modeling_mos_mamba.py

# coding=utf-8
# Copyright 2024 state-spaces/mamba org and HuggingFace Inc. team.
#
# 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.
"""PyTorch MAMBA model."""

import math
from dataclasses import dataclass
from typing import Any, Dict, Optional, Tuple, Union

import torch
import torch.utils.checkpoint
from torch import nn
from torch.nn import CrossEntropyLoss

from transformers.activations import ACT2FN
from transformers.modeling_utils import PreTrainedModel
from transformers.utils import ModelOutput
# from transformers.utils.import_utils import is_causal_conv1d_available, is_mamba_ssm_available
from .configuration_mos_mamba import MoSMambaConfig

import torch.nn.functional as F


# if is_mamba_ssm_available():
#     from mamba_ssm.ops.selective_scan_interface import mamba_inner_fn, selective_scan_fn
#     from mamba_ssm.ops.triton.selective_state_update import selective_state_update
# else:
#     selective_state_update, selective_scan_fn, mamba_inner_fn = None, None, None

# if is_causal_conv1d_available():
#     from causal_conv1d import causal_conv1d_fn, causal_conv1d_update
# else:
#     causal_conv1d_update, causal_conv1d_fn = None, None


try:
    from mamba_ssm.ops.selective_scan_interface import mamba_inner_fn, selective_scan_fn
    from mamba_ssm.ops.triton.selective_state_update import selective_state_update
except:
    selective_state_update, selective_scan_fn, mamba_inner_fn = None, None, None

try:
    from causal_conv1d import causal_conv1d_fn, causal_conv1d_update
except:
    causal_conv1d_update, causal_conv1d_fn = None, None
    

is_fast_path_available = all(
    (selective_state_update, selective_scan_fn, causal_conv1d_fn, causal_conv1d_update, mamba_inner_fn)
)

_CHECKPOINT_FOR_DOC = "state-spaces/mamba-130m-hf"
_CONFIG_FOR_DOC = "MoSMambaConfig"


def load_balancing_loss_func(
    gate_logits: torch.Tensor, num_selectivities: torch.Tensor = None, top_k=2, attention_mask: Optional[torch.Tensor] = None
) -> float:
    r"""
    Computes auxiliary load balancing loss as in Switch Transformer - implemented in Pytorch.

    See Switch Transformer (https://arxiv.org/abs/2101.03961) for more details. This function implements the loss
    function presented in equations (4) - (6) of the paper. It aims at penalizing cases where the routing between
    experts is too unbalanced.

    Args:
        gate_logits (Union[`torch.Tensor`, Tuple[torch.Tensor]):
            Logits from the `gate`, should be a tuple of model.config.num_hidden_layers tensors of
            shape [batch_size X sequence_length, num_selectivities].
        attention_mask (`torch.Tensor`, None):
            The attention_mask used in forward function
            shape [batch_size X sequence_length] if not None.
        num_selectivities (`int`, *optional*):
            Number of experts

    Returns:
        The auxiliary loss.
    """
    if gate_logits is None or not isinstance(gate_logits, tuple):
        return 0

    if isinstance(gate_logits, tuple):
        compute_device = gate_logits[0].device
        concatenated_gate_logits = torch.cat([layer_gate.to(compute_device) for layer_gate in gate_logits], dim=0)

    routing_weights = torch.nn.functional.softmax(concatenated_gate_logits, dim=-1)

    _, selected_experts = torch.topk(routing_weights, top_k, dim=-1)

    expert_mask = torch.nn.functional.one_hot(selected_experts, num_selectivities)

    if attention_mask is None:
        # Compute the percentage of tokens routed to each experts
        tokens_per_expert = torch.mean(expert_mask.float(), dim=0)

        # Compute the average probability of routing to these experts
        router_prob_per_expert = torch.mean(routing_weights, dim=0)
    else:
        batch_size, sequence_length = attention_mask.shape
        num_hidden_layers = concatenated_gate_logits.shape[0] // (batch_size * sequence_length)

        # Compute the mask that masks all padding tokens as 0 with the same shape of expert_mask
        expert_attention_mask = (
            attention_mask[None, :, :, None, None]
            .expand((num_hidden_layers, batch_size, sequence_length, top_k, num_selectivities))
            .reshape(-1, top_k, num_selectivities)
            .to(compute_device)
        )

        # Compute the percentage of tokens routed to each experts
        tokens_per_expert = torch.sum(expert_mask.float() * expert_attention_mask, dim=0) / torch.sum(
            expert_attention_mask, dim=0
        )

        # Compute the mask that masks all padding tokens as 0 with the same shape of tokens_per_expert
        router_per_expert_attention_mask = (
            attention_mask[None, :, :, None]
            .expand((num_hidden_layers, batch_size, sequence_length, num_selectivities))
            .reshape(-1, num_selectivities)
            .to(compute_device)
        )

