add: differential privacy research model

docs/source/en/_toctree.yml

@@ -709,6 +709,8 @@
         title: UL2
       - local: model_doc/umt5
         title: UMT5
+      - local: model_doc/vaultgemma
+        title: VaultGemma
       - local: model_doc/xmod
         title: X-MOD
       - local: model_doc/xglm

docs/source/en/model_doc/vaultgemma.md

@@ -0,0 +1,103 @@
+<!--Copyright 2025 the HuggingFace Team. All rights reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+
+
+⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be rendered properly in your Markdown viewer.
+
+-->
+
+# VaultGemma
+
+## Overview
+
+[VaultGemma](https://services.google.com/fh/files/blogs/vaultgemma_tech_report.pdf) is a text-only decoder model
+derived from [Gemma 2](https://huggingface.co/docs/transformers/en/model_doc/gemma2), notably it drops the norms after
+the Attention and MLP blocks, and uses full attention for all layers instead of alternating between full attention and
+local sliding attention. VaultGemma is available as a pretrained model with 1B parameters that uses a 1024 token
+sequence length.
+
+VaultGemma was trained from scratch with sequence-level differential privacy (DP). Its training data includes the same
+mixture as the [Gemma 2 models](https://huggingface.co/collections/google/gemma-2-release-667d6600fd5220e7b967f315),
+consisting of a number of documents of varying lengths. Additionally, it is trained using
+[DP stochastic gradient descent (DP-SGD)](https://arxiv.org/abs/1607.00133) and provides a
+(ε ≤ 2.0, δ ≤ 1.1e-10)-sequence-level DP guarantee, where a sequence consists of 1024 consecutive tokens extracted from
+heterogeneous data sources. Specifically, the privacy unit of the guarantee is for the sequences after sampling and
+packing of the mixture.
+
+> [!TIP]
+> Click on the VaultGemma models in the right sidebar for more examples of how to apply VaultGemma to different language tasks.
+
+The example below demonstrates how to chat with the model with [`Pipeline`], the [`AutoModel`] class, or from the
+command line.
+
+<hfoptions id="usage">
+<hfoption id="Pipeline">
+
+
+```python
+from transformers import pipeline
+
+pipe = pipeline(
+    task="text-generation",
+    model="google/vaultgemma-1b",
+    dtype="auto",
+    device_map="auto",
+)
+
+text = "Tell me an unknown interesting biology fact about the brain."
+outputs = pipe(text, max_new_tokens=32)
+response = outputs[0]["generated_text"]
+print(response)
+```
+
+</hfoption>
+<hfoption id="AutoModel">
+
+```python
+# pip install accelerate
+from transformers import AutoTokenizer, AutoModelForCausalLM
+
+model_id = "google/vaultgemma-1b"
+tokenizer = AutoTokenizer.from_pretrained(model_id)
+model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto", dtype="auto")
+
+text = "Tell me an unknown interesting biology fact about the brain."
+input_ids = tokenizer(text, return_tensors="pt").to(model.device)
+
+outputs = model.generate(**input_ids, max_new_tokens=32)
+print(tokenizer.decode(outputs[0]))
+```
+
+</hfoption>
+<hfoption id="transformers CLI">
+
+```
+echo -e "Write me a poem about Machine Learning. Answer:" | transformers run --task text2text-generation --model google/vaultgemma-1b-pt --device 0
+```
+
+</hfoption>
+</hfoptions>
+
+## VaultGemmaConfig
+
+[[autodoc]] VaultGemmaConfig
+
+## VaultGemmaModel
+
+[[autodoc]] VaultGemmaModel
+    - forward
+
+## VaultGemmaForCausalLM
+
+[[autodoc]] VaultGemmaForCausalLM

src/transformers/models/__init__.py

@@ -338,6 +338,7 @@
     from .unispeech_sat import *
     from .univnet import *
     from .upernet import *
+    from .vaultgemma import *
     from .video_llava import *
     from .videomae import *
     from .vilt import *

src/transformers/models/auto/configuration_auto.py

@@ -400,6 +400,7 @@
         ("univnet", "UnivNetConfig"),
         ("upernet", "UperNetConfig"),
         ("van", "VanConfig"),
+        ("vaultgemma", "VaultGemmaConfig"),
         ("video_llava", "VideoLlavaConfig"),
         ("videomae", "VideoMAEConfig"),
         ("vilt", "ViltConfig"),
@@ -842,6 +843,7 @@
         ("univnet", "UnivNet"),
         ("upernet", "UPerNet"),
         ("van", "VAN"),
+        ("vaultgemma", "VaultGemma"),
         ("video_llava", "VideoLlava"),
         ("videomae", "VideoMAE"),
         ("vilt", "ViLT"),

