# Copyright 2025 The EasyDeL Author @erfanzar (Erfan Zare Chavoshi).
#
# 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
#
# https://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.
import typing
from eformer.common_types import ColumnWise, Replicated, RowWise
from easydel.infra.base_module import EasyDeLBaseConfig
from easydel.infra.etils import EasyDeLGradientCheckPointers
from easydel.infra.factory import register_config
from easydel.infra.utils import AttnMaskDetail, AttnMaskType
from ..siglip.configuration_siglip import SiglipVisionConfig
[docs]@register_config("gemma3_text")
class Gemma3TextConfig(EasyDeLBaseConfig):
"""
Configuration objects inherit from [`EasyDeLBaseConfig`] and can be used to control the model outputs. Read
the documentation from [`EasyDeLBaseConfig`] for more information.
Args:
vocab_size (`int`, *optional*, defaults to 262208):
Vocabulary size of the Gemma3Text model. Defines the number of different tokens that can be represented
by the `inputs_ids` passed when calling [`Gemma3TextModel`]
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 checkout [this
paper](https://arxiv.org/pdf/2305.13245.pdf). 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 131072):
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 1000000.0):
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 Gemma3Text, every other layer uses sliding window
attention. This is the size of the sliding window.
final_logit_softcapping (`float`, *optional*):
Scaling factor when applying tanh softcapping on the logits.
attn_logit_softcapping (`float`, *optional*):
Scaling factor when applying tanh softcapping on the attention scores.
cache_implementation (`str`, *optional*, defaults to `"hybrid"`): the cache type to be used with `generate`.
rope_scaling (`Dict`, *optional*):
Dictionary containing the scaling configuration for the RoPE embeddings used in gloabl attention.
Expected contents:
`rope_type` (`str`):
The sub-variant of RoPE to use. Can be one of ['default', 'linear', 'dynamic', 'yarn', 'longrope',
'llama3'], with 'default' being the original RoPE implementation.
`factor` (`float`, *optional*):
Used with all rope types except 'default'. The scaling factor to apply to the RoPE embeddings. In
most scaling types, a `factor` of x will enable the model to handle sequences of length x *
original maximum pre-trained length.
`original_max_position_embeddings` (`int`, *optional*):
Used with 'dynamic', 'longrope' and 'llama3'. The original max position embeddings used during
pretraining.
`attention_factor` (`float`, *optional*):
Used with 'yarn' and 'longrope'. The scaling factor to be applied on the attention
computation. If unspecified, it defaults to value recommended by the implementation, using the
`factor` field to infer the suggested value.
`beta_fast` (`float`, *optional*):
Only used with 'yarn'. Parameter to set the boundary for extrapolation (only) in the linear
ramp function. If unspecified, it defaults to 32.
`beta_slow` (`float`, *optional*):
Only used with 'yarn'. Parameter to set the boundary for interpolation (only) in the linear
ramp function. If unspecified, it defaults to 1.
`short_factor` (`List[float]`, *optional*):
Only used with 'longrope'. The scaling factor to be applied to short contexts (<
`original_max_position_embeddings`). Must be a list of numbers with the same length as the hidden
size divided by the number of attention heads divided by 2
`long_factor` (`List[float]`, *optional*):
Only used with 'longrope'. The scaling factor to be applied to long contexts (<
`original_max_position_embeddings`). Must be a list of numbers with the same length as the hidden
size divided by the number of attention heads divided by 2
`low_freq_factor` (`float`, *optional*):
Only used with 'llama3'. Scaling factor applied to low frequency components of the RoPE
`high_freq_factor` (`float`, *optional*):
Only used with 'llama3'. Scaling factor applied to high frequency components of the RoPE
rope_local_base_freq (float, *optional*, defaults to 10000.0):
The base period of the RoPE embeddings for local attention.
sliding_window_pattern (`int`, *optional*, defaults to 6):
Pattern for the sliding window attention.
gradient_checkpointing (`str`, *optional*, defaults to `"nothing_saveable"`):
The gradient checkpointing configuration.
bits (`int`, *optional*):
The number of bits to quantize the model to.
scan_layers (`bool`, *optional*, defaults to `False`):
Whether to use the scan implementation of the layers.
