keras-hub-nightly 0.23.0.dev202510240418__py3-none-any.whl → 0.24.0.dev202511090424__py3-none-any.whl

keras_hub/src/models/dinov3/dinov3_presets.py

@@ -0,0 +1,4 @@
+"""DINOV3 model preset configurations."""
+
+# Metadata for loading pretrained model weights.
+backbone_presets = {}

keras_hub/src/models/gemma3/gemma3_presets.py

@@ -181,4 +181,43 @@ backbone_presets = {
         },
         "kaggle_handle": "kaggle://keras/gemma3/keras/gemma3_instruct_270m/4",
     },
+    "medgemma_instruct_4b": {
+        "metadata": {
+            "description": (
+                "A 4 billion parameter model based on Gemma 3. "
+                "This model is trained for performance on medical text"
+                "and image comprehension and is optimized for medical"
+                "applications that involve a text generation component."
+            ),
+            "params": 4300079472,
+            "path": "gemma3",
+        },
+        "kaggle_handle": "kaggle://keras/medgemma/keras/medgemma_instruct_4b/1",
+    },
+    "medgemma_instruct_27b": {
+        "metadata": {
+            "description": (
+                "A 27 billion parameter model based on Gemma 3. "
+                "This model trained for performance on medical text "
+                "and image comprehension and is optimized for medical "
+                "applications that involve a text generation component."
+            ),
+            "params": 27432406640,
+            "path": "gemma3",
+        },
+        "kaggle_handle": "kaggle://keras/medgemma/keras/medgemma_instruct_27b/1",
+    },
+    "medgemma_instruct_27b_text": {
+        "metadata": {
+            "description": (
+                "A 27 billion parameter text-only model based on Gemma 3. "
+                "This model is trained for performance on medical text "
+                "comprehension and is optimized for medical applications "
+                "that involve a text generation component."
+            ),
+            "params": 27009002240,
+            "path": "gemma3",
+        },
+        "kaggle_handle": "kaggle://keras/medgemma/keras/medgemma_instruct_27b_text/1",
+    },
 }

keras_hub/src/models/siglip/siglip_presets.py

@@ -321,4 +321,19 @@ backbone_presets = {
         },
         "kaggle_handle": "kaggle://keras/siglip/keras/siglip2_so400m_patch16_512/1",
     },
+    "medsiglip_900m_448": {
+        "metadata": {
+            "description": (
+                "A 900 million parameter variant of SigLIP trained to encode "
+                "medical images and text into a common embedding space. "
+                "MedSigLIP contains a vision encoder and a text encoder, and "
+                "supports 448x448 image resolution with up to 64 text tokens."
+            ),
+            "params": 878301426,
+            "official_name": "SigLIP2",
+            "path": "siglip",
+            "model_card": "https://huggingface.co/google/medsiglip-448#medsiglip-model-card",
+        },
+        "kaggle_handle": "kaggle://keras/medsiglip/keras/medsiglip_900m_448/1",
+    },
 }

keras_hub/src/utils/tensor_utils.py

@@ -12,9 +12,11 @@ from packaging import version
 
 try:
     import tensorflow as tf
-    import tensorflow_text as tf_text
 except ImportError:
     tf = None
+try:
+    import tensorflow_text as tf_text
+except ImportError:
     tf_text = None
 
