioToText` task.
+
+    Subclasses are expected to handle the conversion of raw audio data into
+    numerical features suitable for an encoder, and raw text data into token IDs
+    for a decoder.
+
+    All `AudioToTextPreprocessor` layers take a dictionary as input,
+    typically with keys like `"audio"` (for audio data) and `"text"` (for
+    target transcriptions or decoder prompts).
+
+    This layer will always output a `(x, y, sample_weight)` tuple, where `x`
+    is a dictionary containing processed audio features for the encoder and
+    tokenized text inputs for the decoder. `y` contains the target token IDs
+    (decoder input tokens shifted by one position), and `sample_weight`
+    indicates padding in `y`. The exact keys and structure of features within
+    `x` will depend on the specific subclass and the paired `AudioToText` model.
+
+    An `AudioToTextPreprocessor` includes `generate_preprocess` and
+    `generate_postprocess` methods for use during inference with an
+    `AudioToText` model's `generate()` method.
+
+    All `AudioToTextPreprocessor` tasks include a `from_preset()` constructor
+    which can be used to load a pre-trained configuration, including tokenizer
+    vocabularies and audio feature extraction settings. Calling `from_preset()`
+    on this base class can instantiate the correct subclass registered for the
+    given preset.
+
+    Examples:
+    ```python
+    preprocessor = keras_hub.models.AudioToTextPreprocessor.from_preset(
+        "moonshine_base_en",
+        decoder_sequence_length=10
+    )
+
+    # Process a single audio-text pair.
+    x = {
+        "audio": keras.random.normal((1, 16000, 1)),
+        "text": ["the quick brown fox"]
+    }
+    x, y, sample_weight = preprocessor(x)
+
+    # Process a batch of audio-text pairs.
+    x = {
+        "audio": keras.random.normal((2, 16000, 1)),
+        "text": ["first sentence", "second sentence"]
+    }
+    x, y, sample_weight = preprocessor(x)
+
+    # With a `tf.data.Dataset`.
+    audio_tf = keras.ops.convert_to_tensor(batch_input["audio"])
+    text_tf = batch_input["text"] # List of strings
+    x = {"audio": audio_tf, "text": text_tf}
+    ds = tf.data.Dataset.from_tensor_slices(x)
+    ds = ds.map(preprocessor, num_parallel_calls=tf.data.AUTOTUNE)
+    ds = ds.batch(2) # Batching after map
+
+    # Generate preprocess and postprocess.
+    x = preprocessor.generate_preprocess({
+        "audio": keras.random.normal((1, 16000, 1)),
+        "text": ["optional prompt text"]
+    })
+    x = preprocessor.generate_postprocess({
+        "decoder_token_ids": keras.ops.array([[10, 20, 30, 2, 0]]),
+        "decoder_padding_mask": keras.ops.array([
+            [True, True, True, True, False]
+        ])
+    })
+    ```
+    """
+
+    # TODO: Fill in once audio to text task model requirements are clearer.

{keras_hub_nightly-0.21.0.dev202505220409 → keras_hub_nightly-0.21.0.dev202505240409}/keras_hub/src/models/gemma/gemma_presets.py

