# 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 as tp
from eformer.common_types import (
EMPTY,
MODE_TRAIN,
TP,
ColumnWise,
DynamicShardingAxes,
Replicated,
RowWise,
)
from easydel.infra.base_module import EasyDeLBaseConfig
from easydel.infra.etils import EasyDeLGradientCheckPointers
from easydel.infra.factory import register_config
from easydel.layers.moe.utils import get_moe_partition_spec
from easydel.layers.rotary_embedding import RopeConfig
[docs]class ExpertTensorParallel(DynamicShardingAxes):
"""Expert Tensor Parallelism (EPxTP) sharding axes."""
axes: tp.ClassVar = [TP, EMPTY, EMPTY]
mode: tp.ClassVar = MODE_TRAIN
[docs]@register_config("deepseek_v2")
class DeepseekV2Config(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 102400):
Vocabulary size of the DeepseekV2 model. Defines the number of different tokens that can be represented by
the `inputs_ids` passed to the forward method.
hidden_size (`int`, *optional*, defaults to 4096):
Dimensionality of the encoder layers and the pooler layer.
intermediate_size (`int`, *optional*, defaults to 11008):
Dimensionality of the "intermediate" (i.e., feed-forward) layer in the Transformer encoder.
moe_intermediate_size (`int`, *optional*, defaults to 1407):
Dimensionality of the "intermediate" (i.e., feed-forward) layer in the MoE layer.
num_hidden_layers (`int`, *optional*, defaults to 30):
Number of hidden layers in the Transformer encoder.
num_attention_heads (`int`, *optional*, defaults to 32):
Number of attention heads for each attention layer in the Transformer encoder.
num_key_value_heads (`int`, *optional*, defaults to 32):
Number of key and value heads for each attention layer in the Transformer encoder.
n_shared_experts (`int`, *optional*):
Number of shared experts.
n_routed_experts (`int`, *optional*):
Number of routed experts.
ep_size (`int`, *optional*, defaults to 1):
Expert parallel size.
routed_scaling_factor (`float`, *optional*, defaults to 1.0):
Routed scaling factor.
kv_lora_rank (`int`, *optional*, defaults to 512):
KV LoRA rank.
q_lora_rank (`int`, *optional*, defaults to 1536):
Q LoRA rank.
qk_rope_head_dim (`int`, *optional*, defaults to 64):
QK rope head dimension.
v_head_dim (`int`, *optional*, defaults to 128):
V head dimension.
qk_nope_head_dim (`int`, *optional*, defaults to 128):
QK nope head dimension.
topk_method (`str`, *optional*, defaults to `"gready"`):
Top-k method.
n_group (`int`, *optional*):
Number of groups.
topk_group (`int`, *optional*):
Top-k group.
num_experts_per_tok (`int`, *optional*):
Number of experts per token.
moe_layer_freq (`int`, *optional*, defaults to 1):
MoE layer frequency.
first_k_dense_replace (`int`, *optional*, defaults to 0):
First k dense replace.
norm_topk_prob (`bool`, *optional*, defaults to `False`):
Whether to normalize top-k probabilities.
scoring_func (`str`, *optional*, defaults to `"softmax"`):
Scoring function.
aux_loss_alpha (`float`, *optional*, defaults to 0.001):
Auxiliary loss alpha.
seq_aux (`bool`, *optional*, defaults to `True`):
Whether to use sequence auxiliary loss.
hidden_act (`str` or `function`, *optional*, defaults to `"silu"`):
The non-linear activation function (function or string) to use in the encoder and pooler. If string,
`"gelu"`, `"relu"`, `"swish"` and `"gelu_new"` are supported.
max_position_embeddings (`int`, *optional*, defaults to 2048):
The maximum sequence length that this model might ever be used with. Typically set this to something large
just in case (e.g., 2048 or 4096).
