modeling_arcee_kda.py

# coding=utf-8
# Copyright 2022 EleutherAI and the HuggingFace Inc. team. All rights reserved.
#
# This code is based on EleutherAI's GPT-NeoX library and the GPT-NeoX
# and OPT implementations in this library. It has been modified from its
# original forms to accommodate minor architectural differences compared
# to GPT-NeoX and OPT used by the Meta AI team that trained the model.
#
# 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
#
#     http://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 math
from typing import Callable, List, Optional, Tuple, Union, Any

import torch
from torch import nn
from torch.nn import functional as F
from einops import rearrange

from transformers.activations import ACT2FN
from transformers.generation import GenerationMixin
from transformers.modeling_attn_mask_utils import AttentionMaskConverter
from transformers.modeling_flash_attention_utils import FlashAttentionKwargs
from transformers.modeling_outputs import (
    BaseModelOutputWithPast,
    CausalLMOutputWithPast,
)
from transformers.modeling_utils import ALL_ATTENTION_FUNCTIONS, PreTrainedModel
from transformers.processing_utils import Unpack
from transformers.pytorch_utils import ALL_LAYERNORM_LAYERS
from transformers.utils import (
    TransformersKwargs,
    add_start_docstrings,
    logging,
    replace_return_docstrings,
    auto_docstring,
)
from transformers.utils.generic import check_model_inputs
from .configuration_arcee_kda import ArceeKDAConfig


try:
    from fla.layers.kda import KimiDeltaAttention
    from fla.models.utils import Cache
except ImportError as e:
    print(e)
    raise ImportError("Plese run `pip install -U flash-linear-attention fla-core`")

logger = logging.get_logger(__name__)

_CHECKPOINT_FOR_DOC = "arcee-train/Arcee-4B-Base"
_CONFIG_FOR_DOC = "ArceeKDAConfig"


class ArceeRMSNorm(nn.Module):
    def __init__(self, hidden_size, eps=1e-6):
        """
        ArceeRMSNorm is equivalent to T5LayerNorm
        """
        super().__init__()
        self.weight = nn.Parameter(torch.ones(hidden_size))
        self.variance_epsilon = eps

    def forward(self, hidden_states):
        input_dtype = hidden_states.dtype
        hidden_states = hidden_states.to(torch.float32)
        variance = hidden_states.pow(2).mean(-1, keepdim=True)
        hidden_states = hidden_states * torch.rsqrt(variance + self.variance_epsilon)
        return self.weight * hidden_states.to(input_dtype)

    def extra_repr(self):
        return f"{tuple(self.weight.shape)}, eps={self.variance_epsilon}"


ALL_LAYERNORM_LAYERS.append(ArceeRMSNorm)


class ArceeMLP(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.config = config
        self.hidden_size = config.hidden_size
        self.intermediate_size = config.intermediate_size
        self.up_proj = nn.Linear(
            self.hidden_size, self.intermediate_size, bias=config.mlp_bias
        )
        self.down_proj = nn.Linear(
            self.intermediate_size, self.hidden_size, bias=config.mlp_bias
        )
        self.act_fn = ACT2FN[config.hidden_act]

    def forward(self, x):
        down_proj = self.down_proj(self.act_fn(self.up_proj(x)))
        return down_proj


def repeat_kv(hidden_states: torch.Tensor, n_rep: int) -> torch.Tensor:
    """
    This is the equivalent of torch.repeat_interleave(x, dim=1, repeats=n_rep). The hidden states go from (batch,
    num_key_value_heads, seqlen, head_dim) to (batch, num_attention_heads, seqlen, head_dim)
    """
    batch, num_key_value_heads, slen, head_dim = hidden_states.shape
    if n_rep == 1:
        return hidden_states
    hidden_states = hidden_states[:, :, None, :, :].expand(
        batch, num_key_value_heads, n_rep, slen, head_dim
    )
    return hidden_states.reshape(batch, num_key_value_heads * n_rep, slen, head_dim)


def eager_attention_forward(
    module: nn.Module,
    query: torch.Tensor,
    key: torch.Tensor,
    value: torch.Tensor,
    attention_mask: Optional[torch.Tensor],
    scaling: float,
    dropout: float = 0.0,
    **kwargs,
):
    key_states = repeat_kv(key, module.num_key_value_groups)
    value_states = repeat_kv(value, module.num_key_value_groups)