@@ -61,7 +61,7 @@ backbone_presets = {
             "params": 8537680896,
             "path": "gemma",
         },
-        "kaggle_handle": "kaggle://keras/gemma/keras/gemma_7b_en/3",
+        "kaggle_handle": "kaggle://keras/gemma/keras/gemma_7b_en/4",
     },
     "gemma_instruct_7b_en": {
         "metadata": {
@@ -71,7 +71,7 @@ backbone_presets = {
             "params": 8537680896,
             "path": "gemma",
         },
-        "kaggle_handle": "kaggle://keras/gemma/keras/gemma_instruct_7b_en/3",
+        "kaggle_handle": "kaggle://keras/gemma/keras/gemma_instruct_7b_en/4",
     },
     "gemma_1.1_instruct_7b_en": {
         "metadata": {
@@ -82,7 +82,7 @@ backbone_presets = {
             "params": 8537680896,
             "path": "gemma",
         },
-        "kaggle_handle": "kaggle://keras/gemma/keras/gemma_1.1_instruct_7b_en/4",
+        "kaggle_handle": "kaggle://keras/gemma/keras/gemma_1.1_instruct_7b_en/5",
     },
     "code_gemma_7b_en": {
         "metadata": {
@@ -94,7 +94,7 @@ backbone_presets = {
             "params": 8537680896,
             "path": "gemma",
         },
-        "kaggle_handle": "kaggle://keras/codegemma/keras/code_gemma_7b_en/2",
+        "kaggle_handle": "kaggle://keras/codegemma/keras/code_gemma_7b_en/3",
     },
     "code_gemma_instruct_7b_en": {
         "metadata": {
@@ -106,7 +106,7 @@ backbone_presets = {
             "params": 8537680896,
             "path": "gemma",
         },
-        "kaggle_handle": "kaggle://keras/codegemma/keras/code_gemma_instruct_7b_en/2",
+        "kaggle_handle": "kaggle://keras/codegemma/keras/code_gemma_instruct_7b_en/3",
     },
     "code_gemma_1.1_instruct_7b_en": {
         "metadata": {
@@ -118,7 +118,7 @@ backbone_presets = {
             "params": 8537680896,
             "path": "gemma",
         },
-        "kaggle_handle": "kaggle://keras/codegemma/keras/code_gemma_1.1_instruct_7b_en/2",
+        "kaggle_handle": "kaggle://keras/codegemma/keras/code_gemma_1.1_instruct_7b_en/3",
     },
     "gemma2_2b_en": {
         "metadata": {
@@ -144,7 +144,7 @@ backbone_presets = {
             "params": 9241705984,
             "path": "gemma",
         },
-        "kaggle_handle": "kaggle://keras/gemma2/keras/gemma2_9b_en/3",
+        "kaggle_handle": "kaggle://keras/gemma2/keras/gemma2_9b_en/4",
     },
     "gemma2_instruct_9b_en": {
         "metadata": {
@@ -154,7 +154,7 @@ backbone_presets = {
             "params": 9241705984,
             "path": "gemma",
         },
-        "kaggle_handle": "kaggle://keras/gemma2/keras/gemma2_instruct_9b_en/3",
+        "kaggle_handle": "kaggle://keras/gemma2/keras/gemma2_instruct_9b_en/4",
     },
     "gemma2_27b_en": {
         "metadata": {
@@ -162,7 +162,7 @@ backbone_presets = {
             "params": 27227128320,
             "path": "gemma",
         },
-        "kaggle_handle": "kaggle://keras/gemma2/keras/gemma2_27b_en/2",
+        "kaggle_handle": "kaggle://keras/gemma2/keras/gemma2_27b_en/3",
     },
     "gemma2_instruct_27b_en": {
         "metadata": {
@@ -172,7 +172,7 @@ backbone_presets = {
             "params": 27227128320,
             "path": "gemma",
         },
-        "kaggle_handle": "kaggle://keras/gemma2/keras/gemma2_instruct_27b_en/2",
+        "kaggle_handle": "kaggle://keras/gemma2/keras/gemma2_instruct_27b_en/3",
     },
     "shieldgemma_2b_en": {
         "metadata": {

{keras_hub_nightly-0.21.0.dev202505220409 → keras_hub_nightly-0.21.0.dev202505240409}/keras_hub/src/models/gemma3/gemma3_presets.py

