ColossalAI/examples/language/openmoe/model/openmoe_policy.py

from functools import partial
from typing import Callable, Dict, List, Optional, Union

import torch
import torch.nn as nn
import torch.nn.functional as F
from torch import Tensor
from torch.nn import Module
from transformers.modeling_outputs import CausalLMOutputWithPast
from transformers.utils import logging

from colossalai.moe.manager import MOE_MANAGER
from colossalai.pipeline.stage_manager import PipelineStageManager
from colossalai.shardformer.layer import FusedRMSNorm, Linear1D_Col
from colossalai.shardformer.policies.base_policy import ModulePolicyDescription, Policy, SubModuleReplacementDescription

from .modeling_openmoe import OpenMoeDecoderLayer, OpenMoeForCausalLM, OpenMoeModel

__all__ = ["OpenMoePolicy", "OpenMoeForCausalLMPolicy"]


class OpenMoePolicy(Policy):
    def config_sanity_check(self):
        pass

    def preprocess(self):
        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]:
        policy = {}

        if self.shard_config.enable_sequence_parallelism:
            self.shard_config.enable_sequence_parallelism = False
            raise NotImplementedError(
                "openmoe doesn't support sequence parallelism now, will ignore the sequence parallelism flag."
            )

        if self.shard_config.enable_tensor_parallelism:
            raise NotImplementedError("Tensor parallelism is not supported for openmoe model now.")

        # optimization configuration
        if self.shard_config.enable_fused_normalization:
            self.append_or_create_submodule_replacement(
                description=[
                    SubModuleReplacementDescription(
                        suffix="input_layernorm",
                        target_module=FusedRMSNorm,
                    ),
                    SubModuleReplacementDescription(
                        suffix="post_attention_layernorm",
                        target_module=FusedRMSNorm,
                    ),
                    SubModuleReplacementDescription(
                        suffix="pre_extra_mlp_layernorm",
                        target_module=FusedRMSNorm,
                        ignore_if_not_exist=True,
                    ),
                ],
                policy=policy,
                target_key=OpenMoeDecoderLayer,
            )

            self.append_or_create_submodule_replacement(
                description=SubModuleReplacementDescription(
                    suffix="norm",
                    target_module=FusedRMSNorm,
                ),
                policy=policy,
                target_key=OpenMoeModel,
            )

        if self.shard_config.enable_flash_attention:
            raise NotImplementedError("Flash attention has already been replaced in openmoe.")

        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:
            stage_manager = self.pipeline_stage_manager
            if self.model.__class__.__name__ == "OpenMoeModel":
                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__ == "OpenMoeModel":
            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

    def distribute_layers(self, num_layers: int, num_stages: int) -> List[int]:
        """Divide layers into stages"""
        if num_layers == 24 and num_stages == 4:
            return [7, 7, 7, 3]
        elif num_layers == 24 and num_stages == 2:
            return [15, 9]
        elif num_layers == 12 and num_stages == 4:
            return [5, 5, 5, 1]
        elif num_layers == 12 and num_stages == 2:
            return [8, 4]
        else:
            print(f"num_layers: {num_layers}, num_stages: {num_stages} not optimized, use origin pp policy")
            return super().distribute_layers(num_layers, num_stages)


class OpenMoeModelPolicy(OpenMoePolicy):
    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=OpenMoeModel,
                new_forward=OpenMoePipelineForwards.openmoe_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 OpenMoeForCausalLMPolicy(OpenMoePolicy):
    def module_policy(self):
        policy = super().module_policy()

        if self.shard_config.enable_tensor_parallelism:
            # add a new item for casual lm
            # TODO: recursively assign ep group foe all modules
            new_item = {
                OpenMoeForCausalLM: ModulePolicyDescription(
                    sub_module_replacement=[
                        SubModuleReplacementDescription(
                            suffix="lm_head",
                            target_module=Linear1D_Col,
                            kwargs=dict(gather_output=True),
                        )
                    ]
                )
            }
            policy.update(new_item)