        # Compute the average probability of routing to these experts
        router_prob_per_expert = torch.sum(routing_weights * router_per_expert_attention_mask, dim=0) / torch.sum(
            router_per_expert_attention_mask, dim=0
        )

    overall_loss = torch.sum(tokens_per_expert * router_prob_per_expert.unsqueeze(0))
    return overall_loss * num_selectivities


class MixtralBlockSparseTop2MLP(nn.Module):
    def __init__(self, intermediate_size, hidden_size, ssm_size):
        super().__init__()
        self.ffn_dim = intermediate_size
        self.hidden_dim = hidden_size
        self.ssm_dim = ssm_size

        self.w1 = nn.Linear(self.hidden_dim, self.ffn_dim, bias=False)
        self.w2 = nn.Linear(self.ffn_dim, self.ssm_dim, bias=False)
        self.w3 = nn.Linear(self.hidden_dim, self.ffn_dim, bias=False)
        self.w4 = nn.Linear(self.ffn_dim, self.hidden_dim, bias=False)

        self.act_fn = ACT2FN['silu']

    def forward(self, hidden_states):
        current_hidden_states = self.act_fn(self.w1(hidden_states)) * self.w3(hidden_states)
        current_hidden_states = self.w4(current_hidden_states)

        return current_hidden_states

class MixtureOfSelectivity(nn.Module):
    def __init__(self, intermediate_size, ssm_size):
        super().__init__()
        self.intermediate_size = intermediate_size
        self.ssm_dim = ssm_size

#         self.w1 = nn.Linear(self.hidden_dim, self.ffn_dim, bias=False)
        self.w2 = nn.Linear(self.intermediate_size, self.ssm_dim, bias=False)


    def forward(self, hidden_states):
#         current_hidden_states = self.act_fn(self.w1(hidden_states)) * self.w3(hidden_states)
        return self.w2(hidden_states)

class MoSMambaCache:
    """
    Arguments:
        config: MoSMambaConfig
        batch_size: int
        dtype: torch.dtype
        device: torch.device

    Attributes:
        seqlen_offset: int
        dtype: torch.dtype
        conv_states: Dict[int, torch.Tensor] # layer_idx -> [batch_size, intermediate_size, conv_kernel_size]
        ssm_states: Dict[int, torch.Tensor] # layer_idx -> [batch_size, intermediate_size, ssm_state_size]
    """

    def __init__(
        self, config: MoSMambaConfig, batch_size: int, dtype: torch.dtype = torch.float16, device: Optional[str] = None
    ):
        self.seqlen_offset = 0
        self.dtype = dtype
        intermediate_size = config.intermediate_size
        ssm_state_size = config.state_size
        conv_kernel_size = config.conv_kernel

        self.conv_states = {
            i: torch.zeros(batch_size, intermediate_size, conv_kernel_size, device=device, dtype=dtype)
            for i in range(config.num_hidden_layers)
        }
        self.ssm_states = {
            i: torch.zeros(batch_size, intermediate_size, ssm_state_size, device=device, dtype=dtype)
            for i in range(config.num_hidden_layers)
        }


class MoSMambaMixer(nn.Module):
    """
    Compute ∆, A, B, C, and D the state space parameters and compute the `contextualized_states`.
    A, D are input independent (see MoSMamba paper [1] Section 3.5.2 "Interpretation of A" for why A isn't selective)
    ∆, B, C are input-dependent (this is a key difference between MoSMamba and the linear time invariant S4,
    and is why MoSMamba is called **selective** state spaces)
    """

    def __init__(self, config: MoSMambaConfig, layer_idx: int):
        super().__init__()
        self.hidden_size = config.hidden_size
        self.ssm_state_size = config.state_size
        self.conv_kernel_size = config.conv_kernel
        self.intermediate_size = config.intermediate_size
        self.time_step_rank = int(config.time_step_rank)
        self.layer_idx = layer_idx
        self.use_conv_bias = config.use_conv_bias
        self.conv1d = nn.Conv1d(
            in_channels=self.intermediate_size,
            out_channels=self.intermediate_size,
            bias=config.use_conv_bias,
            kernel_size=config.conv_kernel,
            groups=self.intermediate_size,
            padding=config.conv_kernel - 1,
        )

        self.activation = config.hidden_act
        self.act = ACT2FN[config.hidden_act]

        # num experts
        self.num_selectivities = config.num_selectivities

        # num selected experts
        self.top_k = config.num_selectivities_per_tok

        # projection of the input hidden states
        self.in_proj = nn.Linear(self.hidden_size, self.intermediate_size * 2, bias=config.use_bias)
        # selective projection used to make dt, B and C input dependant
#         self.x_proj = nn.Linear(self.intermediate_size, self.time_step_rank + self.ssm_state_size * 2, bias=False

#         self.x_proj = nn.ModuleList([nn.Linear(self.intermediate_size, self.time_step_rank + self.ssm_state_size * 2, bias=False) for _ in range(self.num_selectivities)])
#         for i in range(self.num_selectivities):
#             self.x_proj.add_module("x_proj_"+str(i), nn.Linear(self.intermediate_size, self.time_step_rank + self.ssm_state_size * 2, bias=False))

#         self.x_proj_0 = nn.Linear(self.intermediate_size, self.time_step_rank + self.ssm_state_size * 2, bias=False)
#         self.x_proj_1 = nn.Linear(self.intermediate_size, self.time_step_rank + self.ssm_state_size * 2, bias=False)
#         self.x_proj_2 = nn.Linear(self.intermediate_size, self.time_step_rank + self.ssm_state_size * 2, bias=False)
#         self.x_proj_3 = nn.Linear(self.intermediate_size, self.time_step_rank + self.ssm_state_size * 2, bias=False)
#         self.x_proj_4 = nn.Linear(self.intermediate_size, self.time_step_rank + self.ssm_state_size * 2, bias=False)
#         self.x_proj_5 = nn.Linear(self.intermediate_size, self.time_step_rank + self.ssm_state_size * 2, bias=False)


