modeling_rasphi.py

import torch
import torch.nn as nn
from transformers.modeling_outputs import MoeModelOutputWithPast, MoeCausalLMOutputWithPast
from transformers.modeling_utils import PreTrainedModel, PretrainedConfig
from torch.nn import CrossEntropyLoss
from typing import Optional, Tuple, Union, List
import torch.nn.functional as F
import math

ACT2FN = {
    "relu": F.relu,
    "silu": F.silu,
    "gelu": F.gelu,
    "tanh": torch.tanh,
    "sigmoid": torch.sigmoid,
}

class RasphiDecoderLayer(nn.Module):
    def __init__(self, config: RasphiConfig, layer_idx: int):
        super().__init__()
        self.layer_idx = layer_idx
        self.hidden_size = config.hidden_size
        self.reasoning_hidden_size = config.reasoning_hidden_size
        self.content_hidden_size = config.content_hidden_size

        # Attention layers
        self.reasoning_self_attn = RasphiAttention(config, self.reasoning_hidden_size, layer_idx)
        self.content_self_attn = RasphiAttention(config, self.content_hidden_size, layer_idx)

        # MoE layers
        self.reasoning_moe = RasphiSparseMoeBlock(config, is_reasoning=True)
        self.content_moe = RasphiSparseMoeBlock(config, is_reasoning=False)

        # Layer norms
        self.reasoning_input_layernorm = nn.LayerNorm(self.reasoning_hidden_size, eps=config.rms_norm_eps)
        self.reasoning_post_attention_layernorm = nn.LayerNorm(self.reasoning_hidden_size, eps=config.rms_norm_eps)
        self.content_input_layernorm = nn.LayerNorm(self.content_hidden_size, eps=config.rms_norm_eps)
        self.content_post_attention_layernorm = nn.LayerNorm(self.content_hidden_size, eps=config.rms_norm_eps)

        # Stream interaction
        self.stream_interaction = config.stream_interaction
        if self.stream_interaction in ["attention", "both"]:
            self.reasoning_to_content_attn = RasphiAttention(config, self.content_hidden_size, layer_idx)
            self.content_to_reasoning_attn = RasphiAttention(config, self.reasoning_hidden_size, layer_idx)
        if self.stream_interaction in ["mlp", "both"]:
            self.reasoning_to_content_mlp = nn.Linear(self.reasoning_hidden_size, self.content_hidden_size)
            self.content_to_reasoning_mlp = nn.Linear(self.content_hidden_size, self.reasoning_hidden_size)

    def forward(
        self,
        reasoning_hidden_states: torch.Tensor,
        content_hidden_states: torch.Tensor,
        attention_mask: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.LongTensor] = None,
        past_key_value: Optional[Tuple[torch.Tensor]] = None,
        output_attentions: Optional[bool] = False,
        output_router_logits: Optional[bool] = False,
        use_cache: Optional[bool] = False,
    ) -> Tuple[torch.FloatTensor, ...]:
        # Self Attention for both streams
        reasoning_residual = reasoning_hidden_states
        content_residual = content_hidden_states

        reasoning_hidden_states = self.reasoning_input_layernorm(reasoning_hidden_states)
        content_hidden_states = self.content_input_layernorm(content_hidden_states)

        reasoning_self_attn_output, reasoning_self_attn_weights, reasoning_present_key_value = self.reasoning_self_attn(
            hidden_states=reasoning_hidden_states,
            attention_mask=attention_mask,
            position_ids=position_ids,
            past_key_value=past_key_value[0] if past_key_value is not None else None,
            output_attentions=output_attentions,
            use_cache=use_cache,
        )

        content_self_attn_output, content_self_attn_weights, content_present_key_value = self.content_self_attn(
            hidden_states=content_hidden_states,
            attention_mask=attention_mask,
            position_ids=position_ids,
            past_key_value=past_key_value[1] if past_key_value is not None else None,
            output_attentions=output_attentions,
            use_cache=use_cache,
        )

        reasoning_hidden_states = reasoning_residual + reasoning_self_attn_output
        content_hidden_states = content_residual + content_self_attn_output