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-6):
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*):
The index of the padding token in the vocabulary.
bos_token_id (`int`, *optional*, defaults to 100000):
The index of the beginning of sequence token in the vocabulary.
eos_token_id (`int`, *optional*, defaults to 100001):
The index of the end of sequence token in the vocabulary.
pretraining_tp (`int`, *optional*, defaults to 1):
Pretraining TP.
tie_word_embeddings (`bool`, *optional*, defaults to `False`):
Whether to tie the weights of the input embeddings and the output embeddings.
rope_theta (`float`, *optional*, defaults to 10000.0):
The theta value to use for rotary position embeddings.
attention_bias (`bool`, *optional*, defaults to `False`):
Whether to use attention bias.
attention_dropout (`float`, *optional*, defaults to 0.0):
The dropout ratio for the attention probabilities.
gradient_checkpointing (`str`, *optional*, defaults to `"nothing_saveable"`):
The gradient checkpointing configuration.
use_scan_mlp (`bool`, *optional*, defaults to `False`):
Whether to use scan for MLP.
scan_mlp_chunk_size (`int`, *optional*, defaults to 1024):
The chunk size for scan MLP.
bits (`int`, *optional*):
The number of bits to quantize the model to.
rope_scaling (`tp.Dict[str, tp.Union[str, float]]`, *optional*):
The rope scaling configuration.
"""
model_type: str = "deepseek_v2"
def __init__(
self,
vocab_size=102400,
hidden_size=4096,
intermediate_size=11008,
moe_intermediate_size=1407,
num_hidden_layers=30,
num_attention_heads=32,
num_key_value_heads=32,
n_shared_experts=None,
n_routed_experts=None,
ep_size=1,
routed_scaling_factor=1.0,
kv_lora_rank=512,
q_lora_rank=1536,
qk_rope_head_dim=64,
v_head_dim=128,
qk_nope_head_dim=128,
topk_method="gready",
n_group=None,
topk_group=None,
num_experts_per_tok=None,
moe_layer_freq=1,
first_k_dense_replace=0,
norm_topk_prob=False,
scoring_func="softmax",
aux_loss_alpha=0.001,
seq_aux=True,
hidden_act="silu",
max_position_embeddings=2048,
initializer_range=0.02,
rms_norm_eps=1e-6,
use_cache=True,
pad_token_id=None,
bos_token_id=100000,
eos_token_id=100001,
pretraining_tp=1,
tie_word_embeddings=False,
rope_theta=10000.0,
attention_bias=False,
attention_dropout=0.0,
gradient_checkpointing: EasyDeLGradientCheckPointers = EasyDeLGradientCheckPointers.NONE,
use_scan_mlp: bool = False,
scan_mlp_chunk_size: int = 1024,
bits: int | None = None,
rope_scaling: dict[str, str | float] | None = None,
**kwargs,
):
"""Initialize a new DeepseekV2Config instance.
Args:
vocab_size (int, optional): Size of the vocabulary. Defaults to 102400.
hidden_size (int, optional): Dimensionality of the embeddings and hidden states. Defaults to 4096.
intermediate_size (int, optional): Dimensionality of the MLP layer. Defaults to 11008.
moe_intermediate_size (int, optional): Dimensionality of the MoE intermediate layer. Defaults to 1407.
num_hidden_layers (int, optional): Number of hidden layers in the model. Defaults to 30.
num_attention_heads (int, optional): Number of attention heads. Defaults to 32.
num_key_value_heads (int, optional): Number of key/value heads (for GQA). Defaults to 32.
n_shared_experts (int, optional): Number of shared MoE experts. Defaults to None.
n_routed_experts (int, optional): Number of routed MoE experts. Defaults to None.
ep_size (int, optional): Expert parallelism size. Defaults to 1.
routed_scaling_factor (float, optional): Scaling factor for routed experts. Defaults to 1.0.
kv_lora_rank (int, optional): Rank for KV LoRA. Defaults to 512.
q_lora_rank (int, optional): Rank for Q LoRA. Defaults to 1536.