        # self.x_proj2 = nn.ModuleList([MixtralBlockSparseTop2MLP(self.intermediate_size,self.hidden_size, self.time_step_rank + self.ssm_state_size * 2) for _ in range(self.num_selectivities)])
        self.x_proj = nn.ModuleList()
        for i in range(self.num_selectivities):
          self.x_proj.add_module(f"w{i}",nn.Linear(self.intermediate_size, self.time_step_rank + self.ssm_state_size * 2, bias=False))

        self.gate = nn.Linear(self.hidden_size, self.num_selectivities, bias=False)

        # time step projection (discretization)
        self.dt_proj = nn.Linear(self.time_step_rank, self.intermediate_size, bias=True)

        # S4D real initialization. These are not discretized!
        # The core is to load them, compute the discrete states, then write the updated state. Keeps the memory bounded
        A = torch.arange(1, self.ssm_state_size + 1, dtype=torch.float32)[None, :]
        A = A.expand(self.intermediate_size, -1).contiguous()

        self.A_log = nn.Parameter(torch.log(A))
        self.D = nn.Parameter(torch.ones(self.intermediate_size))
        self.out_proj = nn.Linear(self.intermediate_size, self.hidden_size, bias=config.use_bias)
        self.use_bias = config.use_bias
        
        self.jitter_noise = 0.001

        self.register_parameter("A_log", self.A_log)
        self.register_parameter("D", self.D)

#         for i in enumerate(self.x_proj):
#             self.register_parameter("x_proj_"+str(i), x)


    def cuda_kernels_forward(self, hidden_states: torch.Tensor, x_proj, cache_params: Optional[MoSMambaCache] = None):
        # 1. Gated MLP's linear projection
#         router_logits =
        batch_size, seq_len, _ = hidden_states.shape

        projected_states = self.in_proj(hidden_states).transpose(1, 2)

        if projected_states.shape[-1] == 0:
          hidden_states, gate = projected_states.chunk(2, dim=1)
          dtype = hidden_states.dtype

          if cache_params is not None:
            ssm_state = cache_params.ssm_states[self.layer_idx].clone()
            if cache_params.seqlen_offset > 0:
                conv_state = cache_params.conv_states[self.layer_idx]                   # [batch, intermediate_size, conv_kernel_size]
                conv_state = torch.roll(conv_state, shifts=-1, dims=-1)
                conv_state[:, :, -1] = hidden_states[:, :, 0]
                cache_params.conv_states[self.layer_idx].copy_(conv_state)
                hidden_states = torch.sum(conv_state * self.conv1d.weight[:, 0, :], dim=-1)
                if self.use_conv_bias:
                    hidden_states += self.conv1d.bias
                hidden_states = self.act(hidden_states).to(dtype).unsqueeze(-1)         # [batch, intermediate_size, 1] : decoding
            else:
                conv_state = nn.functional.pad(
                    hidden_states,
                    (self.conv_kernel_size - hidden_states.shape[-1], 0)
                )
                cache_params.conv_states[self.layer_idx].copy_(conv_state)
                if hidden_states.shape[-1] == 0:
                  hidden_states = hidden_states.permute(2,1,0)
                hidden_states = self.act(self.conv1d(hidden_states)[..., :seq_len])     # [batch, intermediate_size, seq_len]
          else:
            ssm_state = torch.zeros(
                (batch_size, self.intermediate_size, self.ssm_state_size),
                device=hidden_states.device, dtype=dtype
            )
            # print(hidden_states.shape)
            # print(self.conv1d)
            if hidden_states.shape[-1] == 0:
              hidden_states = hidden_states.permute(2,1,0)
            hidden_states = self.act(self.conv1d(hidden_states)[..., :seq_len])         # [batch, intermediate_size, seq_len]

          scan_output = (hidden_states * self.D[None, :, None])
          scan_output = (scan_output * self.act(gate))
          if cache_params is not None:
              cache_params.ssm_states[self.layer_idx].copy_(ssm_state)

          # 4. Final linear projection
          contextualized_states = self.out_proj(scan_output.transpose(1, 2))             # [batch, seq_len, hidden_size]
          return contextualized_states

        elif self.training and cache_params is None:  # Doesn't support outputting the states -> used for training
            contextualized_states = mamba_inner_fn(
                projected_states,
                self.conv1d.weight,
                self.conv1d.bias if self.use_conv_bias else None,
                x_proj.weight,
                self.dt_proj.weight,
                self.out_proj.weight,
                self.out_proj.bias.float() if self.use_bias else None,
                -torch.exp(self.A_log.float()),
                None,  # input-dependent B
                None,  # input-dependent C
                self.D.float(),
                delta_bias=self.dt_proj.bias.float(),
                delta_softplus=True,
            )

        else:
            hidden_states, gate = projected_states.chunk(2, dim=1)
            conv_weights = self.conv1d.weight.view(self.conv1d.weight.size(0), self.conv1d.weight.size(2))