        # Stream Interaction
        if self.stream_interaction in ["attention", "both"]:
            reasoning_to_content, _, _ = self.reasoning_to_content_attn(
                hidden_states=content_hidden_states,
                attention_mask=attention_mask,
                position_ids=position_ids,
                past_key_value=None,
                output_attentions=False,
                use_cache=False,
                key_value_states=reasoning_hidden_states,
            )
            content_to_reasoning, _, _ = self.content_to_reasoning_attn(
                hidden_states=reasoning_hidden_states,
                attention_mask=attention_mask,
                position_ids=position_ids,
                past_key_value=None,
                output_attentions=False,
                use_cache=False,
                key_value_states=content_hidden_states,
            )
            reasoning_hidden_states = reasoning_hidden_states + content_to_reasoning
            content_hidden_states = content_hidden_states + reasoning_to_content

        if self.stream_interaction in ["mlp", "both"]:
            reasoning_to_content = self.reasoning_to_content_mlp(reasoning_hidden_states)
            content_to_reasoning = self.content_to_reasoning_mlp(content_hidden_states)
            reasoning_hidden_states = reasoning_hidden_states + content_to_reasoning
            content_hidden_states = content_hidden_states + reasoning_to_content

        # MoE for both streams
        reasoning_residual = reasoning_hidden_states
        content_residual = content_hidden_states

        reasoning_hidden_states = self.reasoning_post_attention_layernorm(reasoning_hidden_states)
        content_hidden_states = self.content_post_attention_layernorm(content_hidden_states)

        reasoning_moe_output, reasoning_router_logits = self.reasoning_moe(reasoning_hidden_states)
        content_moe_output, content_router_logits = self.content_moe(content_hidden_states)

        reasoning_hidden_states = reasoning_residual + reasoning_moe_output
        content_hidden_states = content_residual + content_moe_output

        outputs = (reasoning_hidden_states, content_hidden_states)

        if use_cache:
            outputs += ((reasoning_present_key_value, content_present_key_value),)
        if output_attentions:
            outputs += (reasoning_self_attn_weights, content_self_attn_weights)
        if output_router_logits:
            outputs += (reasoning_router_logits, content_router_logits)

        return outputs

class RasphiModel(PreTrainedModel):
    config_class = RasphiConfig
    base_model_prefix = "model"
    supports_gradient_checkpointing = True
    _no_split_modules = ["RasphiDecoderLayer"]
    _skip_keys_device_placement = "past_key_values"
    _supports_flash_attn_2 = True
    _supports_sdpa = True
    _supports_cache_class = True

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

        self.reasoning_embed_tokens = nn.Embedding(config.vocab_size, config.reasoning_hidden_size, self.padding_idx)
        self.content_embed_tokens = nn.Embedding(config.vocab_size, config.content_hidden_size, self.padding_idx)
        
        self.layers = nn.ModuleList([RasphiDecoderLayer(config, layer_idx) for layer_idx in range(config.num_hidden_layers)])
        
        self.reasoning_norm = nn.LayerNorm(config.reasoning_hidden_size, eps=config.rms_norm_eps, elementwise_affine=True)
        self.content_norm = nn.LayerNorm(config.content_hidden_size, eps=config.rms_norm_eps, elementwise_affine=True)

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

    def get_input_embeddings(self):
        return (self.reasoning_embed_tokens, self.content_embed_tokens)

    def set_input_embeddings(self, value):
        self.reasoning_embed_tokens = value[0]
        self.content_embed_tokens = value[1]

    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,
        use_cache: Optional[bool] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        output_router_logits: Optional[bool] = None,
        return_dict: Optional[bool] = None,
    ) -> Union[Tuple, MoeModelOutputWithPast]:
        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
        output_router_logits = (
            output_router_logits if output_router_logits is not None else self.config.output_router_logits
        )
        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 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")

        if inputs_embeds is None:
            reasoning_inputs_embeds = self.reasoning_embed_tokens(input_ids)
            content_inputs_embeds = self.content_embed_tokens(input_ids)
        else:
            reasoning_inputs_embeds = inputs_embeds[:, :, :self.config.reasoning_hidden_size]
            content_inputs_embeds = inputs_embeds[:, :, self.config.reasoning_hidden_size:]

        reasoning_hidden_states = reasoning_inputs_embeds
        content_hidden_states = content_inputs_embeds