src/transformers/models/auto/modeling_auto.py

@@ -384,6 +384,7 @@ class _BaseModelWithGenerate(PreTrainedModel, GenerationMixin):
         ("unispeech-sat", "UniSpeechSatModel"),
         ("univnet", "UnivNetModel"),
         ("van", "VanModel"),
+        ("vaultgemma", "VaultGemmaModel"),
         ("video_llava", "VideoLlavaModel"),
         ("videomae", "VideoMAEModel"),
         ("vilt", "ViltModel"),
@@ -732,6 +733,7 @@ class _BaseModelWithGenerate(PreTrainedModel, GenerationMixin):
         ("starcoder2", "Starcoder2ForCausalLM"),
         ("transfo-xl", "TransfoXLLMHeadModel"),
         ("trocr", "TrOCRForCausalLM"),
+        ("vaultgemma", "VaultGemmaForCausalLM"),
         ("whisper", "WhisperForCausalLM"),
         ("xglm", "XGLMForCausalLM"),
         ("xlm", "XLMWithLMHeadModel"),

src/transformers/models/vaultgemma/__init__.py

@@ -0,0 +1,29 @@
+# coding=utf-8
+# Copyright 2025 the HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from typing import TYPE_CHECKING
+
+from ...utils import _LazyModule
+from ...utils.import_utils import define_import_structure
+
+
+if TYPE_CHECKING:
+    from .configuration_vaultgemma import *
+    from .modeling_vaultgemma import *
+else:
+    import sys
+
+    _file = globals()["__file__"]
+    sys.modules[__name__] = _LazyModule(__name__, _file, define_import_structure(_file), module_spec=__spec__)

