You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
ColossalAI/colossalai/shardformer/policies/deepseek.py

382 lines
16 KiB

from functools import partial
from typing import Callable, Dict, List, Union
import torch.nn as nn
from torch import Tensor
from torch.nn import Module
from transformers.utils import is_flash_attn_greater_or_equal_2_10
from colossalai.shardformer.layer import FusedRMSNorm, Linear1D_Col
from colossalai.shardformer.layer.embedding import PaddingEmbedding, VocabParallelEmbedding1D
from colossalai.shardformer.layer.linear import Linear1D_Row
from colossalai.shardformer.modeling.deepseek import (
DeepseekMoEGate_Col,
DeepseekPipelineForwards,
EPDeepseekMoE,
get_deepseek_flash_attention_forward,
get_deepseek_flash_attention_model_forward,
)
from colossalai.shardformer.policies.base_policy import ModulePolicyDescription, Policy, SubModuleReplacementDescription
__all__ = ["DeepseekPolicy", "DeepseekForCausalLMPolicy"]
class DeepseekPolicy(Policy):
def config_sanity_check(self):
pass
def preprocess(self):
self.tie_weight = self.tie_weight_check()
self.origin_attn_implement = self.model.config._attn_implementation
"""
Because transformers library's bug for AutoModel/AutoConfig, who pop “attn_implement” twice from modeling_utils.py and configuration_utils.py.
This bug causes attn_cls to be set to sdpa. Here we assign it to "flash_attention_2".
"""
# self.origin_attn_implement = "flash_attention_2"
if self.shard_config.enable_tensor_parallelism:
# Resize embedding
vocab_size = self.model.config.vocab_size
world_size = self.shard_config.tensor_parallel_size
if vocab_size % world_size != 0:
new_vocab_size = vocab_size + world_size - vocab_size % world_size
self.model.resize_token_embeddings(new_vocab_size)
return self.model
def module_policy(self) -> Dict[Union[str, nn.Module], ModulePolicyDescription]:
ATTN_IMPLEMENTATION = {
"eager": "DeepseekAttention",
"flash_attention_2": "DeepseekFlashAttention2",
"sdpa": "DeepseekSdpaAttention",
}
policy = {}
attn_cls = ATTN_IMPLEMENTATION[self.origin_attn_implement]
sp_mode = self.shard_config.sequence_parallelism_mode or None
sp_size = self.shard_config.sequence_parallel_size or None
sp_group = self.shard_config.sequence_parallel_process_group or None
sp_partial_derived = sp_mode in ["split_gather", "ring"]
tp_size = self.shard_config.tensor_parallel_size
# modified for both SP and TP
num_q_heads = self.model.config.num_attention_heads
num_kv_heads = getattr(self.model.config, "num_key_value_heads", None)
if sp_mode == "all_to_all":
num_q_heads //= sp_size
decoder_attribute_replacement = {
"num_heads": num_q_heads,
}
if getattr(self.model.config, "num_key_value_heads", False):
num_kv_heads //= sp_size
decoder_attribute_replacement["num_key_value_heads"] = num_kv_heads
policy[attn_cls] = ModulePolicyDescription(
attribute_replacement=decoder_attribute_replacement,
)
if self.shard_config.enable_sequence_parallelism:
if self.pipeline_stage_manager is not None:
# NOTE: we are replacing model forward for both sequence parallelism and pipeline parallelism
# if both are enabled, one of them will be ignored
raise NotImplementedError("Sequence parallelism is not supported with pipeline parallelism.")
self.append_or_create_method_replacement(
description={
"forward": get_deepseek_flash_attention_forward(self.shard_config, sp_mode, sp_size, sp_group),
},
policy=policy,
target_key=attn_cls,
)
if self.pipeline_stage_manager is None:
self.append_or_create_method_replacement(
description={
"forward": get_deepseek_flash_attention_model_forward(
self.shard_config,
sp_mode=sp_mode,
sp_size=sp_size,
sp_group=sp_group,
),
},
policy=policy,
target_key="DeepseekModel",
)
embedding_cls = None
if self.shard_config.enable_tensor_parallelism:
embedding_cls = VocabParallelEmbedding1D
else:
if self.tie_weight:
embedding_cls = PaddingEmbedding
if self.shard_config.enable_tensor_parallelism:
# tensor parallelism for non-moe params
assert (
self.model.config.num_attention_heads % self.shard_config.tensor_parallel_size == 0
), f"The number of attention heads must be divisible by tensor parallel size."