    attn_weights = torch.matmul(query, key_states.transpose(2, 3)) * scaling
    if attention_mask is not None:
        causal_mask = attention_mask[:, :, :, : key_states.shape[-2]]
        attn_weights = attn_weights + causal_mask

    attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(
        query.dtype
    )
    attn_weights = nn.functional.dropout(
        attn_weights, p=dropout, training=module.training
    )
    attn_output = torch.matmul(attn_weights, value_states)
    attn_output = attn_output.transpose(1, 2).contiguous()

    return attn_output, attn_weights


class ArceeNopeAttention(nn.Module):
    def __init__(self, config: ArceeKDAConfig, layer_idx: int):
        super().__init__()
        self.config = config
        self.layer_idx = layer_idx
        self.head_dim = getattr(
            config, "head_dim", config.hidden_size // config.num_attention_heads
        )
        self.num_key_value_groups = (
            config.num_attention_heads // config.num_key_value_heads
        )
        self.scaling = self.head_dim**-0.5
        self.attention_dropout = config.attention_dropout
        self.is_causal = True

        self.q_proj = nn.Linear(
            config.hidden_size,
            config.num_attention_heads * self.head_dim,
            bias=config.attention_bias,
        )
        self.k_proj = nn.Linear(
            config.hidden_size,
            config.num_key_value_heads * self.head_dim,
            bias=config.attention_bias,
        )
        self.v_proj = nn.Linear(
            config.hidden_size,
            config.num_key_value_heads * self.head_dim,
            bias=config.attention_bias,
        )
        self.o_proj = nn.Linear(
            config.num_attention_heads * self.head_dim,
            config.hidden_size,
            bias=config.attention_bias,
        )

        self.gate_proj = nn.Linear(
            config.hidden_size, config.num_attention_heads * self.head_dim, bias=False
        )

    def forward(
        self,
        hidden_states: torch.Tensor,
        attention_mask: Optional[torch.Tensor],
        past_key_values: Optional[Cache] = None,
        **kwargs: Unpack[FlashAttentionKwargs],
    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Cache]]:
        input_shape = hidden_states.shape[:-1]
        hidden_shape = (*input_shape, -1, self.head_dim)

        query_states = self.q_proj(hidden_states).view(hidden_shape)
        key_states = self.k_proj(hidden_states).view(hidden_shape)
        value_states = self.v_proj(hidden_states).view(hidden_shape)
        gate_states = self.gate_proj(hidden_states)

        if past_key_values is not None:
            cache_has_content = past_key_values.get_seq_length(self.layer_idx) > 0
            k_cached, v_cached = past_key_values.update(
                attn_state=(key_states.flatten(-2, -1), value_states.flatten(-2, -1)),
                layer_idx=self.layer_idx,
                offset=hidden_states.shape[1],
                cache_kwargs=dict(window_size=None),
            )["attn_state"]
            if cache_has_content:
                batch_size = key_states.shape[0]
                key_states = k_cached.view(
                    batch_size, -1, self.config.num_key_value_heads, self.head_dim
                )
                value_states = v_cached.view(
                    batch_size, -1, self.config.num_key_value_heads, self.head_dim
                )

        attention_interface: Callable = eager_attention_forward
        if self.config._attn_implementation != "eager":
            attention_interface = ALL_ATTENTION_FUNCTIONS[
                self.config._attn_implementation
            ]

        attn_output, attn_weights = attention_interface(
            self,
            query_states.transpose(1, 2),
            key_states.transpose(1, 2),
            value_states.transpose(1, 2),
            attention_mask,
            dropout=0.0 if not self.training else self.attention_dropout,
            scaling=self.scaling,
            **kwargs,
        )

        attn_output = attn_output.reshape(*input_shape, -1).contiguous()
        attn_output = attn_output * F.sigmoid(gate_states)
        attn_output = self.o_proj(attn_output)
        return attn_output, attn_weights, past_key_values


class ArceeKDADecoderLayer(nn.Module):
    def __init__(self, config: ArceeKDAConfig, layer_idx: int):
        super().__init__()
        self.hidden_size = config.hidden_size

        if config.is_kda_layer(layer_idx):
            self.self_attn = KimiDeltaAttention(
                layer_idx=layer_idx,
                hidden_size=config.hidden_size,
                **config.linear_attn_config,
            )
            self.is_linear_attn = True
        else:
            self.self_attn = ArceeNopeAttention(config=config, layer_idx=layer_idx)
            self.is_linear_attn = False

        self.mlp = ArceeMLP(config)
        self.input_layernorm = ArceeRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
        self.post_attention_layernorm = ArceeRMSNorm(
            config.hidden_size, eps=config.rms_norm_eps
        )

    def forward(
        self,
        hidden_states: torch.Tensor,
        attention_mask: Optional[torch.Tensor] = None,
        past_key_values: Optional[Cache] = None,
        use_cache: Optional[bool] = False,
        **kwargs: Unpack[FlashAttentionKwargs],
    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Cache]]:
        residual = hidden_states

        hidden_states = self.input_layernorm(hidden_states)