src/transformers/models/vaultgemma/configuration_vaultgemma.py

@@ -0,0 +1,182 @@
+#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
+#           This file was automatically generated from src/transformers/models/vaultgemma/modular_vaultgemma.py.
+#               Do NOT edit this file manually as any edits will be overwritten by the generation of
+#             the file from the modular. If any change should be done, please apply the change to the
+#                          modular_vaultgemma.py file directly. One of our CI enforces this.
+#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
+# coding=utf-8
+# Copyright 2025 the HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from ...configuration_utils import PretrainedConfig, layer_type_validation
+
+
+class VaultGemmaConfig(PretrainedConfig):
+    r"""
+    This is the configuration class to store the configuration of a [`VaultGemmaModel`]. It is used to instantiate an VaultGemma
+    model according to the specified arguments, defining the model architecture. Instantiating a configuration with the
+    defaults will yield a similar configuration to that of the VaultGemma-7B.
+    e.g. [google/vaultgemma-7b](https://huggingface.co/google/vaultgemma-7b)
+    Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
+    documentation from [`PretrainedConfig`] for more information.
+    Args:
+        vocab_size (`int`, *optional*, defaults to 256000):
+            Vocabulary size of the VaultGemma model. Defines the number of different tokens that can be represented by the
+            `inputs_ids` passed when calling [`VaultGemmaModel`]
+        hidden_size (`int`, *optional*, defaults to 2304):
+            Dimension of the hidden representations.
+        intermediate_size (`int`, *optional*, defaults to 9216):
+            Dimension of the MLP representations.
+        num_hidden_layers (`int`, *optional*, defaults to 26):
+            Number of hidden layers in the Transformer decoder.
+        num_attention_heads (`int`, *optional*, defaults to 8):
+            Number of attention heads for each attention layer in the Transformer decoder.
+        num_key_value_heads (`int`, *optional*, defaults to 4):
+            This is the number of key_value heads that should be used to implement Grouped Query Attention. If
+            `num_key_value_heads=num_attention_heads`, the model will use Multi Head Attention (MHA), if
+            `num_key_value_heads=1` the model will use Multi Query Attention (MQA) otherwise GQA is used. When
+            converting a multi-head checkpoint to a GQA checkpoint, each group key and value head should be constructed
+            by meanpooling all the original heads within that group. For more details, check out [this
+            paper](https://huggingface.co/papers/2305.13245). If it is not specified, will default to
+            `num_attention_heads`.
+        head_dim (`int`, *optional*, defaults to 256):
+            The attention head dimension.
+        hidden_activation (`str` or `function`, *optional*, defaults to `"gelu_pytorch_tanh"`):
+            The non-linear activation function (function or string) in the decoder. Will default to `"gelu_pytorch_tanh"`
+            if not specified. `"gelu_pytorch_tanh"` uses an approximation of the `"gelu"` activation function.
+        max_position_embeddings (`int`, *optional*, defaults to 8192):
+            The maximum sequence length that this model might ever be used with.
+        initializer_range (`float`, *optional*, defaults to 0.02):
+            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
+        rms_norm_eps (`float`, *optional*, defaults to 1e-06):
+            The epsilon used by the rms normalization layers.
+        use_cache (`bool`, *optional*, defaults to `True`):
+            Whether or not the model should return the last key/values attentions (not used by all models). Only
+            relevant if `config.is_decoder=True`.
+        pad_token_id (`int`, *optional*, defaults to 0):
+            Padding token id.
+        eos_token_id (`int`, *optional*, defaults to 1):
+            End of stream token id.
+        bos_token_id (`int`, *optional*, defaults to 2):
+            Beginning of stream token id.
+        tie_word_embeddings (`bool`, *optional*, defaults to `True`):
+            Whether to tie weight embeddings
+        rope_theta (`float`, *optional*, defaults to 10000.0):
+            The base period of the RoPE embeddings.
+        attention_bias (`bool`, defaults to `False`, *optional*, defaults to `False`):
+            Whether to use a bias in the query, key, value and output projection layers during self-attention.
+        attention_dropout (`float`, *optional*, defaults to 0.0):
+            The dropout ratio for the attention probabilities.
+        query_pre_attn_scalar (`float`, *optional*, defaults to 256):
+            scaling factor used on the attention scores
+        sliding_window (`int`, *optional*, defaults to 4096):
+            in VaultGemma, every other layer uses sliding window attention. This is the size of the sliding window.
+        layer_types (`list`, *optional*):
+            Attention pattern for each layer.
+        final_logit_softcapping (`float`, *optional*, defaults to 30.0):
+            scaling factor when applying tanh softcapping on the logits.
+        attn_logit_softcapping (`float`, *optional*, defaults to 50.0):
+            scaling factor when applying tanh softcapping on the attention scores.
+
+    ```python
+    >>> from transformers import VaultGemmaModel, VaultGemmaConfig
+    >>> # Initializing a VaultGemma vaultgemma-7b style configuration
+    >>> configuration = VaultGemmaConfig()
+    >>> # Initializing a model from the vaultgemma-7b style configuration
+    >>> model = VaultGemmaModel(configuration)
+    >>> # Accessing the model configuration
+    >>> configuration = model.config
+    ```"""
+
+    model_type = "vaultgemma"
+    keys_to_ignore_at_inference = ["past_key_values"]
+    base_model_tp_plan = {
+        "layers.*.self_attn.q_proj": "colwise",
+        "layers.*.self_attn.k_proj": "colwise",
+        "layers.*.self_attn.v_proj": "colwise",
+        "layers.*.self_attn.o_proj": "rowwise",
+        "layers.*.mlp.gate_proj": "colwise",
+        "layers.*.mlp.up_proj": "colwise",
+        "layers.*.mlp.down_proj": "rowwise",
+    }
+    base_model_pp_plan = {
+        "embed_tokens": (["input_ids"], ["inputs_embeds"]),
+        "layers": (["hidden_states", "attention_mask"], ["hidden_states"]),
+        "norm": (["hidden_states"], ["hidden_states"]),
+    }
+
+    def __init__(
+        self,
+        vocab_size=256000,
+        hidden_size=2304,
+        intermediate_size=9216,
+        num_hidden_layers=26,
+        num_attention_heads=8,
+        num_key_value_heads=4,
+        head_dim=256,
+        hidden_activation="gelu_pytorch_tanh",
+        max_position_embeddings=8192,
+        initializer_range=0.02,
+        rms_norm_eps=1e-6,
+        use_cache=True,
+        pad_token_id=0,
+        eos_token_id=1,
+        bos_token_id=2,
+        tie_word_embeddings=True,
+        rope_theta=10000.0,
+        attention_bias=False,
+        attention_dropout=0.0,
+        query_pre_attn_scalar=256,
+        sliding_window=4096,
+        layer_types=None,
+        final_logit_softcapping=30.0,
+        attn_logit_softcapping=50.0,
+        **kwargs,
+    ):
+        super().__init__(
+            pad_token_id=pad_token_id,
+            bos_token_id=bos_token_id,
+            eos_token_id=eos_token_id,
+            tie_word_embeddings=tie_word_embeddings,
+            **kwargs,
+        )
+        self.vocab_size = vocab_size
+        self.max_position_embeddings = max_position_embeddings
+        self.hidden_size = hidden_size
+        self.intermediate_size = intermediate_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        self.head_dim = head_dim
+        self.num_key_value_heads = num_key_value_heads
+        self.initializer_range = initializer_range
+        self.rms_norm_eps = rms_norm_eps
+        self.use_cache = use_cache
+        self.rope_theta = rope_theta
+        self.attention_bias = attention_bias
+        self.attention_dropout = attention_dropout
+        self.hidden_activation = hidden_activation
+        self.query_pre_attn_scalar = query_pre_attn_scalar
+        self.sliding_window = sliding_window
+        self.final_logit_softcapping = final_logit_softcapping
+        self.attn_logit_softcapping = attn_logit_softcapping
+        self.layer_types = layer_types
+
+        if self.layer_types is None:
+            self.layer_types = [
+                "sliding_attention" if bool((i + 1) % 2) else "full_attention" for i in range(self.num_hidden_layers)
+            ]
+        layer_type_validation(self.layer_types)
+
+
+__all__ = ["VaultGemmaConfig"]