"""
model_type: str = "gemma3_text"
def __init__(
self,
vocab_size=262_208,
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=131_072,
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=1_000_000.0,
attention_bias=False,
attention_dropout=0.0,
query_pre_attn_scalar=256,
sliding_window=4096,
final_logit_softcapping=None,
attn_logit_softcapping=None,
cache_implementation="hybrid",
rope_scaling=None,
rope_local_base_freq=10_000.0,
sliding_window_pattern=6,
layer_types: list[str] | None = None,
gradient_checkpointing: EasyDeLGradientCheckPointers = EasyDeLGradientCheckPointers.NONE,
bits: int | None = None,
scan_layers: bool = False,
use_bidirectional_attention: bool = False,
**kwargs,
):
"""The __init__ function is called when the class is instantiated.
It sets up the attributes of an object, which are sometimes called fields or properties.
The __init__ function can accept arguments, but self must be the first one.
"""
self.gradient_checkpointing = gradient_checkpointing
self.bits = bits
self.scan_layers = scan_layers
super().__init__(
bos_token_id=bos_token_id,
eos_token_id=eos_token_id,
pad_token_id=pad_token_id,
tie_word_embeddings=tie_word_embeddings,
bits=bits,
**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.cache_implementation = cache_implementation
self.rope_local_base_freq = rope_local_base_freq
self.sliding_window_pattern = sliding_window_pattern
self.layer_types = layer_types
self.use_bidirectional_attention = use_bidirectional_attention
if use_bidirectional_attention:
self.sliding_window = (self.sliding_window // 2) + 1
if self.layer_types is None:
self.layer_types = [
"sliding_attention" if bool((i + 1) % self.sliding_window_pattern) else "full_attention"
for i in range(self.num_hidden_layers)
]
self.rope_scaling = rope_scaling
[docs] def get_partition_rules(self, *args, **kwargs):
"""
Get the partition rules for the model.
Returns:
`tp.Tuple[tp.Tuple[str, PartitionSpec]]`: The partition rules.
"""
pmag = self.partition_manager
return (
(r"embed_tokens/embedding", pmag.resolve(ColumnWise)),
(r"self_attn/(q_proj|k_proj|v_proj)/kernel", pmag.resolve(ColumnWise)),
(r"self_attn/o_proj/kernel", pmag.resolve(RowWise)),
(r"self_attn/.*proj/bias", pmag.resolve(Replicated)),
(r"self_attn/(q_norm|k_norm)/kernel", pmag.resolve(Replicated)),
(r"mlp/(gate_proj|up_proj)/kernel", pmag.resolve(ColumnWise)),
(r"mlp/down_proj/kernel", pmag.resolve(RowWise)),
(r"mlp/.*proj/bias", pmag.resolve(Replicated)),
(
r".*(input_layernorm|post_attention_layernorm|pre_feedforward_layernorm|post_feedforward_layernorm|norm)/kernel",
pmag.resolve(Replicated),
),
(r"vision_tower/.*", pmag.resolve(Replicated)),
(r"multi_modal_projector/mm_input_projection_weight", pmag.resolve(ColumnWise)),
(r"multi_modal_projector/mm_soft_emb_norm/kernel", pmag.resolve(Replicated)),
(r"lm_head/kernel", pmag.resolve(ColumnWise)),
(r"language_model/lm_head/kernel", pmag.resolve(ColumnWise)),
(r"score/kernel", pmag.resolve(RowWise)),
(r".*bias", pmag.resolve(Replicated)),
(r".*", pmag.resolve(Replicated)),
)
[docs] def get_mask_details(self) -> dict[int, AttnMaskDetail]:
"""Retrieve attention mask details for each layer in the model.
This method generates a dictionary mapping layer indices to their corresponding attention mask details.
If a sliding window is defined, each layer is assigned a sliding window attention mask with the specified size.
Returns:
dict[int, AttnMaskDetail]: A dictionary where keys are layer indices (int) and values are AttnMaskDetail
objects specifying the attention mask type and size for each layer.
Notes:
- If `self.sliding_window` is None, an empty dictionary is returned.
- The method iterates over `self.num_hidden_layers` to assign mask details for each layer.
- The attention mask type is set to `AttnMaskType.SLIDING` when a sliding window is defined.