qk_rope_head_dim (int, optional): Head dimension for QK with RoPE. Defaults to 64.
v_head_dim (int, optional): Head dimension for V. Defaults to 128.
qk_nope_head_dim (int, optional): Head dimension for QK without RoPE. Defaults to 128.
topk_method (str, optional): Method for top-k expert selection. Defaults to "gready".
n_group (int, optional): Number of expert groups. Defaults to None.
topk_group (int, optional): Top-k groups. Defaults to None.
num_experts_per_tok (int, optional): Number of experts per token. Defaults to None.
moe_layer_freq (int, optional): Frequency of MoE layers. Defaults to 1.
first_k_dense_replace (int, optional): First k dense layers to replace. Defaults to 0.
norm_topk_prob (bool, optional): Whether to normalize top-k probabilities. Defaults to False.
scoring_func (str, optional): Scoring function for expert selection. Defaults to "softmax".
aux_loss_alpha (float, optional): Weight for auxiliary loss. Defaults to 0.001.
seq_aux (bool, optional): Whether to use sequence auxiliary loss. Defaults to True.
hidden_act (str, optional): Activation function. Defaults to "silu".
max_position_embeddings (int, optional): Maximum sequence length. Defaults to 2048.
initializer_range (float, optional): Range for weight initialization. Defaults to 0.02.
rms_norm_eps (float, optional): Epsilon for RMS normalization. Defaults to 1e-6.
use_cache (bool, optional): Whether to use KV cache for generation. Defaults to True.
pad_token_id (int, optional): ID for padding token. Defaults to None.
bos_token_id (int, optional): ID for beginning of sequence token. Defaults to 100000.
eos_token_id (int, optional): ID for end of sequence token. Defaults to 100001.
pretraining_tp (int, optional): Tensor parallelism size during pretraining. Defaults to 1.
tie_word_embeddings (bool, optional): Whether to tie input/output embeddings. Defaults to False.
rope_theta (float, optional): Base value for RoPE. Defaults to 10000.0.
attention_bias (bool, optional): Whether to use bias in attention. Defaults to False.
attention_dropout (float, optional): Dropout rate for attention. Defaults to 0.0.
gradient_checkpointing (EasyDeLGradientCheckPointers, optional):
Checkpointing strategy. Defaults to EasyDeLGradientCheckPointers.NONE.
use_scan_mlp (bool, optional): Whether to use scan for MLP computation. Defaults to False.
scan_mlp_chunk_size (int, optional): Chunk size for scan MLP. Defaults to 1024.
bits (int, optional): Quantization bits. Defaults to None.
rope_scaling (Dict[str, Union[str, float]], optional): RoPE scaling configuration. Defaults to None.
**kwargs: Additional arguments.
"""
self.vocab_size = vocab_size
self.max_position_embeddings = max_position_embeddings
self.hidden_size = hidden_size
self.intermediate_size = intermediate_size
self.moe_intermediate_size = moe_intermediate_size
self.num_hidden_layers = num_hidden_layers
self.num_attention_heads = num_attention_heads
self.n_shared_experts = n_shared_experts
self.n_routed_experts = n_routed_experts
self.ep_size = ep_size
self.routed_scaling_factor = routed_scaling_factor
self.kv_lora_rank = kv_lora_rank
self.q_lora_rank = q_lora_rank
self.qk_rope_head_dim = qk_rope_head_dim
self.v_head_dim = v_head_dim
self.qk_nope_head_dim = qk_nope_head_dim
self.topk_method = topk_method
self.n_group = n_group
self.topk_group = topk_group
self.num_experts_per_tok = num_experts_per_tok
self.moe_layer_freq = moe_layer_freq
self.first_k_dense_replace = first_k_dense_replace
self.norm_topk_prob = norm_topk_prob
self.scoring_func = scoring_func