        # decoder layers
        all_reasoning_hidden_states = () if output_hidden_states else None
        all_content_hidden_states = () if output_hidden_states else None
        all_reasoning_self_attns = () if output_attentions else None
        all_content_self_attns = () if output_attentions else None
        all_reasoning_router_logits = () if output_router_logits else None
        all_content_router_logits = () if output_router_logits else None
        next_decoder_cache = None

        for decoder_layer in self.layers:
            if output_hidden_states:
                all_reasoning_hidden_states += (reasoning_hidden_states,)
                all_content_hidden_states += (content_hidden_states,)

            layer_outputs = decoder_layer(
                reasoning_hidden_states,
                content_hidden_states,
                attention_mask=attention_mask,
                position_ids=position_ids,
                past_key_value=past_key_values,
                output_attentions=output_attentions,
                output_router_logits=output_router_logits,
                use_cache=use_cache,
            )

            reasoning_hidden_states = layer_outputs[0]
            content_hidden_states = layer_outputs[1]

            if use_cache:
                next_decoder_cache = layer_outputs[2 if output_attentions else 1]

            if output_attentions:
                all_reasoning_self_attns += (layer_outputs[2],)
                all_content_self_attns += (layer_outputs[3],)

            if output_router_logits:
                all_reasoning_router_logits += (layer_outputs[-2],)
                all_content_router_logits += (layer_outputs[-1],)

        reasoning_hidden_states = self.reasoning_norm(reasoning_hidden_states)
        content_hidden_states = self.content_norm(content_hidden_states)

        # add hidden states from the last decoder layer
        if output_hidden_states:
            all_reasoning_hidden_states += (reasoning_hidden_states,)
            all_content_hidden_states += (content_hidden_states,)

        next_cache = None
        if use_cache:
            next_cache = next_decoder_cache

        if not return_dict:
            return tuple(
                v
                for v in [reasoning_hidden_states, content_hidden_states, next_cache, all_reasoning_hidden_states, 
                          all_content_hidden_states, all_reasoning_self_attns, all_content_self_attns, 
                          all_reasoning_router_logits, all_content_router_logits]
                if v is not None
            )

        return MoeModelOutputWithPast(
            last_hidden_state=(reasoning_hidden_states, content_hidden_states),
            past_key_values=next_cache,
            hidden_states=(all_reasoning_hidden_states, all_content_hidden_states),
            attentions=(all_reasoning_self_attns, all_content_self_attns),
            router_logits=(all_reasoning_router_logits, all_content_router_logits),
        )

class RasphiSparseMoeBlock(nn.Module):
    def __init__(self, config: RasphiConfig, is_reasoning: bool):
        super().__init__()
        self.hidden_dim = config.reasoning_hidden_size if is_reasoning else config.content_hidden_size
        self.ffn_dim = config.intermediate_size
        self.num_experts = config.num_reasoning_experts if is_reasoning else config.num_content_experts
        self.top_k = config.num_experts_per_tok
        
        # gating
        self.gate = nn.Linear(self.hidden_dim, self.num_experts, bias=False)

        self.experts = nn.ModuleList([RasphiBlockSparseTop2MLP(config, is_reasoning) for _ in range(self.num_experts)])

