# 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.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("xerxes2")
class Xerxes2Config(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 256128):
Vocabulary size of the xerxes 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 16384):
Dimensionality of the "intermediate" (i.e., feed-forward) layer in the Transformer encoder.
num_hidden_layers (`int`, *optional*, defaults to 32):
Number of hidden layers in the Transformer encoder.
num_attention_heads (`int`, *optional*, defaults to 16):
Number of attention heads for each attention layer in the Transformer encoder.
num_key_value_heads (`int`, *optional*, defaults to 16):
Number of key and value heads for each attention layer in the Transformer encoder.
head_dim (`int`, *optional*, defaults to 256):
Dimensionality of the attention head.
max_position_embeddings (`int`, *optional*, defaults to 6144):
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*, defaults to 0):
The index of the padding token in the vocabulary.
eos_token_id (`int`, *optional*, defaults to 1):
The index of the end of sequence token in the vocabulary.
bos_token_id (`int`, *optional*, defaults to 2):
The index of the beginning of sequence token in the vocabulary.
tie_word_embeddings (`bool`, *optional*, defaults to `True`):
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.
softmax_scale (`float`, *optional*, defaults to `14.9666295471`):
softmax scale for attention module.
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.
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 = "xerxes2"
def __init__(
self,
vocab_size: int = 256128,
hidden_size: int = 4096,
intermediate_size: int = 16384,
moe_intermediate_size: int = 8192,
decoder_sparse_step: int = 1,
num_experts_per_tok: int = 8,
num_experts: int = 128,
norm_topk_prob: int = False,
output_router_logits: int = False,
router_aux_loss_coef: int = 0.001,
num_hidden_layers: int = 32,
num_attention_heads: int = 32,
max_position_embeddings: int = 16384,
initializer_range: float = 0.02,
rms_norm_eps: float = 1e-6,
use_cache: bool = True,
pad_token_id: int = 0,
eos_token_id: int = 1,
bos_token_id: int = 2,
tie_word_embeddings: bool = False,
rope_theta: float = 10000.0,
bits: int | None = None,
scan_layers: bool = False,
q_lora_dim: int | None = 1536,
kv_lora_dim: int = 512,
qk_rope_head_dim: int = 64,
qk_nope_head_dim: int = 128,
vhead_dim: int = 128,
mlp_only_layers: list[int] | None = None,
hidden_act: str | None = None,
rope_scaling: dict | None = None,
**kwargs,
):
self.bits = bits
self.scan_layers = scan_layers
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.decoder_sparse_step = decoder_sparse_step
self.num_experts = num_experts
self.num_experts_per_tok = num_experts_per_tok
self.norm_topk_prob = norm_topk_prob
self.output_router_logits = output_router_logits
self.router_aux_loss_coef = router_aux_loss_coef
self.num_hidden_layers = num_hidden_layers
self.num_attention_heads = num_attention_heads
self.initializer_range = initializer_range
self.rms_norm_eps = rms_norm_eps
self.use_cache = use_cache
self.rope_theta = rope_theta
self.q_lora_dim = q_lora_dim
self.kv_lora_dim = kv_lora_dim
self.qk_rope_head_dim = qk_rope_head_dim
self.qk_nope_head_dim = qk_nope_head_dim
self.vhead_dim = vhead_dim
self.mlp_only_layers = [] if mlp_only_layers is None else mlp_only_layers
self.hidden_act = hidden_act if hidden_act is not None else "silu"
self.rope_scaling = rope_scaling
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,
)
[docs] def get_partition_rules(self, *args, **kwargs):
"""
Get the partition rules for the model.
Args:
fully_sharded_data_parallel (`bool`, *optional*, defaults to `True`):
Whether to use fully sharded data parallelism.
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/qa_proj/kernel", pmag.resolve(ColumnWise)),
(r"self_attn/qb_proj/kernel", pmag.resolve(ColumnWise)),
(r"self_attn/qc_proj/kernel", pmag.resolve(ColumnWise)),
(r"self_attn/kv_mqa_proj/kernel", pmag.resolve(ColumnWise)),
(r"self_attn/kvi_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/(qa_norm|kv_norm)/scale", pmag.resolve(Replicated)),
(r"self_attn/(qa_norm|kv_norm)/bias", pmag.resolve(Replicated)),
# Standard MLP rules
(r"mlp/gate_up_proj/kernel", pmag.resolve(ColumnWise)),
(r"mlp/down_proj/kernel", pmag.resolve(RowWise)),
(r"mlp/.*proj/bias", pmag.resolve(Replicated)),
# MoE specific rules
(r"mlp/gate/kernel", pmag.resolve(Replicated if self.use_expert_tensor_mode else ColumnWise)),
(r"mlp/gate/bias", pmag.resolve(Replicated)),
(
r"mlp/experts/gate_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/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"mlp/experts/.*/bias", pmag.resolve(Replicated)),
# Layer norms
(
r".*/(input_layernorm|post_attention_layernorm|pre_feedforward_layernorm|post_feedforward_layernorm|norm)/kernel",
pmag.resolve(Replicated),
),
(r"lm_head/kernel", pmag.resolve(ColumnWise)),
(r"score/kernel", pmag.resolve(RowWise)),
(r".*bias", pmag.resolve(Replicated)),
(r".*", pmag.resolve(Replicated)),
)
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