        # Self Attention
        hidden_states, attn_weights, past_key_values = self.self_attn(
            hidden_states=hidden_states,
            attention_mask=attention_mask,
            past_key_values=past_key_values,
            use_cache=use_cache,
            **kwargs,
        )
        hidden_states = residual + hidden_states

        # Fully Connected
        residual = hidden_states
        hidden_states = self.post_attention_layernorm(hidden_states)
        hidden_states = self.mlp(hidden_states)
        hidden_states = residual + hidden_states

        return (hidden_states, attn_weights, past_key_values)


Arcee_START_DOCSTRING = r"""
    This model inherits from [`PreTrainedModel`]. Check the superclass documentation for the generic methods the
    library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads
    etc.)

    This model is also a PyTorch [torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) subclass.
    Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage
    and behavior.

    Parameters:
        config ([`ArceeKDAConfig`]):
            Model configuration class with all the parameters of the model. Initializing with a config file does not
            load the weights associated with the model, only the configuration. Check out the
            [`~PreTrainedModel.from_pretrained`] method to load the model weights.
"""


@add_start_docstrings(
    "The bare Arcee-KDA Model outputting raw hidden-states without any specific head on top.",
    Arcee_START_DOCSTRING,
)
class ArceeKDAPreTrainedModel(PreTrainedModel):
    config_class = ArceeKDAConfig
    base_model_prefix = "model"
    supports_gradient_checkpointing = True
    _no_split_modules = ["ArceeKDADecoderLayer"]
    _skip_keys_device_placement = ["past_key_values"]
    _supports_flash_attn_2 = True
    _supports_sdpa = True
    _supports_flex_attn = True
    _supports_attention_backend = True
    _can_record_outputs = {
        "hidden_states": ArceeKDADecoderLayer,
        "attentions": ArceeNopeAttention,
    }
    _is_stateful = True
    _supports_cache_class = True

    def _init_weights(self, module):
        std = self.config.initializer_range
        if (
            isinstance(module, KimiDeltaAttention)
            and next(module.parameters()).device.type != "meta"
        ):
            with torch.no_grad():
                module.A_log.copy_(nn.init.uniform_(module.A_log, a=1, b=16).log())
                dt = torch.exp(
                    nn.init.uniform_(module.dt_bias) * (math.log(0.1) - math.log(0.001))
                    + math.log(0.001),
                ).clamp(min=1e-4)
                inv_dt = dt + torch.log(-torch.expm1(-dt))
                module.dt_bias.copy_(inv_dt)
                module.dt_bias._is_hf_initialized = True
        if isinstance(module, (nn.Linear, nn.Conv1d)):
            # Slightly different from the TF version which uses truncated_normal for initialization
            # cf https://github.com/pytorch/pytorch/pull/5617
            nn.init.normal_(module.weight, mean=0.0, std=self.config.initializer_range)
            if module.bias is not None and not getattr(
                module.bias, "_is_hf_initialized", False
            ):
                nn.init.zeros_(module.bias)
        elif isinstance(module, nn.Embedding):
            nn.init.normal_(module.weight, mean=0.0, std=self.config.initializer_range)
        elif hasattr(module, "reset_parameters"):
            module.reset_parameters()


@add_start_docstrings(
    "The bare Arcee-KDA Model outputting raw hidden-states without any specific head on top.",
    Arcee_START_DOCSTRING,
)
class ArceeKDAModel(ArceeKDAPreTrainedModel):
    """
    Transformer decoder consisting of *config.num_hidden_layers* layers. Each layer is a [`ArceeKDADecoderLayer`]

    Args:
        config: ArceeKDAConfig
    """

    def __init__(self, config: ArceeKDAConfig):
        super().__init__(config)
        self.padding_idx = config.pad_token_id
        self.vocab_size = config.vocab_size

        self.embed_tokens = nn.Embedding(
            config.vocab_size, config.hidden_size, self.padding_idx
        )
        self.layers = nn.ModuleList(
            [
                ArceeKDADecoderLayer(config, layer_idx)
                for layer_idx in range(config.num_hidden_layers)
            ]
        )
        self.norm = ArceeRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
        self.gradient_checkpointing = False