        # Jitter parameters
        self.router_jitter_noise = config.router_jitter_noise
        self.input_jitter_noise = config.input_jitter_noise
        
    def forward(self, hidden_states: torch.Tensor) -> Tuple[torch.Tensor, torch.Tensor]:
        batch_size, sequence_length, hidden_dim = hidden_states.shape
        if self.training and self.input_jitter_noise > 0:
            hidden_states *= torch.empty_like(hidden_states).uniform_(1.0 - self.input_jitter_noise, 1.0 + self.input_jitter_noise)
        hidden_states = hidden_states.view(-1, hidden_dim)
        
        router_logits = self.gate(hidden_states)

        routing_weights, selected_experts = sparsemixer(
            router_logits, 
            top_k=self.top_k, 
            jitter_eps=self.router_jitter_noise, 
            training=self.training,
        )

        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
        expert_mask = torch.nn.functional.one_hot(selected_experts, num_classes=self.num_experts).permute(2, 1, 0)

        # Loop over all available experts in the model and perform the computation on each expert
        for expert_idx in range(self.num_experts):
            expert_layer = self.experts[expert_idx]
            idx, top_x = torch.where(expert_mask[expert_idx])

            if top_x.shape[0] == 0:
                continue

            # 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)
            current_state = hidden_states[None, top_x.tolist()].reshape(-1, hidden_dim)
            current_hidden_states = expert_layer(current_state) * routing_weights[top_x.tolist(), idx.tolist(), None]

            # Add the expert output to the final hidden states
            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 RasphiBlockSparseTop2MLP(nn.Module):
    def __init__(self, config: RasphiConfig, is_reasoning: bool):
        super().__init__()
        self.ffn_dim = config.intermediate_size
        self.hidden_dim = config.reasoning_hidden_size if is_reasoning else config.content_hidden_size

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

        self.act_fn = ACT2FN[config.hidden_act]

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

class RasphiPreTrainedModel(PreTrainedModel):
    config_class = RasphiConfig
    base_model_prefix = "rasphi"
    supports_gradient_checkpointing = True
    _no_split_modules = ["RasphiDecoderLayer"]

    def _init_weights(self, module):
        std = self.config.initializer_range
        if isinstance(module, nn.Linear):
            module.weight.data.normal_(mean=0.0, std=std)
            if module.bias is not None:
                module.bias.data.zero_()
        elif isinstance(module, nn.Embedding):
            module.weight.data.normal_(mean=0.0, std=std)
            if module.padding_idx is not None:
                module.weight.data[module.padding_idx].zero_()

class RasphiForCausalLM(RasphiPreTrainedModel):
    _tied_weights_keys = ["lm_head.weight"]

    def __init__(self, config):
        super().__init__(config)
        self.model = RasphiModel(config)
        self.vocab_size = config.vocab_size
        self.lm_head = nn.Linear(config.content_hidden_size, config.vocab_size, bias=config.lm_head_bias)
        self.router_aux_loss_coef = config.router_aux_loss_coef
        self.num_experts = config.num_content_experts  # We use content experts for language modeling
        self.num_experts_per_tok = config.num_experts_per_tok

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

    def get_input_embeddings(self):
        return self.model.get_input_embeddings()[1]  # Return content embeddings

    def set_input_embeddings(self, value):
        self.model.set_input_embeddings((self.model.get_input_embeddings()[0], value))

    def get_output_embeddings(self):
        return self.lm_head

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

    def prepare_inputs_for_generation(self, input_ids, past_key_values=None, **kwargs):
        token_type_ids = kwargs.get("token_type_ids", None)
        # only last token for inputs_ids if past is defined in kwargs
        if past_key_values:
            input_ids = input_ids[:, -1].unsqueeze(-1)
            if token_type_ids is not None:
                token_type_ids = token_type_ids[:, -1].unsqueeze(-1)

        attention_mask = kwargs.get("attention_mask", None)
        position_ids = kwargs.get("position_ids", None)

        if attention_mask is not None and position_ids is None:
            # create position_ids on the fly for batch generation
            position_ids = attention_mask.long().cumsum(-1) - 1
            position_ids.masked_fill_(attention_mask == 0, 1)
            if past_key_values:
                position_ids = position_ids[:, -1].unsqueeze(-1)
        else:
            position_ids = None