assert (
self.model.config.num_key_value_heads % self.shard_config.tensor_parallel_size == 0
), f"The number of key_value heads must be divisible by tensor parallel size."
decoder_attribute_replacement = {
"self_attn.hidden_size": self.model.config.hidden_size // self.shard_config.tensor_parallel_size,
}
num_q_heads //= tp_size
decoder_attribute_replacement = {
"self_attn.hidden_size": self.model.config.hidden_size // self.shard_config.tensor_parallel_size,
"self_attn.num_heads": num_q_heads,
}
if num_kv_heads:
num_kv_heads //= tp_size
decoder_attribute_replacement["self_attn.num_key_value_heads"] = num_kv_heads
policy["DeepseekDecoderLayer"] = ModulePolicyDescription(
attribute_replacement=decoder_attribute_replacement,
sub_module_replacement=[
SubModuleReplacementDescription(
suffix="self_attn.q_proj",
target_module=Linear1D_Col,
kwargs={"fp8_communication": self.shard_config.fp8_communication},
),
SubModuleReplacementDescription(
suffix="self_attn.k_proj",
target_module=Linear1D_Col,
kwargs={"fp8_communication": self.shard_config.fp8_communication},
),
SubModuleReplacementDescription(
suffix="self_attn.v_proj",
target_module=Linear1D_Col,
kwargs={"fp8_communication": self.shard_config.fp8_communication},
),
SubModuleReplacementDescription(
suffix="self_attn.o_proj",
target_module=Linear1D_Row,
kwargs={"fp8_communication": self.shard_config.fp8_communication},
),
SubModuleReplacementDescription(
suffix="mlp.gate",
target_module=DeepseekMoEGate_Col,
kwargs={
"gather_output": True,
"fp8_communication": self.shard_config.fp8_communication,
"config": self.model.config,
},
ignore_if_not_exist=True,
),
],
)
if embedding_cls is not None:
self.append_or_create_submodule_replacement(
description=SubModuleReplacementDescription(
suffix="embed_tokens",
target_module=embedding_cls,
kwargs={
"make_vocab_size_divisible_by": self.shard_config.make_vocab_size_divisible_by,
"fp8_communication": self.shard_config.fp8_communication,
},
),
policy=policy,
target_key="DeepseekModel",
)
if self.shard_config.ep_group:
# expert parallel
self.append_or_create_submodule_replacement(
description=[
SubModuleReplacementDescription(
suffix="mlp",
target_module=EPDeepseekMoE,
kwargs={
"ep_group": self.shard_config.ep_group,
"tp_group": self.shard_config.tensor_parallel_process_group,
"moe_dp_group": self.shard_config.moe_dp_group,
"fp8_communication": self.shard_config.fp8_communication,
},
)
],
policy=policy,
target_key="DeepseekDecoderLayer",
)
# optimization configuration
if self.shard_config.enable_fused_normalization:
self.append_or_create_submodule_replacement(
description=[
SubModuleReplacementDescription(
suffix="input_layernorm",
target_module=FusedRMSNorm,
kwargs={"sp_partial_derived": sp_partial_derived},
),
SubModuleReplacementDescription(
suffix="post_attention_layernorm",
target_module=FusedRMSNorm,
kwargs={"sp_partial_derived": sp_partial_derived},
),
],
policy=policy,
target_key="DeepseekDecoderLayer",
)
self.append_or_create_submodule_replacement(
description=SubModuleReplacementDescription(
suffix="norm",
target_module=FusedRMSNorm,
kwargs={"sp_partial_derived": sp_partial_derived},
),
policy=policy,
target_key="DeepseekModel",
)
if self.shard_config.enable_flash_attention:
# NOTE: there is a bug for toggling flash attention in AutoModel, which has to be used for deepseek right now
from transformers.dynamic_module_utils import get_class_from_dynamic_module
flash_attn_cls = get_class_from_dynamic_module(
"deepseek-ai/deepseek-moe-16b-base--modeling_deepseek.DeepseekFlashAttention2",
"deepseek-ai/deepseek-moe-16b-base",
)
class TargetFlashAttn:
def __init__(self):
raise RuntimeError("This class should not be instantiated")
@staticmethod
def from_native_module(original_attn: nn.Module, *args, **kwargs) -> nn.Module:
original_attn.__class__ = flash_attn_cls
original_attn._flash_attn_uses_top_left_mask = not is_flash_attn_greater_or_equal_2_10()
return original_attn
self.append_or_create_submodule_replacement(
description=SubModuleReplacementDescription(
suffix="self_attn",
target_module=TargetFlashAttn,
),
policy=policy,
target_key="DeepseekDecoderLayer",
)
return policy
def postprocess(self):
return self.model
def set_pipeline_forward(self, model_cls: nn.Module, new_forward: Callable, policy: Dict) -> None:
"""If under pipeline parallel setting, replacing the original forward method of huggingface
to customized forward method, and add this changing to policy."""
