# Copyright 2024 Google Inc. HuggingFace Inc. team. All rights reserved.
#
#
# 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.
from collections.abc import Callable

import torch
import torch.nn as nn
from huggingface_hub.dataclasses import strict

from ...activations import ACT2FN
from ...cache_utils import Cache, DynamicCache
from ...configuration_utils import PreTrainedConfig
from ...masking_utils import create_causal_mask, create_sliding_window_causal_mask
from ...modeling_flash_attention_utils import FlashAttentionKwargs
from ...modeling_layers import GradientCheckpointingLayer
from ...modeling_outputs import BaseModelOutputWithPast, CausalLMOutputWithPast
from ...modeling_rope_utils import (
    ROPE_INIT_FUNCTIONS,
    RopeParameters,
    dynamic_rope_update,
)
from ...modeling_utils import ALL_ATTENTION_FUNCTIONS
from ...processing_utils import Unpack
from ...utils import TransformersKwargs, auto_docstring, logging
from ...utils.generic import maybe_autocast
from ..gemma.modeling_gemma import (
    GemmaAttention,
    GemmaForCausalLM,
    GemmaForSequenceClassification,
    GemmaForTokenClassification,
    GemmaMLP,
    GemmaModel,
    GemmaPreTrainedModel,
    GemmaRMSNorm,
    GemmaRotaryEmbedding,
    apply_rotary_pos_emb,
    repeat_kv,
)


logger = logging.get_logger(__name__)


@auto_docstring(checkpoint="google/gemma2-7b")
@strict
class Gemma2Config(PreTrainedConfig):
    r"""
    query_pre_attn_scalar (`float`, *optional*, defaults to 256):
        scaling factor used on the attention scores
    final_logit_softcapping (`float`, *optional*, defaults to 30.0):
        scaling factor when applying tanh softcapping on the logits.
    attn_logit_softcapping (`float`, *optional*, defaults to 50.0):
        scaling factor when applying tanh softcapping on the attention scores.
    use_bidirectional_attention (`bool`, *optional*):
        If True, the model will attend to all text tokens instead of using a causal mask.

    ```python
    >>> from transformers import Gemma2Model, Gemma2Config
    >>> # Initializing a Gemma2 gemma2-7b style configuration
    >>> configuration = Gemma2Config()
    >>> # Initializing a model from the gemma2-7b style configuration
    >>> model = Gemma2Model(configuration)
    >>> # Accessing the model configuration
    >>> configuration = model.config
    ```"""

    model_type = "gemma2"
    keys_to_ignore_at_inference = ["past_key_values"]
    base_model_tp_plan = {
        "layers.*.self_attn.q_proj": "colwise",
        "layers.*.self_attn.k_proj": "colwise",
        "layers.*.self_attn.v_proj": "colwise",
        "layers.*.self_attn.o_proj": "rowwise",
        "layers.*.mlp.gate_proj": "colwise",
        "layers.*.mlp.up_proj": "colwise",
        "layers.*.mlp.down_proj": "rowwise",
    }
    base_model_pp_plan = {
        "embed_tokens": (["input_ids"], ["inputs_embeds"]),
        "layers": (["hidden_states", "attention_mask"], ["hidden_states"]),
        "norm": (["hidden_states"], ["hidden_states"]),
    }

    vocab_size: int = 256000
    hidden_size: int = 2304
    intermediate_size: int = 9216
    num_hidden_layers: int = 26
    num_attention_heads: int = 8
    num_key_value_heads: int = 4
    head_dim: int = 256
    hidden_activation: str = "gelu_pytorch_tanh"
    max_position_embeddings: int = 8192
    initializer_range: float = 0.02
    rms_norm_eps: float = 1e-6
    use_cache: bool = True
    pad_token_id: int | None = 0
    eos_token_id: int | list[int] | None = 1
    bos_token_id: int | None = 2
    tie_word_embeddings: bool = True
    rope_parameters: RopeParameters | dict | None = None
    attention_bias: bool = False
    attention_dropout: int | float | None = 0.0
    query_pre_attn_scalar: int = 256
    sliding_window: int | None = 4096
    layer_types: list[str] | None = None
    final_logit_softcapping: float | None = 30.0
    attn_logit_softcapping: float | None = 50.0
    use_bidirectional_attention: bool | None = None