        return {
            "input_ids": input_ids,
            "past_key_values": past_key_values,
            "use_cache": kwargs.get("use_cache"),
            "position_ids": position_ids,
            "attention_mask": attention_mask,
            "token_type_ids": token_type_ids,
        }

    def forward(
        self,
        input_ids: Optional[torch.LongTensor] = None,
        attention_mask: Optional[torch.FloatTensor] = 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,
        output_router_logits: Optional[bool] = None,
        return_dict: Optional[bool] = None,
    ) -> Union[Tuple, MoeCausalLMOutputWithPast]:
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        outputs = self.model(
            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,
            output_router_logits=output_router_logits,
            return_dict=return_dict,
        )

        hidden_states = outputs[0]
        content_hidden_states = hidden_states[1]  # Use content stream for language modeling
        logits = self.lm_head(content_hidden_states)

        loss = None
        if labels is not None:
            # 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, self.config.vocab_size), shift_labels.view(-1))

        aux_loss = None
        if output_router_logits:
            aux_loss = load_balancing_loss_func(
                outputs.router_logits[1] if return_dict else outputs[-1][1],  # Use content stream router logits
                self.num_experts,
                self.num_experts_per_tok,
                attention_mask,
            )
            if labels is not None:
                loss += self.router_aux_loss_coef * aux_loss

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

        return MoeCausalLMOutputWithPast(
            loss=loss,
            aux_loss=aux_loss,
            logits=logits,
            past_key_values=outputs.past_key_values,
            hidden_states=outputs.hidden_states,
            attentions=outputs.attentions,
            router_logits=outputs.router_logits,
        )

    @staticmethod
    def _reorder_cache(past, beam_idx):
        return tuple(
            tuple(past_state.index_select(0, beam_idx) for past_state in layer_past)
            for layer_past in past
        )


#—Model > Rasphi changes start—#
class RasphiAttention(nn.Module):
    def __init__(self, config: RasphiConfig, hidden_size: int, layer_idx: Optional[int] = None):
        super().__init__()
        self.config = config
        self.layer_idx = layer_idx
        self.hidden_size = hidden_size
        self.num_heads = config.num_attention_heads
        self.head_dim = hidden_size // self.num_heads
        self.num_key_value_heads = config.num_key_value_heads
        self.num_key_value_groups = self.num_heads // self.num_key_value_heads
        self.max_position_embeddings = config.max_position_embeddings
        self.rope_theta = config.rope_theta
        self.is_causal = True
        self.attention_dropout = config.attention_dropout

        if (self.head_dim * self.num_heads) != self.hidden_size:
            raise ValueError(
                f"hidden_size must be divisible by num_heads (got `hidden_size`: {self.hidden_size}"
                f" and `num_heads`: {self.num_heads})."
            )

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

        if getattr(config, 'rope_scaling', None) is None:
            self.rotary_emb = RasphiMoERotaryEmbedding(
                self.head_dim,
                max_position_embeddings=self.max_position_embeddings,
                base=self.rope_theta,
            )
        else:
            scaling_type = self.config.rope_scaling["type"]
            if scaling_type == "linear":
                self.rotary_emb = LinearScalingRotaryEmbedding(
                    self.head_dim,
                    max_position_embeddings=self.max_position_embeddings,
                    scaling_factor=self.config.rope_scaling["factor"],
                    base=self.rope_theta,
                )
            elif scaling_type == "dynamic":
                self.rotary_emb = DynamicNTKScalingRotaryEmbedding(
                    self.head_dim,
                    max_position_embeddings=self.max_position_embeddings,
                    scaling_factor=self.config.rope_scaling["factor"],
                    base=self.rope_theta,
                )
            else:
                raise ValueError(f"Unknown RoPE scaling type {scaling_type}")

    def _shape(self, tensor: torch.Tensor, seq_len: int, bsz: int):
        return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()

    def forward(
        self,
        hidden_states: torch.Tensor,
        attention_mask: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.LongTensor] = None,
        past_key_value: Optional[Tuple[torch.Tensor]] = None,
        output_attentions: bool = False,
        use_cache: bool = False,
        key_value_states: Optional[torch.Tensor] = None,
    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
        bsz, q_len, _ = hidden_states.size()

