Add files using upload-large-folder tool

config.json

@@ -1,37 +1,37 @@
 {
-    "architectures": [
-      "LongcatCausalLM"
-    ],
-    "attention_bias": false,
-    "attention_dropout": 0.0,
-    "auto_map": {
-      "AutoConfig": "configuration_longcat.LongcatConfig",
-      "AutoModel": "modeling_longcat.LongcatModel",
-      "AutoModelForCausalLM": "modeling_longcat.LongcatForCausalLM"
-    },
-    "vocab_size": 131072,
-    "hidden_size": 6144,
-    "ffn_hidden_size": 12288,
-    "expert_ffn_hidden_size": 2048,
-    "num_layers": 28,
-    "num_attention_heads": 64,
-    "kv_lora_rank": 512,
-    "q_lora_rank": 1536,
-    "qk_rope_head_dim": 64,
-    "v_head_dim": 128,
-    "qk_nope_head_dim": 128,
-    "mla_scale_q_lora": true,
-    "mla_scale_kv_lora": true,
-    "routed_scaling_factor": 6.0,
-    "n_routed_experts": 512,
-    "max_position_embeddings": 131072,
-    "rms_norm_eps": 1e-5,
-    "use_cache": true,
-    "bos_token_id": 1,
-    "eos_token_id": 2,
-    "rope_theta": 10000000.0,
-    "attention_method": "MLA",
-    "zero_expert_num": 256,
-    "zero_expert_type": "identity",
-    "moe_topk": 12
-  }
+  "architectures": [
+      "LongcatFlashForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "auto_map": {
+      "AutoConfig": "configuration_longcat_flash.LongcatFlashConfig",
+      "AutoModel": "modeling_longcat_flash.LongcatFlashModel",
+      "AutoModelForCausalLM": "modeling_longcat_flash.LongcatFlashForCausalLM"
+  },
+  "vocab_size": 131072,
+  "hidden_size": 6144,
+  "ffn_hidden_size": 12288,
+  "expert_ffn_hidden_size": 2048,
+  "num_layers": 28,
+  "num_attention_heads": 64,
+  "kv_lora_rank": 512,
+  "q_lora_rank": 1536,
+  "qk_rope_head_dim": 64,
+  "v_head_dim": 128,
+  "qk_nope_head_dim": 128,
+  "mla_scale_q_lora": true,
+  "mla_scale_kv_lora": true,
+  "routed_scaling_factor": 6.0,
+  "n_routed_experts": 512,
+  "max_position_embeddings": 131072,
+  "rms_norm_eps": 1e-5,
+  "use_cache": true,
+  "bos_token_id": 1,
+  "eos_token_id": 2,
+  "rope_theta": 10000000.0,
+  "attention_method": "MLA",
+  "zero_expert_num": 256,
+  "zero_expert_type": "identity",
+  "moe_topk": 12
+}

configuration_longcat_flash.py

@@ -0,0 +1,216 @@
+
+"""LongcatFlash model configuration"""
+
+from transformers.configuration_utils import PretrainedConfig
+from transformers.modeling_rope_utils import rope_config_validation
+
+
+LONGCAT_PRETRAINED_CONFIG_ARCHIVE_MAP = {}
+
+
+class LongcatFlashConfig(PretrainedConfig):
+    r"""
+    This is the configuration class to store the configuration of a [`LongcatFlashModel`]. It is used to instantiate an LongcatFlash
+    model according to the specified arguments, defining the model architecture. Instantiating a configuration with the
+    defaults will yield a similar configuration to that of the LongcatFlash.
+    Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
+    documentation from [`PretrainedConfig`] for more information.
+
+
+    Args:
+        vocab_size (`int`, *optional*, defaults to 131072):
+            Vocabulary size of the Deep model. Defines the number of different tokens that can be represented by the
+            `inputs_ids` passed when calling [`LongcatFlashModel`]
+        hidden_size (`int`, *optional*, defaults to 7168):
+            Dimension of the hidden representations.
+        ffn_hidden_size (`int`, *optional*, defaults to 18432):
+            Dimension of the MLP representations.
+        expert_ffn_hidden_size (`int`, *optional*, defaults to 2048):
+            Dimension of the MoE representations.
+        num_layers (`int`, *optional*, defaults to 61):
+            Number of hidden layers in the Transformer decoder.