        if self.pipeline_stage_manager:
            # set None as default
            self.set_pipeline_forward(
                model_cls=OpenMoeForCausalLM,
                new_forward=OpenMoePipelineForwards.llama_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]]:
        llama_model = self.model.model
        if self.pipeline_stage_manager and self.pipeline_stage_manager.num_stages > 1:
            if (
                id(llama_model.embed_tokens.weight) == id(self.model.lm_head.weight)
                and self.pipeline_stage_manager.num_stages > 1
            ):
                # tie weights
                return [
                    {
                        0: llama_model.embed_tokens.weight,
                        self.pipeline_stage_manager.num_stages - 1: self.model.lm_head.weight,
                    }
                ]
        return []


class OpenMoePipelineForwards:
    """
    This class serves as a micro library for forward function substitution of Llama models
    under pipeline setting.
    """

    @staticmethod
    def openmoe_model_forward(
        self: OpenMoeModel,
        input_ids: torch.LongTensor = None,
        attention_mask: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.LongTensor] = None,
        past_key_values: Optional[List[torch.FloatTensor]] = None,
        inputs_embeds: Optional[torch.FloatTensor] = None,
        use_cache: Optional[bool] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
        stage_manager: Optional[PipelineStageManager] = None,
        hidden_states: Optional[torch.FloatTensor] = None,
        stage_index: Optional[List[int]] = None,
        past_router_aux_loss: Optional[torch.FloatTensor] = None,
        past_router_z_loss: Optional[torch.FloatTensor] = None,
    ):
        # reset moe loss for different data
        MOE_MANAGER.reset_loss()

        logger = logging.get_logger(__name__)

        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
        output_hidden_states = (
            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
        )
        use_cache = use_cache if use_cache is not None else self.config.use_cache

        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        # retrieve input_ids and inputs_embeds
        if stage_manager.is_first_stage():
            if input_ids is not None and inputs_embeds is not None:
                raise ValueError("You cannot specify both decoder_input_ids and decoder_inputs_embeds at the same time")
            elif input_ids is not None:
                batch_size, seq_length = input_ids.shape
            elif inputs_embeds is not None:
                batch_size, seq_length, _ = inputs_embeds.shape
            else:
                raise ValueError("You have to specify either decoder_input_ids or decoder_inputs_embeds")
            device = input_ids.device if input_ids is not None else inputs_embeds.device
            if inputs_embeds is None:
                inputs_embeds = self.embed_tokens(input_ids)
            hidden_states = inputs_embeds
        else:
            input_shape = hidden_states.shape[:-1]
            batch_size, seq_length = input_shape
            device = hidden_states.device

        seq_length_with_past = seq_length
        past_key_values_length = 0

        # TODO(jianghai): left the recording kv-value tensors as () or None type, this feature may be added in the future.
        if output_attentions:
            logger.warning_once("output_attentions=True is not supported for pipeline models at the moment.")
            output_attentions = False
        if output_hidden_states:
            logger.warning_once("output_hidden_states=True is not supported for pipeline models at the moment.")
            output_hidden_states = False
        if use_cache:
            logger.warning_once("use_cache=True is not supported for pipeline models at the moment.")
            use_cache = False

        if past_key_values is not None:
            past_key_values_length = past_key_values[0][0].shape[2]
            seq_length_with_past = seq_length_with_past + past_key_values_length

        if position_ids is None:
            position_ids = torch.arange(
                past_key_values_length,
                seq_length + past_key_values_length,
                dtype=torch.long,
                device=device,
            )
            position_ids = position_ids.unsqueeze(0).view(-1, seq_length)
        else:
            position_ids = position_ids.view(-1, seq_length).long()

        # embed positions, for the first stage, hidden_states is the input embeddings,
        # for the other stages, hidden_states is the output of the previous stage
        if attention_mask is None:
            attention_mask = torch.ones(
                (batch_size, seq_length_with_past),
                dtype=torch.bool,
                device=hidden_states.device,
            )
        attention_mask = self._prepare_decoder_attention_mask(
            attention_mask,
            (batch_size, seq_length),
            hidden_states,
            past_key_values_length,
        )

        if self.gradient_checkpointing and self.training:
            if use_cache:
                logger.warning_once(
                    "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
                )
                use_cache = False