    def __post_init__(self, **kwargs):
        if self.layer_types is None:
            self.layer_types = [
                "sliding_attention" if bool((i + 1) % 2) else "full_attention" for i in range(self.num_hidden_layers)
            ]

        super().__post_init__(**kwargs)

    def validate_architecture(self):
        """Part of `@strict`-powered validation. Validates the architecture of the config."""
        if self.hidden_size % self.num_attention_heads != 0:
            raise ValueError(
                f"The hidden size ({self.hidden_size}) is not a multiple of the number of attention "
                f"heads ({self.num_attention_heads})."
            )


class Gemma2RMSNorm(GemmaRMSNorm):
    pass


class Gemma2MLP(GemmaMLP):
    def __init__(self, config):
        super().__init__(config)
        self.act_fn = ACT2FN[config.hidden_activation]


class Gemma2RotaryEmbedding(GemmaRotaryEmbedding):
    def __init__(self, config: Gemma2Config, device=None):
        nn.Module.__init__()
        self.max_seq_len_cached = config.max_position_embeddings
        self.original_max_seq_len = config.max_position_embeddings

        self.config = config

        self.rope_type = self.config.rope_parameters["rope_type"]
        rope_init_fn: Callable = self.compute_default_rope_parameters
        if self.rope_type != "default":
            rope_init_fn = ROPE_INIT_FUNCTIONS[self.rope_type]
        inv_freq, self.attention_scaling = rope_init_fn(self.config, device)

        self.register_buffer("inv_freq", inv_freq, persistent=False)
        self.register_buffer("original_inv_freq", inv_freq.clone(), persistent=False)

    @torch.no_grad()
    @dynamic_rope_update  # power user: used with advanced RoPE types (e.g. dynamic rope)
    def forward(self, x, position_ids):
        inv_freq_expanded = self.inv_freq[None, :, None].float().expand(position_ids.shape[0], -1, 1).to(x.device)
        position_ids_expanded = position_ids[:, None, :].float()

        device_type = x.device.type if isinstance(x.device.type, str) and x.device.type != "mps" else "cpu"
        with maybe_autocast(device_type=device_type, enabled=False):  # Force float32
            freqs = (inv_freq_expanded.float() @ position_ids_expanded.float()).transpose(1, 2)
            emb = torch.cat((freqs, freqs), dim=-1)
            cos = emb.cos() * self.attention_scaling
            sin = emb.sin() * self.attention_scaling

        return cos.to(dtype=x.dtype), sin.to(dtype=x.dtype)


def eager_attention_forward(
    module: nn.Module,
    query: torch.Tensor,
    key: torch.Tensor,
    value: torch.Tensor,
    attention_mask: torch.Tensor | None,
    dropout: float | int = 0.0,
    scaling: float | None = None,
    softcap: float | None = None,
    **kwargs,
) -> tuple[torch.Tensor, torch.Tensor]:
    if scaling is None:
        scaling = module.head_dim**-0.5

    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 softcap is not None:
        attn_weights = attn_weights / softcap
        attn_weights = torch.tanh(attn_weights)
        attn_weights = attn_weights * softcap
    if attention_mask is not None:
        attn_weights = attn_weights + attention_mask

    # upcast attention to fp32
    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 Gemma2Attention(GemmaAttention):
    def __init__(self, config: Gemma2Config, layer_idx: int):
        self.layer_type = config.layer_types[layer_idx] if hasattr(config, "layer_types") else None

        super().__init__(config, layer_idx)
        self.attn_logit_softcapping = self.config.attn_logit_softcapping
        self.attention_dropout = self.config.attention_dropout
        self.is_causal = not getattr(config, "use_bidirectional_attention", False)
        self.scaling = config.query_pre_attn_scalar**-0.5
        self.sliding_window = config.sliding_window if self.layer_type == "sliding_attention" else None