 

keras_hub/src/utils/transformers/convert_dinov3.py

@@ -0,0 +1,106 @@
+import numpy as np
+
+from keras_hub.src.models.dinov3.dinov3_backbone import DINOV3Backbone
+
+backbone_cls = DINOV3Backbone
+
+
+def convert_backbone_config(transformers_config):
+    image_size = transformers_config["image_size"]
+    return {
+        "patch_size": transformers_config["patch_size"],
+        "num_layers": transformers_config["num_hidden_layers"],
+        "hidden_dim": transformers_config["hidden_size"],
+        "num_heads": transformers_config["num_attention_heads"],
+        "intermediate_dim": transformers_config["intermediate_size"],
+        "layer_scale_init_value": transformers_config["layerscale_value"],
+        "num_register_tokens": transformers_config["num_register_tokens"],
+        "use_mask_token": True,
+        "hidden_activation": transformers_config["hidden_act"],
+        "use_gated_mlp": transformers_config["use_gated_mlp"],
+        "use_query_bias": transformers_config["query_bias"],
+        "use_key_bias": transformers_config["key_bias"],
+        "use_value_bias": transformers_config["value_bias"],
+        "use_proj_bias": transformers_config["proj_bias"],
+        "use_mlp_bias": transformers_config["mlp_bias"],
+        "attention_dropout": transformers_config["attention_dropout"],
+        "drop_path_rate": transformers_config["drop_path_rate"],
+        "layer_norm_eps": transformers_config["layer_norm_eps"],
+        "image_shape": (image_size, image_size, 3),
+        "rope_theta": transformers_config["rope_theta"],
+        "apply_layernorm": False,
+    }
+
+
+def convert_weights(backbone, loader, transformers_config):
+    if not isinstance(backbone, DINOV3Backbone):
+        raise ValueError(
+            "The provided backbone must be an instance of DINOV3Backbone. "
+            f"Received: {type(backbone)}"
+        )
+
+    def port_ln(keras_variable, weight_key):
+        loader.port_weight(keras_variable.gamma, f"{weight_key}.weight")
+        loader.port_weight(keras_variable.beta, f"{weight_key}.bias")
+
+    def port_dense(keras_variable, weight_key):
+        loader.port_weight(
+            keras_variable.kernel,
+            f"{weight_key}.weight",
+            hook_fn=lambda x, _: x.T,
+        )
+        if keras_variable.bias is not None:
+            loader.port_weight(keras_variable.bias, f"{weight_key}.bias")
+
+    # Embedding.
+    loader.port_weight(
+        keras_variable=backbone.embeddings.cls_token,
+        hf_weight_key="embeddings.cls_token",
+    )
+    if backbone.use_mask_token:
+        loader.port_weight(
+            keras_variable=backbone.embeddings.mask_token,
+            hf_weight_key="embeddings.mask_token",
+        )
+    if backbone.num_register_tokens > 0:
+        loader.port_weight(
+            keras_variable=backbone.embeddings.register_tokens,
+            hf_weight_key="embeddings.register_tokens",
+        )
+    loader.port_weight(
+        keras_variable=backbone.embeddings.patch_embeddings.projection.kernel,
+        hf_weight_key="embeddings.patch_embeddings.weight",
+        hook_fn=lambda x, _: np.transpose(x, (2, 3, 1, 0)),
+    )
+    loader.port_weight(
+        keras_variable=backbone.embeddings.patch_embeddings.projection.bias,
+        hf_weight_key="embeddings.patch_embeddings.bias",
+    )
+
+    # Encoder.
+    for i, layer in enumerate(backbone.encoder.layers):
+        prefix = f"layer.{i}"
+        port_ln(layer.norm1, f"{prefix}.norm1")
+        port_dense(layer.attention.query_dense, f"{prefix}.attention.q_proj")
+        port_dense(layer.attention.key_dense, f"{prefix}.attention.k_proj")
+        port_dense(layer.attention.value_dense, f"{prefix}.attention.v_proj")
+        port_dense(layer.attention.output_dense, f"{prefix}.attention.o_proj")
+
+        loader.port_weight(
+            keras_variable=layer.layer_scale1.lambda1,
+            hf_weight_key=f"{prefix}.layer_scale1.lambda1",
+        )
+        port_ln(layer.norm2, f"{prefix}.norm2")
+        if backbone.use_gated_mlp:
+            port_dense(layer.mlp.gate_proj, f"{prefix}.mlp.gate_proj")
+            port_dense(layer.mlp.up_proj, f"{prefix}.mlp.up_proj")
+            port_dense(layer.mlp.down_proj, f"{prefix}.mlp.down_proj")
+        else:
+            port_dense(layer.mlp.up_proj, f"{prefix}.mlp.up_proj")
+            port_dense(layer.mlp.down_proj, f"{prefix}.mlp.down_proj")
+        loader.port_weight(
+            keras_variable=layer.layer_scale2.lambda1,
+            hf_weight_key=f"{prefix}.layer_scale2.lambda1",
+        )
+
+    port_ln(backbone.layernorm, "norm")