"""
mapping = {}
if self.layer_types is not None:
for layer_idx in range(self.num_hidden_layers):
mapping[layer_idx] = AttnMaskDetail(
mask_type=AttnMaskType.from_hf(self.layer_types[layer_idx]),
size=self.sliding_window,
)
return mapping
[docs]@register_config("gemma3")
class Gemma3Config(EasyDeLBaseConfig):
r"""
Args:
text_config (`Union[Gemma3TextConfig, dict]`, *optional*):
The config object of the text backbone.
vision_config (`Union[AutoConfig, dict]`, *optional*):
Custom vision config or dict.
mm_tokens_per_image (`int`, *optional*, defaults to 256):
The number of tokens per image embedding.
boi_token_index (`int`, *optional*, defaults to 255999):
The begin-of-image token index to wrap the image prompt.
eoi_token_index (`int`, *optional*, defaults to 256000):
The end-of-image token index to wrap the image prompt.
image_token_id (`int`, *optional*, defaults to 262144):
The image token index to encode the image prompt.
initializer_range (`float`, *optional*, defaults to 0.02):
The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
Example:
```python
>>> from transformers import Gemma3ForConditionalGeneration, Gemma3Config, SiglipVisionConfig, Gemma3TextConfig
>>> # Initializing a Siglip-like vision config
>>> vision_config = SiglipVisionConfig()
>>> # Initializing a Gemma3 Text config
>>> text_config = Gemma3TextConfig()
>>> # Initializing a Gemma3 gemma-3-4b style configuration
>>> configuration = Gemma3Config(vision_config, text_config)
>>> # Initializing a model from the gemma-3-4b style configuration
>>> model = Gemma3TextConfig(configuration)
>>> # Accessing the model configuration
>>> configuration = model.config
```"""
model_type = "gemma3"
sub_configs: typing.ClassVar = {"text_config": Gemma3TextConfig, "vision_config": SiglipVisionConfig}
def __init__(
self,
text_config: Gemma3TextConfig | None = None,
vision_config: SiglipVisionConfig | None = None,
mm_tokens_per_image: int = 256,
boi_token_index: int = 255_999,
eoi_token_index: int = 256_000,
image_token_id: int = 262_144,
initializer_range: float = 0.02,
**kwargs,
):
"""Initialize a Gemma3Config instance.
Args:
text_config (Optional[Gemma3TextConfig], optional): The configuration for the text model component.
If None is provided, a default Gemma3TextConfig will be used. Defaults to None.
vision_config (Optional[SiglipVisionConfig], optional): The configuration for the vision model component.
If None is provided, a default SiglipVisionConfig will be used. Defaults to None.
mm_tokens_per_image (int, optional): Number of tokens per image embedding. Defaults to 256.
boi_token_index (int, optional): Begin-of-image token index to wrap the image prompt. Defaults to 255_999.
eoi_token_index (int, optional): End-of-image token index to wrap the image prompt. Defaults to 256_000.
image_token_id (int, optional): Image token index to encode the image prompt. Defaults to 262_144.
initializer_range (float, optional): Standard deviation for weight initialization. Defaults to 0.02.
**kwargs: Additional keyword arguments passed to the parent class constructor.
"""
if text_config is None:
text_config = Gemma3TextConfig()
elif isinstance(text_config, dict):
text_config = Gemma3TextConfig(**text_config)
if isinstance(vision_config, dict):
vision_config = SiglipVisionConfig(**vision_config)
else:
vision_config = SiglipVisionConfig()
self.text_config = text_config
self.vision_config = vision_config
self.mm_tokens_per_image = mm_tokens_per_image
self.boi_token_index = boi_token_index
self.eoi_token_index = eoi_token_index
self.image_token_id = image_token_id
self.initializer_range = initializer_range
super().__init__(**kwargs)
[docs] def get_partition_rules(self, *args, **kwargs):
"""Get the partition rules for the model.
Returns:
Tuple[Tuple[str, PartitionSpec]]: A tuple of tuples, where each inner tuple contains a regex pattern
matching parameter names and the corresponding PartitionSpec for sharding those parameters across devices.
"""
text_partitions = self.text_config.get_partition_rules(*args, **kwargs)
vision_partitions = self.vision_config.get_partition_rules(*args, **kwargs)
return text_partitions + vision_partitions