if self.pipeline_stage_manager:
if self.shard_config.enable_sequence_parallelism:
# NOTE: we are replacing model forward for both sequence parallelism and pipeline parallelism
# if both are enabled, one of them will be ignored
raise NotImplementedError("Pipeline parallelism is not supported with sequence parallelism.")
stage_manager = self.pipeline_stage_manager
if self.model.__class__.__name__ == "DeepseekModel":
module = self.model
else:
module = self.model.model
layers_per_stage = stage_manager.distribute_layers(len(module.layers))
stage_index = stage_manager.get_stage_index(layers_per_stage)
method_replacement = {"forward": partial(new_forward, stage_manager=stage_manager, stage_index=stage_index)}
self.append_or_create_method_replacement(
description=method_replacement, policy=policy, target_key=model_cls
)
return
def get_held_layers(self) -> List[Module]:
"""Get pipeline layers for current stage."""
assert self.pipeline_stage_manager is not None
if self.model.__class__.__name__ == "DeepseekModel":
module = self.model
else:
module = self.model.model
stage_manager = self.pipeline_stage_manager
held_layers = []
layers_per_stage = stage_manager.distribute_layers(len(module.layers))
if stage_manager.is_first_stage():
held_layers.append(module.embed_tokens)
start_idx, end_idx = stage_manager.get_stage_index(layers_per_stage)
held_layers.extend(module.layers[start_idx:end_idx])
if stage_manager.is_last_stage():
held_layers.append(module.norm)
return held_layers
class DeepseekModelPolicy(DeepseekPolicy):
def __init__(self) -> None:
super().__init__()
def module_policy(self):
policy = super().module_policy()
if self.pipeline_stage_manager:
# set None as default
self.set_pipeline_forward(
model_cls="DeepseekModel",
new_forward=DeepseekPipelineForwards.deepseek_model_forward,
policy=policy,
)
return policy
def get_held_layers(self) -> List[Module]:
"""Get pipeline layers for current stage."""
held_layers = super().get_held_layers()
return held_layers
def get_shared_params(self) -> List[Dict[int, Tensor]]:
"""No shared params in llama model"""
return []
class DeepseekForCausalLMPolicy(DeepseekPolicy):
def module_policy(self):
policy = super().module_policy()
# TODO: assign pg mesh from plugin to all modules
if self.shard_config.enable_tensor_parallelism:
# add a new item for casual lm
new_item = {
"DeepseekForCausalLM": ModulePolicyDescription(
sub_module_replacement=[
SubModuleReplacementDescription(
suffix="lm_head",
target_module=Linear1D_Col,
kwargs=dict(gather_output=True, fp8_communication=self.shard_config.fp8_communication),
)
]
)
}
policy.update(new_item)
if self.pipeline_stage_manager:
# set None as default
self.set_pipeline_forward(
model_cls="DeepseekForCausalLM",
new_forward=DeepseekPipelineForwards.deepseek_for_causal_lm_forward,
policy=policy,
)
return policy
def get_held_layers(self) -> List[Module]:
"""Get pipeline layers for current stage."""
stage_manager = self.pipeline_stage_manager
held_layers = super().get_held_layers()
if stage_manager.is_last_stage():
held_layers.append(self.model.lm_head)
return held_layers
def get_shared_params(self) -> List[Dict[int, Tensor]]:
deepseek_model = self.model.model
if self.pipeline_stage_manager and self.pipeline_stage_manager.num_stages > 1:
if (
id(deepseek_model.embed_tokens.weight) == id(self.model.lm_head.weight)
and self.pipeline_stage_manager.num_stages > 1
):
# tie weights
return [
{
0: deepseek_model.embed_tokens.weight,
self.pipeline_stage_manager.num_stages - 1: self.model.lm_head.weight,
}
]
return []