keras_hub/src/utils/transformers/preset_loader.py

@@ -8,6 +8,7 @@ from keras_hub.src.utils.transformers import convert_bart
 from keras_hub.src.utils.transformers import convert_bert
 from keras_hub.src.utils.transformers import convert_deit
 from keras_hub.src.utils.transformers import convert_dinov2
+from keras_hub.src.utils.transformers import convert_dinov3
 from keras_hub.src.utils.transformers import convert_distilbert
 from keras_hub.src.utils.transformers import convert_esm
 from keras_hub.src.utils.transformers import convert_gemma
@@ -42,6 +43,8 @@ class TransformersPresetLoader(PresetLoader):
             self.converter = convert_distilbert
         elif model_type in ("dinov2", "dinov2_with_registers"):
             self.converter = convert_dinov2
+        elif model_type == "dinov3_vit":
+            self.converter = convert_dinov3
         elif model_type == "esm":
             self.converter = convert_esm
         elif model_type in ("gemma", "gemma2"):

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__ = "0.23.0.dev202510240418"
+__version__ = "0.24.0.dev202511090424"
 
 
 @keras_hub_export("keras_hub.version")

{keras_hub_nightly-0.23.0.dev202510240418.dist-info → keras_hub_nightly-0.24.0.dev202511090424.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: keras-hub-nightly
-Version: 0.23.0.dev202510240418
+Version: 0.24.0.dev202511090424
 Summary: Pretrained models for Keras.
 Author-email: Keras team <keras-users@googlegroups.com>
 License-Expression: Apache-2.0

{keras_hub_nightly-0.23.0.dev202510240418.dist-info → keras_hub_nightly-0.24.0.dev202511090424.dist-info}/RECORD