@@ -55,7 +55,7 @@ backbone_presets = {
             "params": 11765788416,
             "path": "gemma3",
         },
-        "kaggle_handle": "kaggle://keras/gemma3/keras/gemma3_12b_text/2",
+        "kaggle_handle": "kaggle://keras/gemma3/keras/gemma3_12b_text/3",
     },
     "gemma3_instruct_12b_text": {
         "metadata": {
@@ -66,7 +66,7 @@ backbone_presets = {
             "params": 11765788416,
             "path": "gemma3",
         },
-        "kaggle_handle": "kaggle://keras/gemma3/keras/gemma3_instruct_12b_text/2",
+        "kaggle_handle": "kaggle://keras/gemma3/keras/gemma3_instruct_12b_text/3",
     },
     "gemma3_27b_text": {
         "metadata": {
@@ -77,7 +77,7 @@ backbone_presets = {
             "params": 27009002240,
             "path": "gemma3",
         },
-        "kaggle_handle": "kaggle://keras/gemma3/keras/gemma3_27b_text/3",
+        "kaggle_handle": "kaggle://keras/gemma3/keras/gemma3_27b_text/4",
     },
     "gemma3_instruct_27b_text": {
         "metadata": {
@@ -88,7 +88,7 @@ backbone_presets = {
             "params": 27009002240,
             "path": "gemma3",
         },
-        "kaggle_handle": "kaggle://keras/gemma3/keras/gemma3_instruct_27b_text/2",
+        "kaggle_handle": "kaggle://keras/gemma3/keras/gemma3_instruct_27b_text/3",
     },
     "gemma3_4b": {
         "metadata": {
@@ -121,7 +121,7 @@ backbone_presets = {
             "params": 12187079280,
             "path": "gemma3",
         },
-        "kaggle_handle": "kaggle://keras/gemma3/keras/gemma3_12b/1",
+        "kaggle_handle": "kaggle://keras/gemma3/keras/gemma3_12b/2",
     },
     "gemma3_instruct_12b": {
         "metadata": {
@@ -132,7 +132,7 @@ backbone_presets = {
             "params": 12187079280,
             "path": "gemma3",
         },
-        "kaggle_handle": "kaggle://keras/gemma3/keras/gemma3_instruct_12b/1",
+        "kaggle_handle": "kaggle://keras/gemma3/keras/gemma3_instruct_12b/2",
     },
     "gemma3_27b": {
         "metadata": {
@@ -143,7 +143,7 @@ backbone_presets = {
             "params": 27432062576,
             "path": "gemma3",
         },
-        "kaggle_handle": "kaggle://keras/gemma3/keras/gemma3_27b/1",
+        "kaggle_handle": "kaggle://keras/gemma3/keras/gemma3_27b/2",
     },
     "gemma3_instruct_27b": {
         "metadata": {
@@ -154,6 +154,6 @@ backbone_presets = {
             "params": 27432062576,
             "path": "gemma3",
         },
-        "kaggle_handle": "kaggle://keras/gemma3/keras/gemma3_instruct_27b/1",
+        "kaggle_handle": "kaggle://keras/gemma3/keras/gemma3_instruct_27b/2",
     },
 }

{keras_hub_nightly-0.21.0.dev202505220409 → keras_hub_nightly-0.21.0.dev202505240409}/keras_hub/src/models/llama/llama_presets.py

@@ -8,7 +8,7 @@ backbone_presets = {
             "params": 6738415616,
             "path": "llama",
         },
-        "kaggle_handle": "kaggle://keras/llama2/keras/llama2_7b_en/2",
+        "kaggle_handle": "kaggle://keras/llama2/keras/llama2_7b_en/3",
     },
     "llama2_7b_en_int8": {
         "metadata": {
@@ -30,7 +30,7 @@ backbone_presets = {
             "params": 6738415616,
             "path": "llama",
         },
-        "kaggle_handle": "kaggle://keras/llama2/keras/llama2_instruct_7b_en/2",
+        "kaggle_handle": "kaggle://keras/llama2/keras/llama2_instruct_7b_en/3",
     },
     "llama2_instruct_7b_en_int8": {
         "metadata": {
@@ -52,6 +52,6 @@ backbone_presets = {
             "params": 6738415616,
             "path": "llama",
         },
-        "kaggle_handle": "kaggle://keras/vicuna/keras/vicuna_1.5_7b_en/2",
+        "kaggle_handle": "kaggle://keras/vicuna/keras/vicuna_1.5_7b_en/3",
     },
 }

{keras_hub_nightly-0.21.0.dev202505220409 → keras_hub_nightly-0.21.0.dev202505240409}/keras_hub/src/models/llama3/llama3_presets.py