+        num_attention_heads (`int`, *optional*, defaults to 128):
+            Number of attention heads for each attention layer in the Transformer decoder.
+        num_key_value_heads (`int`, *optional*, defaults to 128):
+            This is the number of key_value heads that should be used to implement Grouped Query Attention. If
+            `num_key_value_heads=num_attention_heads`, the model will use Multi Head Attention (MHA), if
+            `num_key_value_heads=1 the model will use Multi Query Attention (MQA) otherwise GQA is used. When
+            converting a multi-head checkpoint to a GQA checkpoint, each group key and value head should be constructed
+            by meanpooling all the original heads within that group. For more details checkout [this
+            paper](https://arxiv.org/pdf/2305.13245.pdf). If it is not specified, will default to
+            `num_attention_heads`.
+        n_routed_experts (`int`, *optional*, defaults to 256):
+            Number of routed experts.
+        routed_scaling_factor (`float`, *optional*, defaults to 2.5):
+            Scaling factor or routed experts.
+        kv_lora_rank (`int`, *optional*, defaults to 512):
+            Rank of the LoRA matrices for key and value projections.
+        q_lora_rank (`int`, *optional*, defaults to 1536):
+            Rank of the LoRA matrices for query projections.
+        qk_rope_head_dim (`int`, *optional*, defaults to 64):
+            Dimension of the query/key heads that use rotary position embeddings.
+        v_head_dim (`int`, *optional*, defaults to 128):
+            Dimension of the value heads.
+        qk_nope_head_dim (`int`, *optional*, defaults to 128):
+            Dimension of the query/key heads that don't use rotary position embeddings.
+        norm_topk_prob (`bool`, *optional*, defaults to `True`):
+            Whether to normalize the weights of the routed experts.
+        hidden_act (`str` or `function`, *optional*, defaults to `"silu"`):
+            The non-linear activation function (function or string) in the decoder.
+        max_position_embeddings (`int`, *optional*, defaults to 4096):
+            The maximum sequence length that this model might ever be used with.
+        rms_norm_eps (`float`, *optional*, defaults to 1e-06):
+            The epsilon used by the rms normalization layers.
+        use_cache (`bool`, *optional*, defaults to `True`):
+            Whether or not the model should return the last key/values attentions (not used by all models). Only
+            relevant if `config.is_decoder=True`.
+        pad_token_id (`int`, *optional*):
+            Padding token id.
+        bos_token_id (`int`, *optional*, defaults to 0):
+            Beginning of stream token id.
+        eos_token_id (`int`, *optional*, defaults to 1):
+            End of stream token id.
+        tie_word_embeddings (`bool`, *optional*, defaults to `False`):
+            Whether to tie weight embeddings
+        rope_theta (`float`, *optional*, defaults to 10000.0):
+            The base period of the RoPE embeddings.
+        attention_bias (`bool`, defaults to `False`, *optional*, defaults to `False`):
+            Whether to use a bias in the query, key, value and output projection layers during self-attention.
+        attention_dropout (`float`, *optional*, defaults to 0.0):
+            The dropout ratio for the attention probabilities.
+        attention_method (`str`, *optional*, defaults to `"MLA"`):
+            The attention method to use.
+        initializer_range (`float`, *optional*, defaults to 0.006):
+            The initializer range for the model.
+        router_bias (`bool`, *optional*, defaults to `False`):
+            Whether to use a bias in the router.
+        zero_expert_num (`int`, *optional*, defaults to `None`):
+            The number of zero experts to use.
+        zero_expert_type (`str`, *optional*, defaults to `None`):
+            The type of zero expert to use.