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

        query_states = self.q_proj(hidden_states).view(hidden_shape).transpose(1, 2)
        key_states = self.k_proj(hidden_states).view(hidden_shape).transpose(1, 2)
        value_states = self.v_proj(hidden_states).view(hidden_shape).transpose(1, 2)

        cos, sin = position_embeddings
        query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin)

        if past_key_values is not None:
            key_states, value_states = past_key_values.update(key_states, value_states, self.layer_idx)

        attention_interface: Callable = ALL_ATTENTION_FUNCTIONS.get_interface(
            self.config._attn_implementation, eager_attention_forward
        )

        attn_output, attn_weights = attention_interface(
            self,
            query_states,
            key_states,
            value_states,
            attention_mask,
            dropout=self.attention_dropout if self.training else 0.0,
            scaling=self.scaling,
            sliding_window=self.sliding_window,
            softcap=self.attn_logit_softcapping,
            **kwargs,
        )

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


class Gemma2DecoderLayer(GradientCheckpointingLayer):
    def __init__(self, config: Gemma2Config, layer_idx: int):
        super().__init__()
        self.hidden_size = config.hidden_size
        self.config = config
        self.self_attn = Gemma2Attention(config=config, layer_idx=layer_idx)
        self.mlp = Gemma2MLP(config)
        self.input_layernorm = Gemma2RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
        self.post_attention_layernorm = Gemma2RMSNorm(config.hidden_size, eps=config.rms_norm_eps)

        self.pre_feedforward_layernorm = Gemma2RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
        self.post_feedforward_layernorm = Gemma2RMSNorm(config.hidden_size, eps=config.rms_norm_eps)

    def forward(
        self,
        hidden_states: torch.Tensor,
        position_embeddings: tuple[torch.Tensor, torch.Tensor] | None = None,
        attention_mask: torch.Tensor | None = None,
        position_ids: torch.LongTensor | None = None,
        past_key_values: Cache | None = None,
        **kwargs,
    ) -> tuple[torch.FloatTensor, tuple[torch.FloatTensor, torch.FloatTensor] | None]:
        residual = hidden_states

        hidden_states = self.input_layernorm(hidden_states)

        # Self Attention
        hidden_states, _ = self.self_attn(
            hidden_states=hidden_states,
            position_embeddings=position_embeddings,
            attention_mask=attention_mask,
            position_ids=position_ids,
            past_key_values=past_key_values,
            **kwargs,
        )
        hidden_states = self.post_attention_layernorm(hidden_states)
        hidden_states = residual + hidden_states

        residual = hidden_states
        hidden_states = self.pre_feedforward_layernorm(hidden_states)
        hidden_states = self.mlp(hidden_states)
        hidden_states = self.post_feedforward_layernorm(hidden_states)
        hidden_states = residual + hidden_states

        return hidden_states


class Gemma2PreTrainedModel(GemmaPreTrainedModel):
    pass


class Gemma2Model(GemmaModel):
    def __init__(self, config: Gemma2Config):
        super().__init__(config)
        self.layers = nn.ModuleList(
            [Gemma2DecoderLayer(config, layer_idx) for layer_idx in range(config.num_hidden_layers)]
        )
        self.rotary_emb = Gemma2RotaryEmbedding(config)

    def forward(
        self,
        input_ids: torch.LongTensor | None = None,
        attention_mask: torch.Tensor | None = None,
        position_ids: torch.LongTensor | None = None,
        past_key_values: Cache | None = None,
        inputs_embeds: torch.FloatTensor | None = None,
        use_cache: bool | None = None,
        **kwargs: Unpack[TransformersKwargs],
    ) -> BaseModelOutputWithPast:
        if (input_ids is None) ^ (inputs_embeds is not None):
            raise ValueError("You must specify exactly one of input_ids or inputs_embeds")