@@ -8,7 +8,7 @@ backbone_presets = {
             "params": 8030261248,
             "path": "llama3",
         },
-        "kaggle_handle": "kaggle://keras/llama3/keras/llama3_8b_en/4",
+        "kaggle_handle": "kaggle://keras/llama3/keras/llama3_8b_en/5",
     },
     "llama3_8b_en_int8": {
         "metadata": {
@@ -30,7 +30,7 @@ backbone_presets = {
             "params": 8030261248,
             "path": "llama3",
         },
-        "kaggle_handle": "kaggle://keras/llama3/keras/llama3_instruct_8b_en/4",
+        "kaggle_handle": "kaggle://keras/llama3/keras/llama3_instruct_8b_en/5",
     },
     "llama3_instruct_8b_en_int8": {
         "metadata": {

{keras_hub_nightly-0.21.0.dev202505220409 → keras_hub_nightly-0.21.0.dev202505240409}/keras_hub/src/models/mistral/mistral_presets.py

@@ -8,7 +8,7 @@ backbone_presets = {
             "params": 7241732096,
             "path": "mistral",
         },
-        "kaggle_handle": "kaggle://keras/mistral/keras/mistral_7b_en/7",
+        "kaggle_handle": "kaggle://keras/mistral/keras/mistral_7b_en/8",
     },
     "mistral_instruct_7b_en": {
         "metadata": {
@@ -16,7 +16,7 @@ backbone_presets = {
             "params": 7241732096,
             "path": "mistral",
         },
-        "kaggle_handle": "kaggle://keras/mistral/keras/mistral_instruct_7b_en/7",
+        "kaggle_handle": "kaggle://keras/mistral/keras/mistral_instruct_7b_en/8",
     },
     "mistral_0.2_instruct_7b_en": {
         "metadata": {
@@ -24,6 +24,6 @@ backbone_presets = {
             "params": 7241732096,
             "path": "mistral",
         },
-        "kaggle_handle": "kaggle://keras/mistral/keras/mistral_0.2_instruct_7b_en/2",
+        "kaggle_handle": "kaggle://keras/mistral/keras/mistral_0.2_instruct_7b_en/3",
     },
 }

{keras_hub_nightly-0.21.0.dev202505220409 → keras_hub_nightly-0.21.0.dev202505240409}/keras_hub/src/models/mixtral/mixtral_presets.py

@@ -10,7 +10,7 @@ backbone_presets = {
             "params": 46702792704,
             "path": "mixtral",
         },
-        "kaggle_handle": "kaggle://keras/mixtral/keras/mixtral_8_7b_en",
+        "kaggle_handle": "kaggle://keras/mixtral/keras/mixtral_8_7b_en/3",
     },
     "mixtral_8_instruct_7b_en": {
         "metadata": {
@@ -21,6 +21,6 @@ backbone_presets = {
             "params": 46702792704,
             "path": "mixtral",
         },
-        "kaggle_handle": "kaggle://keras/mixtral/keras/mixtral_8_instruct_7b_en",
+        "kaggle_handle": "kaggle://keras/mixtral/keras/mixtral_8_instruct_7b_en/3",
     },
 }

keras_hub_nightly-0.21.0.dev202505240409/keras_hub/src/models/moonshine/__init__.py

@@ -0,0 +1,5 @@
+from keras_hub.src.models.moonshine.moonshine_backbone import MoonshineBackbone
+from keras_hub.src.models.moonshine.moonshine_presets import backbone_presets
+from keras_hub.src.utils.preset_utils import register_presets
+
+register_presets(backbone_presets, MoonshineBackbone)

{keras_hub_nightly-0.21.0.dev202505220409 → keras_hub_nightly-0.21.0.dev202505240409}/keras_hub/src/models/moonshine/moonshine_audio_to_text.py

@@ -1,6 +1,7 @@
 import keras
 
 from keras_hub.src.api_export import keras_hub_export
+from keras_hub.src.models.audio_to_text import AudioToText
 from keras_hub.src.models.moonshine.moonshine_audio_to_text_preprocessor import (  # noqa: E501
     MoonshineAudioToTextPreprocessor,
 )
@@ -9,12 +10,11 @@ from keras_hub.src.models.moonshine.moonshine_backbone import MoonshineBackbone
 from keras_hub.src.models.moonshine.moonshine_backbone import (
     compute_output_lengths,
 )
-from keras_hub.src.models.seq_2_seq_lm import Seq2SeqLM
 from keras_hub.src.utils.tensor_utils import any_equal
 
 
 @keras_hub_export("keras_hub.models.MoonshineAudioToText")
-class MoonshineAudioToText(Seq2SeqLM):
+class MoonshineAudioToText(AudioToText):
     """An end-to-end Moonshine model for audio-to-text tasks.
 