        query_states = self.q_proj(hidden_states)

        if key_value_states is None:
            # self-attention
            key_states = self.k_proj(hidden_states)
            value_states = self.v_proj(hidden_states)
        else:
            # cross-attention
            key_states = self.k_proj(key_value_states)
            value_states = self.v_proj(key_value_states)
            kv_len = key_value_states.size(1)

        query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
        key_states = key_states.view(bsz, -1, self.num_key_value_heads, self.head_dim).transpose(1, 2)
        value_states = value_states.view(bsz, -1, self.num_key_value_heads, self.head_dim).transpose(1, 2)

        kv_seq_len = key_states.shape[-2]
        if past_key_value is not None:
            kv_seq_len += past_key_value[0].shape[-2]
        cos, sin = self.rotary_emb(value_states, seq_len=kv_seq_len)
        query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin, position_ids)

        if past_key_value is not None:
            # reuse k, v, self_attention
            key_states = torch.cat([past_key_value[0], key_states], dim=2)
            value_states = torch.cat([past_key_value[1], value_states], dim=2)

        past_key_value = (key_states, value_states) if use_cache else None

        # repeat k/v heads if n_kv_heads < n_heads
        key_states = repeat_kv(key_states, self.num_key_value_groups)
        value_states = repeat_kv(value_states, self.num_key_value_groups)

        attn_weights = torch.matmul(query_states, key_states.transpose(2, 3)) / math.sqrt(self.head_dim)

        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_states.dtype)
        attn_weights = nn.functional.dropout(attn_weights, p=self.attention_dropout, training=self.training)
        attn_output = torch.matmul(attn_weights, value_states)

        attn_output = attn_output.transpose(1, 2).contiguous()
        attn_output = attn_output.reshape(bsz, q_len, self.hidden_size)

        attn_output = self.o_proj(attn_output)

        if not output_attentions:
            attn_weights = None

        return attn_output, attn_weights, past_key_value

class mp(torch.autograd.Function):
    @staticmethod
    def forward(
        ctx, 
        scores: torch.Tensor, 
        multiplier: torch.Tensor, 
        selected_experts: torch.Tensor,
        masked_gates: torch.Tensor,
        mask_for_one: torch.Tensor,
    ):
        ctx.save_for_backward(multiplier, selected_experts, masked_gates)
        return multiplier * mask_for_one
        
    @staticmethod
    def backward(
        ctx, 
        grad_at_output: torch.Tensor, 
    ):
        multiplier, selected_experts, masked_gates = ctx.saved_tensors
        
        grad_at_output = grad_at_output * multiplier
        
        grad_at_scores_expaned = masked_gates * grad_at_output.mul(-1)
        grad_at_scores_expaned.scatter_add_(
            dim=-1,
            index=selected_experts,
            src=grad_at_output,
        )
        
        return (
            grad_at_scores_expaned, 
            None, 
            None, 
            None, 
            None, 
        )

def sparsemixer(scores, top_k, jitter_eps, training):
    assert top_k == 2
    
    ################ first expert ################
    
    with torch.no_grad():
        # compute mask for sparsity
        mask_logits_threshold, max_ind = scores.max(dim=-1, keepdim=True)
        factor = scores.abs().clamp(min=mask_logits_threshold)
        mask_logits_threshold = (
            (mask_logits_threshold - scores) / factor
        ) > (2 * jitter_eps)

    # apply mask 
    masked_gates = scores.masked_fill(mask_logits_threshold, float('-inf'))
    if training:
        selected_experts = (
            masked_gates - torch.empty_like(masked_gates, memory_format=torch.legacy_contiguous_format).exponential_().log()
        ).max(dim=-1)[1].unsqueeze(-1) # gumbel sampling, more robust than than the multinomial method
    else:
        selected_experts = max_ind
        