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

        if use_cache and past_key_values is None:
            past_key_values = DynamicCache(config=self.config)

        if position_ids is None:
            past_seen_tokens = past_key_values.get_seq_length() if past_key_values is not None else 0
            position_ids = torch.arange(inputs_embeds.shape[1], device=inputs_embeds.device) + past_seen_tokens
            position_ids = position_ids.unsqueeze(0)

        # It may already have been prepared by e.g. `generate`
        if not isinstance(causal_mask_mapping := attention_mask, dict):
            # Prepare mask arguments
            mask_kwargs = {
                "config": self.config,
                "inputs_embeds": inputs_embeds,
                "attention_mask": attention_mask,
                "past_key_values": past_key_values,
                "position_ids": position_ids,
            }
            # Create the masks
            causal_mask_mapping = {
                "full_attention": create_causal_mask(**mask_kwargs),
                "sliding_attention": create_sliding_window_causal_mask(**mask_kwargs),
            }

        # embed positions
        hidden_states = inputs_embeds
        position_embeddings = self.rotary_emb(hidden_states, position_ids)

        for i, decoder_layer in enumerate(self.layers[: self.config.num_hidden_layers]):
            hidden_states = decoder_layer(
                hidden_states,
                attention_mask=causal_mask_mapping[self.config.layer_types[i]],
                position_embeddings=position_embeddings,
                position_ids=position_ids,
                past_key_values=past_key_values,
                **kwargs,
            )

        hidden_states = self.norm(hidden_states)

        return BaseModelOutputWithPast(
            last_hidden_state=hidden_states,
            past_key_values=past_key_values,
        )


class Gemma2ForCausalLM(GemmaForCausalLM):
    def __init__(self, config):
        super().__init__(config)
        self.model = Gemma2Model(config)
        self.post_init()

    def forward(
        self,
        input_ids: torch.LongTensor | None = None,
        attention_mask: torch.Tensor | None = None,
        position_ids: torch.LongTensor | None = None,
        past_key_values: Cache | None = None,
        inputs_embeds: torch.FloatTensor | None = None,
        labels: torch.LongTensor | None = None,
        use_cache: bool | None = None,
        logits_to_keep: int | torch.Tensor = 0,
        **kwargs: Unpack[TransformersKwargs],
    ) -> CausalLMOutputWithPast:
        r"""
        Example:

        ```python
        >>> from transformers import AutoTokenizer, Gemma2ForCausalLM

        >>> model = Gemma2ForCausalLM.from_pretrained("google/gemma-2-9b")
        >>> tokenizer = AutoTokenizer.from_pretrained("google/gemma-2-9b")

        >>> prompt = "What is your favorite condiment?"
        >>> 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]
        "What is your favorite condiment?"
        ```"""
        # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
        outputs: BaseModelOutputWithPast = 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,
            **kwargs,
        )

        hidden_states = outputs.last_hidden_state
        # Only compute necessary logits, and do not upcast them to float if we are not computing the loss
        slice_indices = slice(-logits_to_keep, None) if isinstance(logits_to_keep, int) else logits_to_keep
        logits = self.lm_head(hidden_states[:, slice_indices, :])
        if self.config.final_logit_softcapping is not None:
            logits = logits / self.config.final_logit_softcapping
            logits = torch.tanh(logits)
            logits = logits * self.config.final_logit_softcapping

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

        return CausalLMOutputWithPast(
            loss=loss,
            logits=logits,
            past_key_values=outputs.past_key_values,
            hidden_states=outputs.hidden_states,
            attentions=outputs.attentions,
        )


class Gemma2ForSequenceClassification(GemmaForSequenceClassification):
    pass


class Gemma2ForTokenClassification(GemmaForTokenClassification):
    pass


__all__ = [
    "Gemma2Config",
    "Gemma2ForCausalLM",
    "Gemma2Model",
    "Gemma2PreTrainedModel",
    "Gemma2ForSequenceClassification",
    "Gemma2ForTokenClassification",
]