     A Seq2Seq LM designed for audio-to-text tasks, such as speech recognition.

{keras_hub_nightly-0.21.0.dev202505220409 → keras_hub_nightly-0.21.0.dev202505240409}/keras_hub/src/models/moonshine/moonshine_audio_to_text_preprocessor.py

@@ -6,16 +6,18 @@ except ImportError:
     tf = None
 from keras_hub.src.api_export import keras_hub_export
 from keras_hub.src.layers.preprocessing.start_end_packer import StartEndPacker
+from keras_hub.src.models.audio_to_text_preprocessor import (
+    AudioToTextPreprocessor,
+)
 from keras_hub.src.models.moonshine.moonshine_backbone import MoonshineBackbone
 from keras_hub.src.models.moonshine.moonshine_tokenizer import (
     MoonshineTokenizer,
 )
-from keras_hub.src.models.seq_2_seq_lm_preprocessor import Seq2SeqLMPreprocessor
 from keras_hub.src.utils.tensor_utils import preprocessing_function
 
 
 @keras_hub_export("keras_hub.models.MoonshineAudioToTextPreprocessor")
-class MoonshineAudioToTextPreprocessor(Seq2SeqLMPreprocessor):
+class MoonshineAudioToTextPreprocessor(AudioToTextPreprocessor):
     """Moonshine Seq2Seq LM preprocessor for audio-to-text tasks.
 
     This preprocessor converts raw audio and text inputs into a format suitable

{keras_hub_nightly-0.21.0.dev202505220409 → keras_hub_nightly-0.21.0.dev202505240409}/keras_hub/src/models/moonshine/moonshine_presets.py

@@ -9,7 +9,7 @@ backbone_presets = {
             "params": 27092736,
             "path": "moonshine",
         },
-        "kaggle_handle": "kaggle://keras/moonshine/Keras/moonshine_tiny_en",
+        "kaggle_handle": "kaggle://keras/moonshine/Keras/moonshine_tiny_en/1",
     },
     "moonshine_base_en": {
         "metadata": {
@@ -20,6 +20,6 @@ backbone_presets = {
             "params": 61513920,
             "path": "moonshine",
         },
-        "kaggle_handle": "kaggle://keras/moonshine/Keras/moonshine_base_en",
+        "kaggle_handle": "kaggle://keras/moonshine/Keras/moonshine_base_en/1",
     },
 }

{keras_hub_nightly-0.21.0.dev202505220409 → keras_hub_nightly-0.21.0.dev202505240409}/keras_hub/src/models/pali_gemma/pali_gemma_presets.py