              # print("NON ZERO", hidden_states.shape)
              # 2. Convolution sequence transformation
            if cache_params is not None and cache_params.seqlen_offset > 0:
                hidden_states = causal_conv1d_update(
                    hidden_states.squeeze(-1),
                    cache_params.conv_states[self.layer_idx],
                    conv_weights,
                    self.conv1d.bias,
                    self.activation,
                )
                hidden_states = hidden_states.unsqueeze(-1)
            else:
                if cache_params is not None:
                    conv_states = nn.functional.pad(
                        hidden_states, (self.conv_kernel_size - hidden_states.shape[-1], 0)
                    )
                    # print(conv_states)
                    cache_params.conv_states[self.layer_idx].copy_(conv_states)

                hidden_states = causal_conv1d_fn(
                    hidden_states, conv_weights, self.conv1d.bias, activation=self.activation
                )
            # 3. State Space Model sequence transformation
            # 3.a. input varying initialization of time_step, B and C
            ssm_parameters = x_proj(hidden_states.transpose(1, 2))
            time_step, B, C = torch.split(
                ssm_parameters, [self.time_step_rank, self.ssm_state_size, self.ssm_state_size], dim=-1
            )
            discrete_time_step = self.dt_proj.weight @ time_step.transpose(1, 2)

            A = -torch.exp(self.A_log.float())
            # 3.c perform the recurrence y ← SSM(A, B, C)(x)
            time_proj_bias = self.dt_proj.bias.float() if hasattr(self.dt_proj, "bias") else None

            if cache_params is not None and cache_params.seqlen_offset > 0:
                scan_outputs = selective_state_update(
                    cache_params.ssm_states[self.layer_idx],
                    hidden_states[..., 0],
                    discrete_time_step[..., 0],
                    A,
                    B[:, 0],
                    C[:, 0],
                    self.D,
                    gate[..., 0],
                    time_proj_bias,
                    dt_softplus=True,
                ).unsqueeze(-1)
            else:
                # print("A.shape",A.shape)
                # print("hidden_states", hidden_states.shape)
                # print("discrete_time_step", discrete_time_step.shape)
                # print("GATE.SHAOE", gate.shape)

                scan_outputs, ssm_state = selective_scan_fn(
                    hidden_states,
                    discrete_time_step,
                    A,
                    B.transpose(1, 2),
                    C.transpose(1, 2),
                    self.D.float(),
                    gate,
                    time_proj_bias,
                    delta_softplus=True,
                    return_last_state=True,
                )
                # print("SCANOUTPUTS | SSMSTATE", scan_outputs.shape, ssm_state.shape)
                if ssm_state is not None and cache_params is not None:
                    cache_params.ssm_states[self.layer_idx].copy_(ssm_state)

            # 4. Final linear projection
            contextualized_states = self.out_proj(scan_outputs.transpose(1, 2))
        return contextualized_states

    # fmt: off
    def slow_forward(self, input_states, x_proj, cache_params: Optional[MoSMambaCache]=None):
        batch_size, seq_len, _ = input_states.shape
        dtype = input_states.dtype
        # 1. Gated MLP's linear projection
        projected_states = self.in_proj(input_states).transpose(1, 2)                   # [batch, 2 * intermediate_size, seq_len]
        hidden_states, gate = projected_states.chunk(2, dim=1)

        # 2. Convolution sequence transformation
        if cache_params is not None:
            ssm_state = cache_params.ssm_states[self.layer_idx].clone()
            if cache_params.seqlen_offset > 0:
                conv_state = cache_params.conv_states[self.layer_idx]                   # [batch, intermediate_size, conv_kernel_size]
                conv_state = torch.roll(conv_state, shifts=-1, dims=-1)
                conv_state[:, :, -1] = hidden_states[:, :, 0]
                cache_params.conv_states[self.layer_idx].copy_(conv_state)
                hidden_states = torch.sum(conv_state * self.conv1d.weight[:, 0, :], dim=-1)
                if self.use_conv_bias:
                    hidden_states += self.conv1d.bias
                hidden_states = self.act(hidden_states).to(dtype).unsqueeze(-1)         # [batch, intermediate_size, 1] : decoding
            else:
                conv_state = nn.functional.pad(
                    hidden_states,
                    (self.conv_kernel_size - hidden_states.shape[-1], 0)
                )
                cache_params.conv_states[self.layer_idx].copy_(conv_state)
                if hidden_states.shape[-1] == 0:
                  hidden_states = hidden_states.permute(2,1,0)
                hidden_states = self.act(self.conv1d(hidden_states)[..., :seq_len])     # [batch, intermediate_size, seq_len]
        else:
            ssm_state = torch.zeros(
                (batch_size, self.intermediate_size, self.ssm_state_size),
                device=hidden_states.device, dtype=dtype
            )
            # print(hidden_states.shape)
            # print(self.conv1d)
            if hidden_states.shape[-1] == 0:
              hidden_states = hidden_states.permute(2,1,0)
            hidden_states = self.act(self.conv1d(hidden_states)[..., :seq_len])         # [batch, intermediate_size, seq_len]

        # 3. State Space Model sequence transformation
        # 3.a. Selection:  [batch, seq_len, self.time_step_rank + self.ssm_state_size * 2]
        ssm_parameters = x_proj(hidden_states.transpose(1, 2))
        time_step, B, C = torch.split(
            ssm_parameters, [self.time_step_rank, self.ssm_state_size, self.ssm_state_size], dim=-1
        )
        discrete_time_step = self.dt_proj(time_step)                                    # [batch, seq_len, intermediate_size]
        discrete_time_step = nn.functional.softplus(discrete_time_step).transpose(1, 2) # [batch, intermediate_size, seq_len]