self.aux_loss_alpha = aux_loss_alpha
self.seq_aux = seq_aux
# for backward compatibility
if num_key_value_heads is None:
num_key_value_heads = num_attention_heads
self.num_key_value_heads = num_key_value_heads
self.hidden_act = hidden_act
self.initializer_range = initializer_range
self.rms_norm_eps = rms_norm_eps
self.pretraining_tp = pretraining_tp
self.use_cache = use_cache
self.rope_theta = rope_theta
self.rope_scaling = rope_scaling
self.attention_bias = attention_bias
self.attention_dropout = attention_dropout
self.gradient_checkpointing = gradient_checkpointing
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,
use_scan_mlp=use_scan_mlp,
scan_mlp_chunk_size=scan_mlp_chunk_size,
bits=bits,
**kwargs,
)
[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 # Handles resolving strategies
return (
(r"embed_tokens/embedding", pmag.resolve(ColumnWise)),
(r"self_attn/q_proj/kernel", pmag.resolve(ColumnWise)),
(r"self_attn/q_a_proj/kernel", pmag.resolve(ColumnWise)),
(r"self_attn/q_b_proj/kernel", pmag.resolve(ColumnWise)),
(r"self_attn/kv_a_proj_with_mqa/kernel", pmag.resolve(ColumnWise)),
(r"self_attn/kv_b_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_a_layernorm|kv_a_layernorm)/kernel", pmag.resolve(Replicated)),
(r"mlp/(gate_proj|up_proj)/kernel", pmag.resolve(ColumnWise)),
(r"mlp/down_proj/kernel", pmag.resolve(RowWise)),
(r"mlp/gate/kernel", pmag.resolve(Replicated if self.use_expert_tensor_mode else ColumnWise)),
(
r"mlp/experts/(gate_proj|up_proj)/kernel",
get_moe_partition_spec(
partition_manager=self.partition_manager,
direction="column",
tensors_are_expert=self.use_expert_tensor_mode,
is_bias=False,
fsdp_is_ep_bound=self.fsdp_is_ep_bound,
sp_is_ep_bound=self.sp_is_ep_bound,
module_view=True,
),
),
(
r"mlp/experts/down_proj/kernel",
get_moe_partition_spec(
partition_manager=self.partition_manager,
direction="row",
tensors_are_expert=self.use_expert_tensor_mode,
is_bias=False,
fsdp_is_ep_bound=self.fsdp_is_ep_bound,
sp_is_ep_bound=self.sp_is_ep_bound,
module_view=True,
),
),
(r".*(input_layernorm|post_attention_layernorm|norm)/kernel", pmag.resolve(Replicated)),
(r"lm_head/kernel", pmag.resolve(ColumnWise)),
(r".*bias", pmag.resolve(Replicated)),
(r".*", pmag.resolve(Replicated)),
)
@property
def granted_freq_max_position_embedding(self) -> int:
"""Returns the maximum position embedding size for frequency-based position embeddings.
Returns:
int: The maximum position embedding size, falling back to max_position_embeddings if not explicitly set.
"""
return getattr(
self,
"freq_max_position_embeddings",
self.max_position_embeddings,
)
@property
def granted_mask_max_position_embedding(self) -> int:
"""Returns the maximum position embedding size for mask-based position embeddings.
Returns:
int: The maximum position embedding size, falling back to max_position_embeddings if not explicitly set.
"""
return getattr(
self,
"mask_max_position_embeddings",
self.max_position_embeddings,
)
def _get_rope_config(self) -> RopeConfig:
"""Get RoPE configuration from the instance attributes."""
if not hasattr(self, "rope_scaling") or self.rope_scaling is None:
config = RopeConfig.from_dict(
dict(
rope_type="yarn",
base=10000,
scaling_factor=1.0,
original_max_position_embeddings=4096,
beta_fast=32,
beta_slow=1,
mscale=1,
mscale_all_dim=0,
)
)
else:
config = RopeConfig.from_dict(self.rope_scaling)
if config.original_max_position_embeddings is None:
config.original_max_position_embeddings = getattr(self, "original_max_position_embeddings", None)
return config