@@ -1,11 +1,11 @@
 keras_hub/__init__.py,sha256=bJbUZkqwhZvTb1Tqx1fbkq6mzBYiEyq-Hin3oQIkhdE,558
-keras_hub/layers/__init__.py,sha256=ufJKHxMTFhwp--E3ixfGCZqq89pZOUOxCQYgI5pEUA8,5944
+keras_hub/layers/__init__.py,sha256=hY5hZX5oOxRTFxfPe2hGhrHWJwF1kB7QiwITSS4Xp2A,6061
 keras_hub/metrics/__init__.py,sha256=KYalsMPBnfwim9BdGHFfJ5WxUKFXOQ1QoKIMT_0lwlM,439
-keras_hub/models/__init__.py,sha256=yazrEg57HafE6Fgr-bfDXR3iFrArx6RytOEnV1CJJS8,32068
+keras_hub/models/__init__.py,sha256=XGYkwfBVZiPw5ZjSV5S_n3FnkPf06yYNzxZjXMhiX70,32166
 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=QM9WStK1yy-8Q3X8sOhi5Mxr-jNfz4ET6FAFE-Qit0I,222
+keras_hub/src/version.py,sha256=0oZ2eQ3pK7UNLgYg6OOna2ubpYCdPpH9WrlbvIq-QC0,222
 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
@@ -174,8 +174,13 @@ keras_hub/src/models/depth_anything/interpolate.py,sha256=qwrPGP6wA4jZ-XcSeulhky
 keras_hub/src/models/dinov2/__init__.py,sha256=qacZi82EfAloVND4gDLZjqgR5_yVdz_dc4mMKyCsjOA,257
 keras_hub/src/models/dinov2/dinov2_backbone.py,sha256=QH3lzE1EnxTcOSii9KS1Qx3lq0XcZMsvElB7AL_ejZY,10672
 keras_hub/src/models/dinov2/dinov2_image_converter.py,sha256=gfFROdYV5rOzo3kJFlRvRHYjek8z9YirKfrFwlVJO3g,342
-keras_hub/src/models/dinov2/dinov2_layers.py,sha256=UCcia2kWA1O37SMmUbyhUcSXmUpLfNjk1E6mPTPDrF0,33647
+keras_hub/src/models/dinov2/dinov2_layers.py,sha256=wo80Re043Gjly-XE-sT01QAYq3h793zhmU-Nb6SFN4g,33702
 keras_hub/src/models/dinov2/dinov2_presets.py,sha256=ho493GPH98K4LH1E54UV2qZZ4h7Un9ylbBmMQjNoKh4,2937
+keras_hub/src/models/dinov3/__init__.py,sha256=AI7vTZJBG6Ygb48o6pXtHzxKk0Rek3p7-HffD-Y48cc,257
+keras_hub/src/models/dinov3/dinov3_backbone.py,sha256=WDHipJSG10seRzYG_hARifF52wqhj9enkhuZ6mgJmjw,10511
+keras_hub/src/models/dinov3/dinov3_image_converter.py,sha256=_oHDcI2CoxjbSLxLfkK1zEPcf4Goy0S66igmrXt58cQ,342
+keras_hub/src/models/dinov3/dinov3_layers.py,sha256=w5K2btblrgrULqzPQdbvtkyR5Px2UZkqcZQ7jq2K3Uk,37169
+keras_hub/src/models/dinov3/dinov3_presets.py,sha256=AXXdrgrs9WBrsGlac0TgWV0DIPnvKdlxD3kUhbii1sk,114
 keras_hub/src/models/distil_bert/__init__.py,sha256=3Z0w-Mt3aOR0u9RGzjHQ7B3J3qBF2pGjupDGQ9yyzoc,303
 keras_hub/src/models/distil_bert/distil_bert_backbone.py,sha256=rnAf_GokB3wAeJwVZtgUKQO_bKJIa8RavhL_ykTJpNw,6440
 keras_hub/src/models/distil_bert/distil_bert_masked_lm.py,sha256=axeZd5UcxFr3_Q8H4yG10CINh93wbcyjlPLauqe5N9E,4289
@@ -247,7 +252,7 @@ keras_hub/src/models/gemma3/gemma3_causal_lm_preprocessor.py,sha256=vjt4N-zr0Eb5
 keras_hub/src/models/gemma3/gemma3_decoder_block.py,sha256=CYwYazqwakLNfhOLBl_8Q2TVZcMcOxMtiZtuVlk_hoo,11470