        # 3.b. Discretization: B and C to [batch, seq_len, intermediate_size, ssm_state_size] (SRAM)
        A = -torch.exp(self.A_log.float())                                              # [intermediate_size, ssm_state_size]
        discrete_A = torch.exp(A[None, :, None, :] * discrete_time_step[:, :, :, None]) # [batch, intermediate_size, seq_len, ssm_state_size]
        discrete_B = discrete_time_step[:, :, :, None] * B[:, None, :, :].float()       # [batch, intermediade_size, seq_len, ssm_state_size]
        deltaB_u = discrete_B * hidden_states[:, :, :, None].float()

        # 3.c perform the recurrence y ← SSM(A, B, C)(x)
        scan_outputs = []
        for i in range(seq_len):
            ssm_state = discrete_A[:, :, i, :] * ssm_state + deltaB_u[:, :, i, :]      # [batch, intermediade_size, ssm_state]
            scan_output = torch.matmul(ssm_state.to(dtype), C[:, i, :].unsqueeze(-1))  # [batch, intermediade_size, 1]
            scan_outputs.append(scan_output[:, :, 0])
        # print(scan_outputs)
        scan_output = torch.stack(scan_outputs, dim=-1) if scan_outputs else torch.tensor(scan_outputs)                            # [batch, seq_len, intermediade_size]
        scan_output = scan_output + (hidden_states * self.D[None, :, None])
        scan_output = (scan_output * self.act(gate))

        if cache_params is not None:
            cache_params.ssm_states[self.layer_idx].copy_(ssm_state)

        # 4. Final linear projection
        contextualized_states = self.out_proj(scan_output.transpose(1, 2))             # [batch, seq_len, hidden_size]
        return contextualized_states

    def forward(self, hidden_states, cache_params: Optional[MoSMambaCache] = None):
        batch_size, sequence_length, hidden_dim = hidden_states.shape
        
        if self.training and self.jitter_noise > 0:
            hidden_states *= torch.empty_like(hidden_states).uniform_(1.0 - self.jitter_noise, 1.0 + self.jitter_noise)

#         print('BATCH_SIZE | SEQ LENGTH | HID DIM:',batch_size, sequence_length, hidden_dim)

        hidden_states = hidden_states.view(-1, hidden_dim)

        router_logits = self.gate(hidden_states)

#         print("ROUTER LOGITS:", router_logits, router_logits.size())

        routing_weights = F.softmax(router_logits, dim=1, dtype=torch.float)
        routing_weights, selected_experts = torch.topk(routing_weights, self.top_k, dim=-1)
        # print("ROUTING WEIGHTS", routing_weights, routing_weights.shape)
        # print("SEL EXPERTS", selected_experts, selected_experts.shape)
        routing_weights /= routing_weights.sum(dim=-1, keepdim=True)
        # we cast back to the input dtype
        routing_weights = routing_weights.to(hidden_states.dtype)

#         print(routing_weights .shape)

        final_hidden_states = torch.zeros(
            (batch_size * sequence_length, hidden_dim), dtype=hidden_states.dtype, device=hidden_states.device
        )

        # One hot encode the selected experts to create an expert mask
        # this will be used to easily index which expert is going to be sollicitated
        expert_mask = torch.nn.functional.one_hot(selected_experts, num_classes=self.num_selectivities).permute(2, 1, 0)
        # print("EXPERT MASK", expert_mask, expert_mask.shape)

        # Loop over all available experts in the model and perform the computation on each expert
        for expert_idx in range(self.num_selectivities):
            # expert_layer = self.x_proj[expert_idx]
            expert_layer = self.x_proj.get_submodule(f"w{expert_idx}")
#             expert_layer = getattr(self, f'x_proj_{expert_idx}')
            idx, top_x = torch.where(expert_mask[expert_idx])
#             print("expert_mask[expert_idx]:",expert_mask[expert_idx], expert_mask[expert_idx].shape)


            # Index the correct hidden states and compute the expert hidden state for
            # the current expert. We need to make sure to multiply the output hidden
            # states by `routing_weights` on the corresponding tokens (top-1 and top-2)
#             print("TOP_x:",top_x)
#             print("TOP X.SHAPE:",top_x.shape)
#             print("HIDDEN STATES.SHAPE:",hidden_states.shape)
#             print("HIDDEN STATES[NONE, TOPX].SHAPE:", hidden_states[None, top_x].shape)


#             print("TOP_X | IDX", top_x, idx)

            current_state = hidden_states[None, top_x]
#             print("TOPX", top_x,top_x.shape)
#             print("CURRENT_STATE",current_state.shape)
            current_state = current_state.reshape(-1, hidden_dim)#.reshape(batch_size, sequence_length, hidden_dim )

#             if current_state.shape[1] == 0:
#                 continue


#             print("CURRENT_STATE",current_state)

#             current_state = hidden_states.reshape(batch_size, sequence_length, hidden_dim )

#             print(current_state.shape)
#             if current_state.shape[0] < 1:
#                 print(current_state)
#                 current_state = current_state.reshape(batch_size, 1, hidden_dim)
#             else:
#                 current_state = current_state.reshape(batch_size, sequence_length, hidden_dim)

            # print("current_state.shape", current_state.shape, "ROUTING WEIGHTS",routing_weights[top_x, idx, None].shape)

            current_state = current_state * routing_weights[top_x, idx, None]