    # compute scores for gradients
    masked_gates = torch.softmax(masked_gates, dim=-1)
    multiplier_o = masked_gates.gather(dim=-1, index=selected_experts)
    
    if training:
        # compute midpoint mask 
        max_scores, max_ind = masked_gates.max(dim=-1, keepdim=True)
        mask_for_one = torch.logical_or(
            selected_experts == max_ind,
            torch.rand_like(max_scores) > 0.75 # Heun's third-order method: f(x) - f(0) = .25 f'(x) + .75 f'(x/3.)
        ) 
        # 1 -> 1.0 & 0 -> 1./3: lambda x: (x + 0.5) / 1.5
        mask_for_one = torch.add(0.3333, mask_for_one, alpha=0.6667).type_as(masked_gates)

        multiplier = mp.apply(
            scores, 
            multiplier_o, 
            selected_experts, 
            masked_gates, 
            mask_for_one,
        )
    else:
        multiplier = multiplier_o

    # masked out first expert 
    masked_scores = torch.scatter(
        scores,
        -1,
        selected_experts,
        float('-inf'),
    )
    with torch.no_grad():
        # compute mask for sparsity
        mask_logits_threshold, max_ind = masked_scores.max(dim=-1, keepdim=True)
        factor = scores.abs().clamp(min=mask_logits_threshold)
        mask_logits_threshold = (
            (mask_logits_threshold - scores) / factor
        ) > (2 * jitter_eps)

    # apply mask 
    masked_gates_top2 = masked_scores.masked_fill(mask_logits_threshold, float('-inf'))
    if training:
        selected_experts_top2 = (
            masked_gates_top2 - torch.empty_like(masked_gates_top2, memory_format=torch.legacy_contiguous_format).exponential_().log()
        ).max(dim=-1)[1].unsqueeze(-1) # gumbel sampling, more robust than than the multinomial method
    else:
        selected_experts_top2 = max_ind
    # compute scores for gradients
    masked_gates_top2 = torch.softmax(masked_gates_top2, dim=-1)
    multiplier_top2_o = masked_gates_top2.gather(dim=-1, index=selected_experts_top2)
    
    if training: 
        # compute midpoint mask 
        max_scores, max_ind = masked_gates_top2.max(dim=-1, keepdim=True)
        mask_for_one_top2 = torch.logical_or(
            selected_experts_top2 == max_ind,
            torch.rand_like(max_scores).uniform_() > 0.75 # Heun's third-order method: f(x) - f(0) = .25 f'(x) + .75 f'(x/3.)
        ) 
        # 1 -> 1.0 & 0 -> 1./3: lambda x: (x + 0.5) / 1.5
        mask_for_one_top2 = torch.add(0.3333, mask_for_one_top2, alpha=0.6667).type_as(masked_gates_top2)

        multiplier_top2 = mp.apply(
            scores, 
            multiplier_top2_o, 
            selected_experts_top2, 
            masked_gates_top2, 
            mask_for_one_top2,
        )
    else:
        multiplier_top2 = multiplier_top2_o
    
    multiplier = torch.concat((multiplier, multiplier_top2), dim=-1)
    selected_experts = torch.concat((selected_experts, selected_experts_top2), dim=-1)
    
    return (
        multiplier, 
        selected_experts,
    )

def load_balancing_loss_func(
    gate_logits: torch.Tensor, num_experts: torch.Tensor = None, top_k=2, attention_mask: Optional[torch.Tensor] = None
) -> float:
    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 = F.softmax(concatenated_gate_logits, dim=-1)

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

    expert_mask = F.one_hot(selected_experts, num_experts).permute(2, 1, 0)