@@ -81,7 +81,7 @@ backbone_presets = {
             "path": "pali_gemma2",
             "model_card": "https://www.kaggle.com/models/google/paligemma-2",
         },
-        "kaggle_handle": "kaggle://keras/paligemma2/keras/pali_gemma2_ft_docci_10b_448/2",
+        "kaggle_handle": "kaggle://keras/paligemma2/keras/pali_gemma2_ft_docci_10b_448/3",
     },
     "pali_gemma2_mix_3b_224": {
         "metadata": {
@@ -126,7 +126,7 @@ backbone_presets = {
             "path": "pali_gemma2",
             "model_card": "https://www.kaggle.com/models/google/paligemma-2",
         },
-        "kaggle_handle": "kaggle://keras/paligemma2/keras/pali_gemma2_mix_10b_224/2",
+        "kaggle_handle": "kaggle://keras/paligemma2/keras/pali_gemma2_mix_10b_224/3",
     },
     "pali_gemma2_mix_10b_448": {
         "metadata": {
@@ -141,7 +141,7 @@ backbone_presets = {
             "path": "pali_gemma2",
             "model_card": "https://www.kaggle.com/models/google/paligemma-2",
         },
-        "kaggle_handle": "kaggle://keras/paligemma2/keras/pali_gemma2_mix_10b_448/2",
+        "kaggle_handle": "kaggle://keras/paligemma2/keras/pali_gemma2_mix_10b_448/3",
     },
     "pali_gemma2_mix_28b_224": {
         "metadata": {
@@ -156,7 +156,7 @@ backbone_presets = {
             "path": "pali_gemma2",
             "model_card": "https://www.kaggle.com/models/google/paligemma-2",
         },
-        "kaggle_handle": "kaggle://keras/paligemma2/keras/pali_gemma2_28b_mix_224/2",
+        "kaggle_handle": "kaggle://keras/paligemma2/keras/pali_gemma2_28b_mix_224/3",
     },
     "pali_gemma2_mix_28b_448": {
         "metadata": {
@@ -171,7 +171,7 @@ backbone_presets = {
             "path": "pali_gemma2",
             "model_card": "https://www.kaggle.com/models/google/paligemma-2",
         },
-        "kaggle_handle": "kaggle://keras/paligemma2/keras/pali_gemma2_28b_mix_448/2",
+        "kaggle_handle": "kaggle://keras/paligemma2/keras/pali_gemma2_28b_mix_448/3",
     },
     "pali_gemma2_pt_3b_224": {
         "metadata": {
@@ -231,7 +231,7 @@ backbone_presets = {
             "path": "pali_gemma2",
             "model_card": "https://www.kaggle.com/models/google/paligemma-2",
         },
-        "kaggle_handle": "kaggle://keras/paligemma2/keras/pali_gemma2_pt_10b_224/2",
+        "kaggle_handle": "kaggle://keras/paligemma2/keras/pali_gemma2_pt_10b_224/3",
     },
     "pali_gemma2_pt_10b_448": {
         "metadata": {
@@ -246,7 +246,7 @@ backbone_presets = {
             "path": "pali_gemma2",
             "model_card": "https://www.kaggle.com/models/google/paligemma-2",
         },
-        "kaggle_handle": "kaggle://keras/paligemma2/keras/pali_gemma2_pt_10b_448/2",
+        "kaggle_handle": "kaggle://keras/paligemma2/keras/pali_gemma2_pt_10b_448/3",
     },
     "pali_gemma2_pt_10b_896": {
         "metadata": {
@@ -261,7 +261,7 @@ backbone_presets = {
             "path": "pali_gemma2",
             "model_card": "https://www.kaggle.com/models/google/paligemma-2",
         },
-        "kaggle_handle": "kaggle://keras/paligemma2/keras/pali_gemma2_pt_10b_896/2",
+        "kaggle_handle": "kaggle://keras/paligemma2/keras/pali_gemma2_pt_10b_896/3",
     },
     "pali_gemma2_pt_28b_224": {
         "metadata": {
@@ -276,7 +276,7 @@ backbone_presets = {
             "path": "pali_gemma2",
             "model_card": "https://www.kaggle.com/models/google/paligemma-2",
         },
-        "kaggle_handle": "kaggle://keras/paligemma2/keras/pali_gemma2_pt_28b_224/3",
+        "kaggle_handle": "kaggle://keras/paligemma2/keras/pali_gemma2_pt_28b_224/4",
     },
     "pali_gemma2_pt_28b_448": {
         "metadata": {
@@ -291,7 +291,7 @@ backbone_presets = {
             "path": "pali_gemma2",
             "model_card": "https://www.kaggle.com/models/google/paligemma-2",
         },
-        "kaggle_handle": "kaggle://keras/paligemma2/keras/pali_gemma2_pt_28b_448/2",
+        "kaggle_handle": "kaggle://keras/paligemma2/keras/pali_gemma2_pt_28b_448/3",
     },
     "pali_gemma2_pt_28b_896": {
         "metadata": {
@@ -306,6 +306,6 @@ backbone_presets = {
             "path": "pali_gemma2",
             "model_card": "https://www.kaggle.com/models/google/paligemma-2",
         },
-        "kaggle_handle": "kaggle://keras/paligemma2/keras/pali_gemma2_pt_28b_896/2",
+        "kaggle_handle": "kaggle://keras/paligemma2/keras/pali_gemma2_pt_28b_896/3",
     },
 }