            # print("current_hidden_states.shape", current_state.shape)

            current_hidden_states = current_state[None]




#             print("current_hidden_states[none].shape", current_hidden_states.shape)

            if current_hidden_states.shape[1] != 0:

                if is_fast_path_available and "cuda" in expert_layer.weight.device.type:
                # if is_fast_path_available and "cuda" in expert_layer.w2.weight.device.type:
                    current_hidden_states = self.cuda_kernels_forward(current_hidden_states, expert_layer, cache_params) * routing_weights[top_x, idx, None]
                else:
                    current_hidden_states = self.slow_forward(current_hidden_states, expert_layer, cache_params) * routing_weights[top_x, idx, None]
    #             else:
#                 expert_layer.grad = torch.zeros_like(expert_layer.weight)
#             current_hidden_states = expert_layer(current_state)

            current_hidden_states = current_hidden_states.reshape(-1, hidden_dim)
#             print(current_hidden_states.shape, final_hidden_states.shape)

            # However `index_add_` only support torch tensors for indexing so we'll use
            # the `top_x` tensor here.
            final_hidden_states.index_add_(0, top_x, current_hidden_states.to(hidden_states.dtype))
        final_hidden_states = final_hidden_states.reshape(batch_size, sequence_length, hidden_dim)

        return final_hidden_states, router_logits


class MoSMambaRMSNorm(nn.Module):
    def __init__(self, hidden_size, eps=1e-6):
        """
        MoSMambaRMSNorm is equivalent to T5LayerNorm and LlamaRMSNorm
        """
        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)


class MoSMambaBlock(nn.Module):
    def __init__(self, config, layer_idx):
        super().__init__()
        self.config = config
        self.layer_idx = layer_idx
        self.residual_in_fp32 = config.residual_in_fp32
        self.norm = MoSMambaRMSNorm(config.hidden_size, eps=config.layer_norm_epsilon)
        self.mixer = MoSMambaMixer(config, layer_idx=layer_idx)

    def forward(self, hidden_states, cache_params: Optional[MoSMambaCache] = None, output_router_logits:Optional[bool] = False):
        residual = hidden_states
        hidden_states = self.norm(hidden_states.to(dtype=self.norm.weight.dtype))
        if self.residual_in_fp32:
            residual = residual.to(torch.float32)

        hidden_states, router_logits = self.mixer(hidden_states, cache_params=cache_params)
        hidden_states = residual + hidden_states
        outputs = (hidden_states,)

        if output_router_logits:
            outputs += (router_logits,)
        return outputs


class MoSMambaPreTrainedModel(PreTrainedModel):
    """
    An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
    models.
    """

    config_class = MoSMambaConfig
    base_model_prefix = "backbone"
    _no_split_modules = ["MoSMambaBlock"]
    supports_gradient_checkpointing = True

    def make_tensors_contiguous(self):
        for name, param in self.named_parameters():
            if not param.is_contiguous():
                param.data = param.data.contiguous()

    def save_pretrained(self, save_directory, **kwargs):
        # Make tensors contiguous
        self.make_tensors_contiguous()

        # Call the original save_pretrained method
        super().save_pretrained(save_directory, **kwargs)

    def _init_weights(self, module):
        """Initialize the weights."""
        if isinstance(module, MoSMambaMixer):
            module.A_log._no_weight_decay = True
            module.D._no_weight_decay = True

            dt_init_std = self.config.time_step_rank**-0.5 * self.config.time_step_scale
            if self.config.time_step_init_scheme == "constant":
                nn.init.constant_(module.dt_proj.weight, dt_init_std)
            elif self.config.time_step_init_scheme == "random":
                nn.init.uniform_(module.dt_proj.weight, -dt_init_std, dt_init_std)

            nn.init.xavier_uniform_(module.gate.weight, gain=0.1)

            dt = torch.exp(
                torch.rand(self.config.intermediate_size)
                * (math.log(self.config.time_step_max) - math.log(self.config.time_step_min))
                + math.log(self.config.time_step_min)
            ).clamp(min=self.config.time_step_floor)
            # # Inverse of softplus: https://github.com/pytorch/pytorch/issues/72759
            inv_dt = dt + torch.log(-torch.expm1(-dt))
            with torch.no_grad():
                module.dt_proj.bias.copy_(inv_dt)
            module.dt_proj.bias._no_reinit = True

        if isinstance(module, nn.ModuleList):
            for i in range(len(module)):
                try:
                    nn.init.xavier_uniform_(module.get_submodule(f"w{i}").weight, gain=0.1)
                    print("INIT WEIGHTS FOR X_PROJ")
                    # for :
                    #     if isinstance(submodule, nn.Linear):
                    #         nn.init.xavier_uniform_(submodule.weight, gain=0.1)
                    #         print("INIT WEIGHTS FOR X_PROJ")
                except:
                    pass

        if isinstance(module, nn.Linear):
            if module.bias is not None:
                if not getattr(module.bias, "_no_reinit", False):
                    nn.init.zeros_(module.bias)
            print("Init Weight.... | Module name:", module)
            # nn.init.uniform_(module.weight, -0.001, 0.001)
            
        elif isinstance(module, nn.Embedding):
            nn.init.normal_(module.weight, std=self.config.initializer_range)