        # decoder layers
        all_hidden_states = () if output_hidden_states else None
        all_self_attns = () if output_attentions else None
        next_decoder_cache = () if use_cache else None

        start_idx, end_idx = stage_index[0], stage_index[1]
        for idx, decoder_layer in enumerate(self.layers[start_idx:end_idx], start=start_idx):
            if output_hidden_states:
                all_hidden_states += (hidden_states,)

            past_key_value = past_key_values[idx] if past_key_values is not None else None

            if self.gradient_checkpointing and self.training:

                def create_custom_forward(module):
                    def custom_forward(*inputs):
                        # None for past_key_value
                        return module(*inputs, output_attentions, None)

                    return custom_forward

                layer_outputs = torch.utils.checkpoint.checkpoint(
                    create_custom_forward(decoder_layer),
                    hidden_states,
                    attention_mask,
                    position_ids,
                    None,
                )
            else:
                layer_outputs = decoder_layer(
                    hidden_states,
                    attention_mask=attention_mask,
                    position_ids=position_ids,
                    past_key_value=past_key_value,
                    output_attentions=output_attentions,
                    use_cache=use_cache,
                )

            hidden_states = layer_outputs[0]

            if use_cache:
                next_decoder_cache += (layer_outputs[2 if output_attentions else 1],)
            if output_attentions:
                all_self_attns += (layer_outputs[1],)

        if stage_manager.is_last_stage():
            hidden_states = self.norm(hidden_states)

        # add hidden states from the last decoder layer
        if output_hidden_states:
            all_hidden_states += (hidden_states,)
        next_cache = next_decoder_cache if use_cache else None

        # concat past losses with current ones
        router_aux_loss, router_z_loss = MOE_MANAGER.get_loss()
        if past_router_aux_loss is not None and past_router_z_loss is not None:
            router_aux_loss = past_router_aux_loss + router_aux_loss
            router_z_loss = past_router_z_loss + router_z_loss

        if stage_manager.is_last_stage():
            return tuple(
                [
                    hidden_states,
                    next_cache,
                    all_hidden_states,
                    all_self_attns,
                    router_aux_loss,
                    router_z_loss,
                ]
            )
        # always return dict for imediate stage
        return {
            "hidden_states": hidden_states,
            "router_aux_loss": router_aux_loss,
            "router_z_loss": router_z_loss,
        }

    @staticmethod
    def llama_for_causal_lm_forward(
        self: OpenMoeForCausalLM,
        input_ids: torch.LongTensor = None,
        attention_mask: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.LongTensor] = None,
        past_key_values: Optional[List[torch.FloatTensor]] = None,
        inputs_embeds: Optional[torch.FloatTensor] = None,
        labels: Optional[torch.LongTensor] = None,
        use_cache: Optional[bool] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
        stage_manager: Optional[PipelineStageManager] = None,
        hidden_states: Optional[torch.FloatTensor] = None,
        stage_index: Optional[List[int]] = None,
        chunk_head: Optional[bool] = True,
        past_router_aux_loss: Optional[torch.FloatTensor] = None,
        past_router_z_loss: Optional[torch.FloatTensor] = None,
    ):
        r"""
        Args:
            labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
                Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,
                config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored
                (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`.