{keras_hub_nightly-0.21.0.dev202505220409 → keras_hub_nightly-0.21.0.dev202505240409}/keras_hub/src/models/qwen/qwen_presets.py

@@ -7,7 +7,7 @@ backbone_presets = {
             "params": 494032768,
             "path": "qwen",
         },
-        "kaggle_handle": "kaggle://keras/qwen/keras/qwen2.5_0.5b_en",
+        "kaggle_handle": "kaggle://keras/qwen/keras/qwen2.5_0.5b_en/1",
     },
     "qwen2.5_3b_en": {
         "metadata": {
@@ -15,7 +15,7 @@ backbone_presets = {
             "params": 3085938688,
             "path": "qwen",
         },
-        "kaggle_handle": "kaggle://keras/qwen/keras/qwen2.5_3b_en",
+        "kaggle_handle": "kaggle://keras/qwen/keras/qwen2.5_3b_en/1",
     },
     "qwen2.5_7b_en": {
         "metadata": {
@@ -23,7 +23,7 @@ backbone_presets = {
             "params": 6993420288,
             "path": "qwen",
         },
-        "kaggle_handle": "kaggle://keras/qwen/keras/qwen2.5_7b_en/2",
+        "kaggle_handle": "kaggle://keras/qwen/keras/qwen2.5_7b_en/3",
     },
     "qwen2.5_instruct_0.5b_en": {
         "metadata": {
@@ -34,7 +34,7 @@ backbone_presets = {
             "params": 494032768,
             "path": "qwen",
         },
-        "kaggle_handle": "kaggle://keras/qwen/keras/qwen2.5_instruct_0.5b_en",
+        "kaggle_handle": "kaggle://keras/qwen/keras/qwen2.5_instruct_0.5b_en/1",
     },
     "qwen2.5_instruct_32b_en": {
         "metadata": {
@@ -45,7 +45,7 @@ backbone_presets = {
             "params": 32763876352,
             "path": "qwen",
         },
-        "kaggle_handle": "kaggle://keras/qwen/keras/qwen2.5_instruct_32b_en",
+        "kaggle_handle": "kaggle://keras/qwen/keras/qwen2.5_instruct_32b_en/2",
     },
     "qwen2.5_instruct_72b_en": {
         "metadata": {
@@ -56,6 +56,6 @@ backbone_presets = {
             "params": 72706203648,
             "path": "qwen",
         },
-        "kaggle_handle": "kaggle://keras/qwen/keras/qwen2.5_instruct_72b_en",
+        "kaggle_handle": "kaggle://keras/qwen/keras/qwen2.5_instruct_72b_en/2",
     },
 }

{keras_hub_nightly-0.21.0.dev202505220409 → keras_hub_nightly-0.21.0.dev202505240409}/keras_hub/src/models/qwen_moe/qwen_moe_presets.py

@@ -10,6 +10,6 @@ backbone_presets = {
             "params": 14315784192,
             "path": "qwen-1.5-moe",
         },
-        "kaggle_handle": "kaggle://keras/qwen-1.5-moe/Keras/qwen1.5_moe_2.7b_en",
+        "kaggle_handle": "kaggle://keras/qwen-1.5-moe/Keras/qwen1.5_moe_2.7b_en/3",
     },
 }

{keras_hub_nightly-0.21.0.dev202505220409 → keras_hub_nightly-0.21.0.dev202505240409}/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.dev202505220409"
+__version__ = "0.21.0.dev202505240409"
 
 
 @keras_hub_export("keras_hub.version")

{keras_hub_nightly-0.21.0.dev202505220409 → keras_hub_nightly-0.21.0.dev202505240409}/keras_hub_nightly.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: keras-hub-nightly
-Version: 0.21.0.dev202505220409
+Version: 0.21.0.dev202505240409
 Summary: Pretrained models for Keras.
 Author-email: Keras team <keras-users@googlegroups.com>
 License-Expression: Apache-2.0