        if self.config.rescale_prenorm_residual:
            # Reinitialize selected weights subject to the OpenAI GPT-2 Paper Scheme:
            #   > A modified initialization which accounts for the accumulation on the residual path with model depth. Scale
            #   > the weights of residual layers at initialization by a factor of 1/√N where N is the # of residual layers.
            #   >   -- GPT-2 :: https://openai.com/blog/better-language-models/
            #
            # Reference (Megatron-LM): https://github.com/NVIDIA/Megatron-LM/blob/main/megatron/model/gpt_model.py
            for name, p in module.named_parameters():
                if name in ["out_proj.weight"]:
                    # Special Scaled Initialization --> There are 2 Layer Norms per Transformer Block
                    # Following Pytorch init, except scale by 1/sqrt(2 * n_layer)
                    # We need to reinit p since this code could be called multiple times
                    # Having just p *= scale would repeatedly scale it down
                    nn.init.kaiming_uniform_(p, a=math.sqrt(5))
                    with torch.no_grad():
                        p /= math.sqrt(self.config.num_layers)


@dataclass
class MoSMambaOutput(ModelOutput):
    """
    Class for the MAMBA model outputs.

    Args:
        last_hidden_state (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`):
            Sequence of hidden-states at the output of the last layer of the model.
        cache_params (`MoSMambaCache`):
            The state of the model at the last time step. Can be used in a forward method with the next `input_ids` to
            avoid providing the old `input_ids`.

            Includes both the State space model state matrices after the selective scan, and the Convolutional states
        hidden_states (`tuple(torch.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
            Tuple of `torch.FloatTensor` (one for the output of the embeddings, if the model has an embedding layer, +
            one for the output of each layer) of shape `(batch_size, sequence_length, hidden_size)`.

            Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
    """

    last_hidden_state: Optional[torch.FloatTensor] = None
    cache_params: Optional[MoSMambaCache] = None
    hidden_states: Optional[Tuple[torch.FloatTensor]] = None
    router_logits: Optional[Tuple[torch.FloatTensor]] = None


@dataclass
class MoSMambaCausalLMOutput(ModelOutput):
    """
    Base class for causal language model (or autoregressive) outputs.

    Args:
        loss (`torch.FloatTensor` of shape `(1,)`, *optional*, returned when `labels` is provided):
            Language modeling loss (for next-token prediction).
        logits (`torch.FloatTensor` of shape `(batch_size, sequence_length, config.vocab_size)`):
            Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).
        cache_params (`MoSMambaCache`):
            The state of the model at the last time step. Can be used in a forward method with the next `input_ids` to
            avoid providing the old `input_ids`.

            Includes both the State space model state matrices after the selective scan, and the Convolutional states
        hidden_states (`tuple(torch.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
            Tuple of `torch.FloatTensor` (one for the output of the embeddings, if the model has an embedding layer, +
            one for the output of each layer) of shape `(batch_size, sequence_length, hidden_size)`.

            Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
    """

    loss: Optional[torch.FloatTensor] = None
    logits: Optional[torch.FloatTensor] = None
    cache_params: Optional[MoSMambaCache] = None
    hidden_states: Optional[Tuple[torch.FloatTensor]] = None
    router_logits: Optional[Tuple[torch.FloatTensor]] = None


class MoSMambaModel(MoSMambaPreTrainedModel):
    def __init__(self, config):
        super().__init__(config)

        self.embeddings = nn.Embedding(config.vocab_size, config.hidden_size)
        self.layers = nn.ModuleList([MoSMambaBlock(config, layer_idx=idx) for idx in range(config.num_hidden_layers)])

        self.gradient_checkpointing = False
        self.norm_f = MoSMambaRMSNorm(config.hidden_size, eps=config.layer_norm_epsilon)
        # Initialize weights and apply final processing
        self._register_load_state_dict_pre_hook(self.load_hook)
        self.post_init()
        self.config.output_router_logits = True

    def load_hook(self, state_dict, prefix, *args):
        for k in state_dict:
            if "embedding." in k:
                state_dict[k.replace("embedding.", "embeddings.")] = state_dict.pop(k)
                break

    def get_input_embeddings(self):
        return self.embeddings

    def set_input_embeddings(self, new_embeddings):
        self.embeddings = new_embeddings

    def forward(
        self,
        input_ids: Optional[torch.LongTensor] = None,
        inputs_embeds: Optional[torch.LongTensor] = None,
        cache_params: Optional[MoSMambaCache] = None,
        use_cache: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        output_router_logits: Optional[bool] = None,
        return_dict: Optional[bool] = None,
        **kwargs,  # `attention_mask` is passed by the tokenizer and we don't want it
    ) -> Union[Tuple, MoSMambaOutput]:
        output_hidden_states = (
            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
        )
        output_router_logits = (
            output_router_logits if output_router_logits is not None else self.config.output_router_logits
        )
        use_cache = use_cache if use_cache is not None else (self.config.use_cache if not self.training else False)
        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):  # ^ is python for xor
            raise ValueError(
                "You cannot specify both input_ids and inputs_embeds at the same time, and must specify either one"
            )