        Returns:

        Example:

        ```python
        >>> from transformers import AutoTokenizer, LlamaForCausalLM

        >>> model = LlamaForCausalLM.from_pretrained(PATH_TO_CONVERTED_WEIGHTS)
        >>> tokenizer = AutoTokenizer.from_pretrained(PATH_TO_CONVERTED_TOKENIZER)

        >>> prompt = "Hey, are you consciours? Can you talk to me?"
        >>> inputs = tokenizer(prompt, return_tensors="pt")

        >>> # Generate
        >>> generate_ids = model.generate(inputs.input_ids, max_length=30)
        >>> tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
        "Hey, are you consciours? Can you talk to me?\nI'm not consciours, but I can talk to you."
        ```"""
        logger = logging.get_logger(__name__)
        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
        output_hidden_states = (
            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
        )
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        # TODO(jianghai): left the recording kv-value tensors as () or None type, this feature may be added in the future.
        if output_attentions:
            logger.warning_once("output_attentions=True is not supported for pipeline models at the moment.")
            output_attentions = False
        if output_hidden_states:
            logger.warning_once("output_hidden_states=True is not supported for pipeline models at the moment.")
            output_hidden_states = False

        # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
        outputs = OpenMoePipelineForwards.openmoe_model_forward(
            self.model,
            input_ids=input_ids,
            attention_mask=attention_mask,
            position_ids=position_ids,
            past_key_values=past_key_values,
            inputs_embeds=inputs_embeds,
            use_cache=use_cache,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
            stage_manager=stage_manager,
            hidden_states=hidden_states,
            stage_index=stage_index,
            past_router_aux_loss=past_router_aux_loss,
            past_router_z_loss=past_router_z_loss,
        )

        if stage_manager.is_last_stage():
            (
                hidden_states,
                past_key_values,
                all_hidden_states,
                attentions,
                router_aux_loss,
                router_z_loss,
            ) = outputs

            if self.pretraining_tp > 1:
                lm_head_slices = self.lm_head.weight.split(self.vocab_size // self.pretraining_tp, dim=0)
                logits = [F.linear(hidden_states, lm_head_slices[i]) for i in range(self.pretraining_tp)]
                logits = torch.cat(logits, dim=-1)

            loss = None
            # if no training, just do forward
            if labels is None:
                logits = self.lm_head(hidden_states)
                logits = logits.float()
            # the vocab size for openmoe is 30w+
            # which causes great activation memory in training, up to 20G for one sequence
            # so we use chunk and checkpoint to reduce memory
            else:
                if chunk_head == True:

                    def create_custom_forward(module):
                        def custom_forward(*inputs):
                            logits = module(inputs[0])
                            logits = logits.float()
                            # Shift so that tokens < n predict n
                            shift_logits = logits[..., :-1, :].contiguous().float()
                            shift_labels = inputs[1][..., 1:].contiguous()
                            # Flatten the tokens
                            loss = self._calculate_loss(shift_logits, shift_labels)
                            return loss

                        return custom_forward

                    aux_loss, z_loss = self._calculate_router_loss(router_aux_loss, router_z_loss)
                    loss = aux_loss + z_loss
                    for batch_idx in range(hidden_states.shape[0]):
                        loss = loss + torch.utils.checkpoint.checkpoint(
                            create_custom_forward(self.lm_head),
                            hidden_states[batch_idx : batch_idx + 1, :],
                            labels[batch_idx : batch_idx + 1, :],
                        )
                    logits = None
                else:
                    logits = self.lm_head(hidden_states)
                    logits = logits.float()
                    # Shift so that tokens < n predict n
                    shift_logits = logits[..., :-1, :].contiguous()
                    shift_labels = labels[..., 1:].contiguous()
                    # Flatten the tokens
                    aux_loss, z_loss = self._calculate_router_loss(router_aux_loss, router_z_loss)
                    loss = aux_loss + z_loss
                    loss = loss + self._calculate_loss(shift_logits, shift_labels)

            if not return_dict:
                output = (logits,) + outputs[1:]
                return (loss,) + output if loss is not None else output

            return CausalLMOutputWithPast(
                loss=loss,
                logits=logits,
                past_key_values=past_key_values,
                hidden_states=all_hidden_states,
                attentions=attentions,
            )
        else:
            hidden_states = outputs["hidden_states"]
            router_aux_loss = outputs["router_aux_loss"]
            router_z_loss = outputs["router_z_loss"]
            return {
                "hidden_states": hidden_states,
                "past_router_aux_loss": router_aux_loss,
                "past_router_z_loss": router_z_loss,
            }