    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_experts))
            .reshape(-1, top_k, num_experts)
            .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_experts))
            .reshape(-1, num_experts)
            .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_experts

class RasphiMoERotaryEmbedding(nn.Module):
    def __init__(self, dim, max_position_embeddings=2048, base=10000, device=None):
        super().__init__()
        inv_freq = 1.0 / (base ** (torch.arange(0, dim, 2).float().to(device) / dim))
        self.register_buffer("inv_freq", inv_freq)
        self.max_seq_len_cached = max_position_embeddings
        t = torch.arange(self.max_seq_len_cached, device=self.inv_freq.device, dtype=self.inv_freq.dtype)
        freqs = torch.einsum("i,j->ij", t, self.inv_freq)
        emb = torch.cat((freqs, freqs), dim=-1)
        self.register_buffer("cos_cached", emb.cos()[None, None, :, :], persistent=False)
        self.register_buffer("sin_cached", emb.sin()[None, None, :, :], persistent=False)

    def forward(self, x, seq_len=None):
        if seq_len > self.max_seq_len_cached:
            self._set_cos_sin_cache(seq_len, device=x.device, dtype=x.dtype)

        return (
            self.cos_cached[:, :, :seq_len, ...].to(dtype=x.dtype),
            self.sin_cached[:, :, :seq_len, ...].to(dtype=x.dtype),
        )

class LinearScalingRotaryEmbedding(RasphiMoERotaryEmbedding):
    def __init__(self, dim, max_position_embeddings=2048, base=10000, device=None, scaling_factor=1.0):
        self.scaling_factor = scaling_factor
        super().__init__(dim, max_position_embeddings, base, device)

    def _set_cos_sin_cache(self, seq_len, device, dtype):
        self.max_seq_len_cached = seq_len
        t = torch.arange(self.max_seq_len_cached, device=device, dtype=dtype)
        t = t / self.scaling_factor
        freqs = torch.einsum("i,j->ij", t, self.inv_freq)
        emb = torch.cat((freqs, freqs), dim=-1).to(device)
        self.register_buffer("cos_cached", emb.cos()[None, None, :, :], persistent=False)
        self.register_buffer("sin_cached", emb.sin()[None, None, :, :], persistent=False)

class DynamicNTKScalingRotaryEmbedding(RasphiMoERotaryEmbedding):
    def __init__(self, dim, max_position_embeddings=2048, base=10000, device=None, scaling_factor=1.0):
        self.scaling_factor = scaling_factor
        super().__init__(dim, max_position_embeddings, base, device)

    def _set_cos_sin_cache(self, seq_len, device, dtype):
        self.max_seq_len_cached = seq_len

        if seq_len > self.max_seq_len_cached:
            base = self.base * ((self.scaling_factor * seq_len / self.max_seq_len_cached) - (self.scaling_factor - 1)) ** (self.dim / (self.dim - 2))
            inv_freq = 1.0 / (base ** (torch.arange(0, self.dim, 2).float().to(device) / self.dim))
            self.register_buffer("inv_freq", inv_freq)

        t = torch.arange(self.max_seq_len_cached, device=device, dtype=dtype)
        freqs = torch.einsum("i,j->ij", t, self.inv_freq)
        emb = torch.cat((freqs, freqs), dim=-1).to(device)
        self.register_buffer("cos_cached", emb.cos()[None, None, :, :], persistent=False)
        self.register_buffer("sin_cached", emb.sin()[None, None, :, :], persistent=False)

def apply_rotary_pos_emb(q, k, cos, sin, position_ids):
    # The first two dimensions of cos and sin are always 1, so we can `squeeze` them.
    cos = cos.squeeze(1).squeeze(0)  # [seq_len, dim]
    sin = sin.squeeze(1).squeeze(0)  # [seq_len, dim]
    cos = cos[position_ids].unsqueeze(1)  # [bs, 1, seq_len, dim]
    sin = sin[position_ids].unsqueeze(1)  # [bs, 1, seq_len, dim]
    q_embed = (q * cos) + (rotate_half(q) * sin)
    k_embed = (k * cos) + (rotate_half(k) * sin)
    return q_embed, k_embed

def rotate_half(x):
    """Rotates half the hidden dims of the input."""
    x1 = x[..., : x.shape[-1] // 2]
    x2 = x[..., x.shape[-1] // 2 :]
    return torch.cat((-x2, x1), dim=-1)

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)

from transformers import AutoModelForCausalLM

AutoModelForCausalLM.register("rasphi", RasphiForCausalLM)