        # Initialize weights and apply final processing
        self.post_init()

    def get_input_embeddings(self):
        return self.embed_tokens

    def set_input_embeddings(self, value):
        self.embed_tokens = value

    @auto_docstring
    def forward(
        self,
        input_ids: Optional[torch.LongTensor] = None,
        attention_mask: Optional[torch.Tensor] = None,
        past_key_values: Optional[Cache] = 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,
        **kwargs: Unpack[dict],
    ) -> Union[Tuple, BaseModelOutputWithPast]:
        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
        )

        if (input_ids is None) ^ (inputs_embeds is not None):
            raise ValueError(
                "You must specify exactly one of input_ids or inputs_embeds"
            )

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

        if inputs_embeds is None:
            inputs_embeds = self.embed_tokens(input_ids)

        if use_cache and not isinstance(past_key_values, Cache):
            past_key_values = Cache.from_legacy_cache(past_key_values)

        hidden_states = inputs_embeds

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

        for decoder_layer in self.layers[: self.config.num_hidden_layers]:
            if output_hidden_states:
                all_hidden_states += (hidden_states,)

            if self.gradient_checkpointing and self.training:
                layer_outputs = self._gradient_checkpointing_func(
                    decoder_layer.__call__,
                    hidden_states,
                    attention_mask,
                    past_key_values,
                    use_cache,
                )
            else:
                layer_outputs = decoder_layer(
                    hidden_states,
                    attention_mask=attention_mask,
                    past_key_values=past_key_values,
                    use_cache=use_cache,
                    **kwargs,
                )

            hidden_states, attn_weights, past_key_values = layer_outputs

            if output_attentions:
                all_self_attns += (attn_weights,)

        hidden_states = self.norm(hidden_states)

        # add hidden states from the last decoder layer
        if output_hidden_states:
            all_hidden_states += (hidden_states,)

        output = BaseModelOutputWithPast(
            last_hidden_state=hidden_states,
            past_key_values=past_key_values,
            hidden_states=all_hidden_states,
            attentions=all_self_attns,
        )
        return output if return_dict else output.to_tuple()


class ArceeKDAForCausalLM(ArceeKDAPreTrainedModel, GenerationMixin):
    _tied_weights_keys = ["lm_head.weight"]
    _tp_plan = {"lm_head": "colwise_rep"}
    _pp_plan = {"lm_head": (["hidden_states"], ["logits"])}

    def __init__(self, config):
        super().__init__(config)
        self.model = ArceeKDAModel(config)
        self.vocab_size = config.vocab_size
        self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)

        # Initialize weights and apply final processing
        self.post_init()

    def get_input_embeddings(self):
        return self.model.embed_tokens

    def set_input_embeddings(self, value):
        self.model.embed_tokens = value

    def get_output_embeddings(self):
        return self.lm_head

    def set_output_embeddings(self, new_embeddings):
        self.lm_head = new_embeddings

    def set_decoder(self, decoder):
        self.model = decoder

    def get_decoder(self):
        return self.model

    @auto_docstring
    @replace_return_docstrings(
        output_type=CausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC
    )
    def forward(
        self,
        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,
        **kwargs: Unpack[TransformersKwargs],
    ) -> Union[Tuple, CausalLMOutputWithPast]:
        r"""
            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, ArceeForCausalLM

        >>> model = ArceeForCausalLM.from_pretrained("arcee-train/Arcee-4B-Base")
        >>> tokenizer = AutoTokenizer.from_pretrained("arcee-train/Arcee-4B-Base")

        >>> prompt = "Hey, are you conscious? 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 conscious? Can you talk to me?\nI'm not conscious, but I can talk to you."
        ```"""
        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
        )

        # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
        outputs = 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,
            **kwargs,
        )

        hidden_states = outputs[0]
        logits = self.lm_head(hidden_states)

        loss = None
        if labels is not None:
            loss = self.loss_function(
                logits=logits,
                labels=labels,
                vocab_size=self.config.vocab_size,
                **kwargs,
            )

        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=outputs.past_key_values,
            hidden_states=outputs.hidden_states,
            attentions=outputs.attentions,
        )


__all__ = [
    "ArceeKDAForCausalLM",
    "ArceeKDAModel",
    "ArceeKDAPreTrainedModel",
]