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

        if self.gradient_checkpointing and self.training and use_cache:
            use_cache = False

        if cache_params is None and use_cache:
            cache_params = MoSMambaCache(
                self.config, inputs_embeds.size(0), device=inputs_embeds.device, dtype=inputs_embeds.dtype
            )

        hidden_states = inputs_embeds
        all_hidden_states = () if output_hidden_states else None
        all_router_logits = () if output_router_logits else None
        for mixer_block in self.layers:
            if self.gradient_checkpointing and self.training:
                layer_outputs = self._gradient_checkpointing_func(mixer_block.__call__, hidden_states, cache_params, output_router_logits)
            else:
                layer_outputs = mixer_block(hidden_states, cache_params=cache_params,output_router_logits=output_router_logits)

            hidden_states = layer_outputs[0]

            if output_hidden_states:
                all_hidden_states = all_hidden_states + (hidden_states,)

            if output_router_logits:
                all_router_logits += (layer_outputs[-1],)

        if use_cache:
            cache_params.seqlen_offset += inputs_embeds.shape[1]

        hidden_states = self.norm_f(hidden_states)

        if output_hidden_states:
            all_hidden_states = all_hidden_states + (hidden_states,)


        if not return_dict:
            return tuple(v for v in [hidden_states, cache_params, all_hidden_states, all_router_logits] if v is not None)

        return MoSMambaOutput(
            last_hidden_state=hidden_states,
            cache_params=cache_params if use_cache else None,
            hidden_states=all_hidden_states,
            router_logits=all_router_logits,
        )


class MoSMambaForCausalLM(MoSMambaPreTrainedModel):
    _tied_weights_keys = ["lm_head.weight"]

    def __init__(self, config):
        super().__init__(config)
        self.backbone = MoSMambaModel(config)
        self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
        self.num_selectivities = config.num_selectivities
        self.num_selectivities_per_tok = config.num_selectivities_per_tok
        self.router_aux_loss_coef = 0.02
        # Initialize weights and apply final processing
        self.post_init()

    def get_output_embeddings(self):
        return self.lm_head

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

    def get_input_embeddings(self):
        return self.backbone.get_input_embeddings()

    def set_input_embeddings(self, new_embeddings):
        return self.backbone.set_input_embeddings(new_embeddings)

    def _update_model_kwargs_for_generation(
        self, outputs: ModelOutput, model_kwargs: Dict[str, Any], **kwargs
    ) -> Dict[str, Any]:
        model_kwargs["cache_params"] = outputs.get("cache_params", None)
        return model_kwargs

    def prepare_inputs_for_generation(
        self, input_ids, cache_params: Optional[MoSMambaCache] = None, inputs_embeds=None, attention_mask=None, output_router_logits=False, **kwargs
    ):
        # only last token for inputs_ids if the state is passed along.
        if cache_params is not None:
            input_ids = input_ids[:, -1].unsqueeze(-1)

        if inputs_embeds is not None and cache_params is None:
            model_inputs = {"inputs_embeds": inputs_embeds}
        else:
            model_inputs = {"input_ids": input_ids}

        model_inputs["cache_params"] = cache_params
        model_inputs['output_router_logits'] = output_router_logits
        return model_inputs


    def forward(
        self,
        input_ids: Optional[torch.LongTensor] = None,
        inputs_embeds: Optional[torch.FloatTensor] = None,
        cache_params: Optional[MoSMambaCache] = None,
        labels: Optional[torch.LongTensor] = None,
        output_hidden_states: Optional[bool] = None,
        output_router_logits: Optional[bool] = None,
        return_dict: Optional[bool] = None,
        use_cache: Optional[bool] = None,
        **kwargs,  # for now we need this for generation
    ) -> Union[Tuple, MoSMambaCausalLMOutput]:
        r"""
        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
            Labels for language modeling. Note that the labels **are shifted** inside the model, i.e. you can set
            `labels = input_ids` Indices are selected in `[-100, 0, ..., config.vocab_size]` All labels set to `-100`
            are ignored (masked), the loss is only computed for labels in `[0, ..., config.vocab_size]`
        """
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        output_router_logits = (
            output_router_logits if output_router_logits is not None else self.config.output_router_logits
        )

        mamba_outputs = self.backbone(
            input_ids,
            cache_params=cache_params,
            inputs_embeds=inputs_embeds,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
            use_cache=use_cache,
        )
        hidden_states = mamba_outputs[0]

        logits = self.lm_head(hidden_states.to(self.lm_head.weight.dtype)).float()

        loss = None
        if labels is not None:
            # move labels to correct device to enable model parallelism
            labels = labels.to(logits.device)
            # Shift so that tokens < n predict n
            shift_logits = logits[..., :-1, :].contiguous()
            shift_labels = labels[..., 1:].contiguous()
            # Flatten the tokens
            loss_fct = CrossEntropyLoss()
            loss = loss_fct(shift_logits.view(-1, shift_logits.size(-1)), shift_labels.view(-1))

        aux_loss = None
        if output_router_logits:
            aux_loss = load_balancing_loss_func(
                mamba_outputs.router_logits if return_dict else mamba_outputs[-1],
                self.num_selectivities,
                self.num_selectivities_per_tok,
#                 attention_mask,
            )
            if labels is not None:
                loss += self.router_aux_loss_coef * aux_loss.to(loss.device)  # make sure to reside in the same device

#         print("AUX LOSS:", aux_loss)
#         print("LOSS:", loss)

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

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

        return MoSMambaCausalLMOutput(
            loss=loss,
            logits=logits,
            cache_params=mamba_outputs.cache_params,
            hidden_states=mamba_outputs.hidden_states,
            router_logits=mamba_outputs.router_logits,
        )