 keras_hub/src/models/gemma3/gemma3_image_converter.py,sha256=czi5JrTyKiK0nFzvonviBIX8jjvLHqvGNA9RyheB31k,536
 keras_hub/src/models/gemma3/gemma3_interleave_embeddings.py,sha256=CfYdudk5En9iU6vEnrcrEWIztloD1r8VzF2extqAhAM,4616
-keras_hub/src/models/gemma3/gemma3_presets.py,sha256=FGAHYE4HTLuiceuiCKBJtc1aNd7OgMB59KD0s6Ba_Fg,6105
+keras_hub/src/models/gemma3/gemma3_presets.py,sha256=1GZSwsGRA19RllhZPR-kFjH5y9A6308V3TYfqHAnXUw,7744
 keras_hub/src/models/gemma3/gemma3_tokenizer.py,sha256=ZaBclFIwzJkSXDuZMBQLHUKV8RWEdZ_dsJMvMcc3qXw,3215
 keras_hub/src/models/gemma3/gemma3_vision_encoder.py,sha256=7XI0oBjIfJItV5w90t5bWb3C2KzjhvDnIC7wjIq4Cns,20850
 keras_hub/src/models/gemma3/rms_normalization.py,sha256=fku-JEo2sNy-ytX7ySD1sRzdhRAPmYex_z8oFk1NiG8,833
@@ -467,7 +472,7 @@ keras_hub/src/models/siglip/siglip_image_converter.py,sha256=yjYc0XOyL37WLlr-X6V
 keras_hub/src/models/siglip/siglip_layers.py,sha256=c20n6v3cFsI-Im9GBVTknhj_IpX79I4a-fajBKRMzQA,19893
 keras_hub/src/models/siglip/siglip_loss.py,sha256=n6zmOeL0o7Nwb5iaoEZfrxiAsQoqZ9yLIlaCJsAfTg4,1442
 keras_hub/src/models/siglip/siglip_preprocessor.py,sha256=r1Ej7hVwr5BudFYTHkjW5yc3lk4OYZD1s3t32lKkuec,5660
-keras_hub/src/models/siglip/siglip_presets.py,sha256=gOzSVhLskAthfzq8jWOtQWv14euaqS2ywcZlNfivDOI,13164
+keras_hub/src/models/siglip/siglip_presets.py,sha256=jtIQrNcq14othG1QgwBEfozEmoHdXXW270qylQEmA8E,13864
 keras_hub/src/models/siglip/siglip_text_encoder.py,sha256=xOVvzyQHLX9ne30y4ussar99gNMXPXHYKlkbCX_On2Y,5380
 keras_hub/src/models/siglip/siglip_tokenizer.py,sha256=j_67JbIHJDRk-CbiemG2dgAO6lp3_0_JdnfroZ90G18,2579
 keras_hub/src/models/siglip/siglip_vision_encoder.py,sha256=CaNaFq5thBC3TUXXOf2qknk5vWsauM20ZoaDPYRnXcs,5927
@@ -578,7 +583,7 @@ keras_hub/src/utils/openvino_utils.py,sha256=P1ZvedLv91LZD-UAgAo2dy6WC5305elh1qv
 keras_hub/src/utils/pipeline_model.py,sha256=jgzB6NQPSl0KOu08N-TazfOnXnUJbZjH2EXXhx25Ftg,9084
 keras_hub/src/utils/preset_utils.py,sha256=lyCg_PRcYH1Jy8lGKaO8sgpIbMrP-Ik66EbjGD4gizc,37677
 keras_hub/src/utils/python_utils.py,sha256=N8nWeO3san4YnGkffRXG3Ix7VEIMTKSN21FX5TuL7G8,202
-keras_hub/src/utils/tensor_utils.py,sha256=bGM0pK-x0R4640emul49GfSZ3p4OSvOaVzZZPlm6eiM,16957
+keras_hub/src/utils/tensor_utils.py,sha256=tULr53SZLCczN_BD7XvbAq9c9bFVZTn7aYcLbqmbfx8,16982
 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
@@ -598,6 +603,7 @@ keras_hub/src/utils/transformers/convert_bart.py,sha256=Tk4h9Md9rwN5wjQbGIVrC7qz
 keras_hub/src/utils/transformers/convert_bert.py,sha256=4gQqXCJzC9QWdLPDUAq741K8t_kjPIET050YjUnLeDA,5977