+
+    ```python
+    >>> from transformers import LongcatFlashModel, LongcatFlashConfig
+
+    >>> # Initializing a LongcatFlash style configuration
+    >>> configuration = LongcatFlashConfig()
+
+    >>> # Accessing the model configuration
+    >>> configuration = model.config
+    ```"""
+
+    model_type = "longcat_flash"
+    keys_to_ignore_at_inference = ["past_key_values"]
+    base_model_tp_plan = {
+        "layers.*.self_attn.k_proj": "colwise",
+        "layers.*.self_attn.v_proj": "colwise",
+        "layers.*.self_attn.o_proj": "rowwise",
+        "layers.*.mlp.experts.*.gate_proj": "local_colwise",
+        "layers.*.mlp.experts.*.up_proj": "local_colwise",
+        "layers.*.mlp.experts.*.down_proj": "local_rowwise",
+        "layers.*.mlps.*.gate_proj": "local_colwise",
+        "layers.*.mlps.*.up_proj": "local_colwise",
+        "layers.*.mlps.*.down_proj": "local_rowwise",
+    }
+    base_model_pp_plan = {
+        "embed_tokens": (["input_ids"], ["inputs_embeds"]),
+        "layers": (["hidden_states", "attention_mask"], ["hidden_states"]),
+        "norm": (["hidden_states"], ["hidden_states"]),
+    }
+
+    def __init__(
+        self,
+        vocab_size=131072,
+        hidden_size=7168,
+        ffn_hidden_size=18432,
+        expert_ffn_hidden_size=2048,
+        num_layers=61,
+        num_attention_heads=128,
+        num_key_value_heads=None,
+        n_routed_experts=256,
+        routed_scaling_factor=1,
+        kv_lora_rank=512,
+        q_lora_rank=1536,
+        qk_rope_head_dim=64,
+        v_head_dim=128,
+        qk_nope_head_dim=128,
+        mla_scale_q_lora=True,
+        mla_scale_kv_lora=True,
+        moe_topk=8,
+        norm_topk_prob=False,
+        hidden_act="silu",
+        max_position_embeddings=4096,
+        rms_norm_eps=1e-6,
+        use_cache=True,
+        pad_token_id=None,
+        bos_token_id=0,
+        eos_token_id=1,
+        tie_word_embeddings=False,
+        rope_theta=10000.0,
+        attention_bias=False,
+        attention_dropout=0.0,
+        attention_method='MLA',
+        initializer_range=0.006,
+        router_bias=False,
+        zero_expert_num=None,
+        zero_expert_type=None,
+        **kwargs,
+    ):
+        self.vocab_size = vocab_size
+        self.max_position_embeddings = max_position_embeddings
+        self.hidden_size = hidden_size
+        self.ffn_hidden_size = ffn_hidden_size
+        self.expert_ffn_hidden_size = expert_ffn_hidden_size
+        self.num_layers = num_layers
+        self.num_attention_heads = num_attention_heads
+        self.n_routed_experts = n_routed_experts
+        self.routed_scaling_factor = routed_scaling_factor
+        self.kv_lora_rank = kv_lora_rank
+        self.q_lora_rank = q_lora_rank
+        self.qk_rope_head_dim = qk_rope_head_dim
+        self.v_head_dim = v_head_dim
+        self.qk_nope_head_dim = qk_nope_head_dim
+        self.qk_head_dim = qk_nope_head_dim + qk_rope_head_dim
+        self.moe_topk = moe_topk
+        self.norm_topk_prob = norm_topk_prob
+        self.mla_scale_q_lora = mla_scale_q_lora
+        self.mla_scale_kv_lora = mla_scale_kv_lora
+        self.attention_method = attention_method
+        self.initializer_range = initializer_range
+        self.router_bias = router_bias
+        self.zero_expert_num = zero_expert_num
+        self.zero_expert_type = zero_expert_type
+
+        if self.attention_method == "MLA":
+            self.head_dim = qk_rope_head_dim
+        else:
+            ValueError('attention_method should be one of ["MLA"]')
+
+
+        if num_key_value_heads is None:
+            num_key_value_heads = num_attention_heads
+
+        self.num_key_value_heads = num_key_value_heads
+        self.hidden_act = hidden_act
+        self.rms_norm_eps = rms_norm_eps
+        self.use_cache = use_cache
+        self.rope_theta = rope_theta
+        self.attention_bias = attention_bias
+        self.attention_dropout = attention_dropout
+
+        rope_config_validation(self)
+
+        super().__init__(
+            pad_token_id=pad_token_id,
+            bos_token_id=bos_token_id,
+            eos_token_id=eos_token_id,
+            tie_word_embeddings=tie_word_embeddings,
+            **kwargs,
+        )
+
+    @property
+    def num_hidden_layers(self):
+        return self.num_layers
+
+
+__all__ = ["LongcatFlashConfig"]

modeling_longcat_flash.py

@@ -0,0 +1,644 @@
+from typing import Callable, Optional, Union
+
+import torch
+import torch.nn.functional as F
+from torch import nn
+
+from transformers.activations import ACT2FN
+from transformers.cache_utils import Cache, DynamicCache
+from transformers.generation import GenerationMixin
+from transformers.integrations import use_kernel_forward_from_hub
+from transformers.masking_utils import create_causal_mask
+from transformers.modeling_flash_attention_utils import FlashAttentionKwargs