 keras_hub/src/utils/transformers/convert_deit.py,sha256=ubcqYzMlhWTCE2S_TsXICCMmqjN9RsQPaw_70vArnjo,5306
 keras_hub/src/utils/transformers/convert_dinov2.py,sha256=Zmxz33hKJCcykQOcW8XhG_Yy1l8XqIYam1cjzM69-Mk,6986
+keras_hub/src/utils/transformers/convert_dinov3.py,sha256=rZqowTASKSAQQ1HrwlD9_tY7VAQHY_C4_61ky5wUbvE,4448
 keras_hub/src/utils/transformers/convert_distilbert.py,sha256=SlfIRhSRk5c1ir2HGiDPiXa5XdOId_DbcnZO9lbwyZ8,6498
 keras_hub/src/utils/transformers/convert_esm.py,sha256=rOgGnNY37ZbYnoVC3L-Y-yGGAxTRmYtQV0nJoandH2Y,6214
 keras_hub/src/utils/transformers/convert_gemma.py,sha256=ElCgwBpSN5Q7rV5PJawTsoytPzs5ZjuwoY60YAe8y_A,6533
@@ -613,13 +619,13 @@ keras_hub/src/utils/transformers/convert_qwen_moe.py,sha256=a7R28aln-PdAcNuKAXdr
 keras_hub/src/utils/transformers/convert_smollm3.py,sha256=V2vWES85YSNXNx39I8OwAcOvSpb9KxUscrDr7ra-LPA,5281
 keras_hub/src/utils/transformers/convert_t5gemma.py,sha256=DPOwd61UhjspKuCsk3_EaNvSADGP_f8KLcZARHYVk5Y,9490
 keras_hub/src/utils/transformers/convert_vit.py,sha256=YAmXh519ecSgEO5B4g-aEQg1Bb_6ifFafLMqDTfLn_c,5259
-keras_hub/src/utils/transformers/preset_loader.py,sha256=PmB4wcPaMlqMhrhk2bYt74TvRHRsZgpodfKlzixfr-Q,5219
+keras_hub/src/utils/transformers/preset_loader.py,sha256=alzuIEhDI6gLpEw05wPJVbOJ2LhwmLB_s7JhDqkb4ec,5364
 keras_hub/src/utils/transformers/safetensor_utils.py,sha256=CYUHyA4y-B61r7NDnCsFb4t_UmSwZ1k9L-8gzEd6KRg,3339
 keras_hub/src/utils/transformers/export/gemma.py,sha256=xX_vfQwvFZ_-lQX4kgMNOGKL7fL_1yk7QyGYV2Qyly4,4699
 keras_hub/src/utils/transformers/export/hf_exporter.py,sha256=Qk52c6LIA2eMHUNY9Vy4STJSpnhLMdJ_t-3ljqhSr4k,5081
 keras_hub/tokenizers/__init__.py,sha256=XFOxDmM1Mz9TxiE8ICZK_-yTTyRFywUUiVwRIzz2QZ8,4770
 keras_hub/utils/__init__.py,sha256=jXPqVGBpJr_PpYmqD8aDG-fRMlxH-ulqCR2SZMn288Y,646
-keras_hub_nightly-0.23.0.dev202510240418.dist-info/METADATA,sha256=yjCRNBfVk10uy8m0lllvjp7-EzChQXXzY8LC-0tWjNs,7395
-keras_hub_nightly-0.23.0.dev202510240418.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-keras_hub_nightly-0.23.0.dev202510240418.dist-info/top_level.txt,sha256=N4J6piIWBKa38A4uV-CnIopnOEf8mHAbkNXafXm_CuA,10
-keras_hub_nightly-0.23.0.dev202510240418.dist-info/RECORD,,
+keras_hub_nightly-0.24.0.dev202511090424.dist-info/METADATA,sha256=sQwmYQhujdR3LE7rPYvUYCmmx0jOFzEMRXOv3QqnASA,7395
+keras_hub_nightly-0.24.0.dev202511090424.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+keras_hub_nightly-0.24.0.dev202511090424.dist-info/top_level.txt,sha256=N4J6piIWBKa38A4uV-CnIopnOEf8mHAbkNXafXm_CuA,10
+keras_hub_nightly-0.24.0.dev202511090424.dist-info/RECORD,,