+from transformers.modeling_layers import GradientCheckpointingLayer
+from transformers.modeling_outputs import BaseModelOutputWithPast, CausalLMOutputWithPast
+from transformers.modeling_rope_utils import ROPE_INIT_FUNCTIONS, dynamic_rope_update
+from transformers.modeling_utils import ALL_ATTENTION_FUNCTIONS, PreTrainedModel
+from transformers.processing_utils import Unpack
+from transformers.utils import TransformersKwargs, auto_docstring, can_return_tuple
+from transformers.utils.generic import check_model_inputs
+from .configuration_longcat_flash import LongcatFlashConfig
+
+
+@use_kernel_forward_from_hub("RMSNorm")
+class LongcatFlashRMSNorm(nn.Module):
+    def __init__(self, hidden_size, eps=1e-6):
+        """
+        LongcatFlashRMSNorm is equivalent to T5LayerNorm
+        """
+        super().__init__()
+        self.weight = nn.Parameter(torch.ones(hidden_size))
+        self.variance_epsilon = eps
+
+    def forward(self, hidden_states):
+        input_dtype = hidden_states.dtype
+        hidden_states = hidden_states.to(torch.float32)
+        variance = hidden_states.pow(2).mean(-1, keepdim=True)
+        hidden_states = hidden_states * torch.rsqrt(variance + self.variance_epsilon)
+        return self.weight * hidden_states.to(input_dtype)
+
+    def extra_repr(self):
+        return f"{tuple(self.weight.shape)}, eps={self.variance_epsilon}"
+
+
+class LongcatFlashRotaryEmbedding(nn.Module):
+    def __init__(self, config: LongcatFlashConfig, device=None):
+        super().__init__()
+        # BC: "rope_type" was originally "type"
+        if hasattr(config, "rope_scaling") and isinstance(config.rope_scaling, dict):
+            self.rope_type = config.rope_scaling.get("rope_type", config.rope_scaling.get("type"))
+        else:
+            self.rope_type = "default"
+        self.max_seq_len_cached = config.max_position_embeddings
+        self.original_max_seq_len = config.max_position_embeddings
+
+        self.config = config
+        self.rope_init_fn = ROPE_INIT_FUNCTIONS[self.rope_type]
+
+        inv_freq, self.attention_scaling = self.rope_init_fn(self.config, device)
+        self.register_buffer("inv_freq", inv_freq, persistent=False)
+        self.original_inv_freq = self.inv_freq
+
+    @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 torch.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)
+
+
+class LongcatFlashMLP(nn.Module):
+    def __init__(self, config, hidden_size=None, intermediate_size=None):
+        super().__init__()
+        self.config = config
+        self.hidden_size = config.hidden_size if hidden_size is None else hidden_size
+        self.intermediate_size = config.ffn_hidden_size if intermediate_size is None else intermediate_size
+
+        self.gate_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
+        self.up_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
+        self.down_proj = nn.Linear(self.intermediate_size, self.hidden_size, bias=False)
+        self.act_fn = ACT2FN[config.hidden_act]
+
+    def forward(self, x):
+        down_proj = self.down_proj(self.act_fn(self.gate_proj(x)) * self.up_proj(x))
+        return down_proj
+
+
+class LongcatFlashTopkRouter(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.config = config
+        self.top_k = config.moe_topk
+        self.n_routed_experts = (
+            config.n_routed_experts
+            if config.zero_expert_num is None
+            else config.n_routed_experts + config.zero_expert_num
+        )
+        self.routed_scaling_factor = config.routed_scaling_factor
+        self.norm_topk_prob = config.norm_topk_prob
+        self.router_bias = config.router_bias
+
+        self.classifier = nn.Linear(config.hidden_size, self.n_routed_experts, bias=self.router_bias)
+        self.register_buffer("e_score_correction_bias", torch.zeros((self.n_routed_experts)))
+
+    @torch.no_grad()
+    def get_topk_indices(self, scores):
+        scores_for_choice = scores.view(-1, self.n_routed_experts) + self.e_score_correction_bias.unsqueeze(0)
+        topk_indices = torch.topk(scores_for_choice, k=self.top_k, dim=-1, sorted=False)[1]
+        return topk_indices
+
+    def forward(self, hidden_states):
+        hidden_states = hidden_states.view(-1, self.config.hidden_size)
+        router_logits = F.linear(hidden_states.type(torch.float32), self.classifier.weight.type(torch.float32))
+        scores = router_logits.softmax(dim=-1)
+        topk_indices = self.get_topk_indices(scores)
+        topk_weights = scores.gather(1, topk_indices)
+        if self.norm_topk_prob:
+            denominator = topk_weights.sum(dim=-1, keepdim=True) + 1e-20
+            topk_weights /= denominator
+        topk_weights = topk_weights * self.routed_scaling_factor
+        return topk_indices, topk_weights
+
+
+class LongcatFlashMoE(nn.Module):
+    """
+    moe module.
+    """
+
+    def __init__(self, config):
+        super().__init__()
+        self.config = config
+        self.experts = nn.ModuleList(
+            [
+                LongcatFlashMLP(config, intermediate_size=config.expert_ffn_hidden_size)
+                for _ in range(config.n_routed_experts)
+            ]
+        )
+        self.router = LongcatFlashTopkRouter(config)
+        self.zero_expert_num = config.zero_expert_num
+        self.zero_expert_type = config.zero_expert_type
+
+    def moe(self, hidden_states: torch.Tensor, topk_indices: torch.Tensor, topk_weights: torch.Tensor):
+        final_hidden_states = torch.zeros_like(hidden_states, dtype=topk_weights.dtype)
+        total_experts = len(self.experts) if self.zero_expert_num is None else len(self.experts) + self.zero_expert_num
+
+        expert_mask = torch.nn.functional.one_hot(topk_indices, num_classes=total_experts)
+        expert_mask = expert_mask.permute(2, 0, 1)
+
+        for expert_idx in range(total_experts):
+            expert = self.experts[expert_idx] if expert_idx < len(self.experts) else None
+            mask = expert_mask[expert_idx]
+            token_indices, weight_indices = torch.where(mask)
+
+            if token_indices.numel() > 0:
+                expert_weights = topk_weights[token_indices, weight_indices]
+                expert_input = hidden_states[token_indices]
+
+                if self.zero_expert_num is None or expert_idx < len(self.experts):
+                    expert_output = expert(expert_input)
+                elif self.zero_expert_type == "identity":
+                    expert_output = expert_input
+                else:
+                    raise ValueError("Unknown condition")
+
+                weighted_output = expert_output * expert_weights.unsqueeze(-1)
+                final_hidden_states.index_add_(0, token_indices, weighted_output)
+
+        return final_hidden_states.type(hidden_states.dtype)
+
+    def forward(self, hidden_states):
+        orig_shape = hidden_states.shape
+        topk_indices, topk_weights = self.router(hidden_states)
+        hidden_states = hidden_states.view(-1, hidden_states.shape[-1])
+        hidden_states = self.moe(hidden_states, topk_indices, topk_weights).view(*orig_shape)
+        return hidden_states
+
+
+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)
+
+
+def eager_attention_forward(
+    module: nn.Module,
+    query: torch.Tensor,
+    key: torch.Tensor,
+    value: torch.Tensor,
+    attention_mask: Optional[torch.Tensor],
+    scaling: float,
+    dropout: float = 0.0,
+    **kwargs: Unpack[TransformersKwargs],
+):
+    key_states = repeat_kv(key, module.num_key_value_groups)
+    value_states = repeat_kv(value, module.num_key_value_groups)
+
+    attn_weights = torch.matmul(query, key_states.transpose(2, 3)) * scaling
+    if attention_mask is not None:
+        causal_mask = attention_mask[:, :, :, : key_states.shape[-2]]
+        attn_weights = attn_weights + causal_mask
+
+    attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(query.dtype)
+    attn_weights = nn.functional.dropout(attn_weights, p=dropout, training=module.training)
+    attn_output = torch.matmul(attn_weights, value_states)
+    attn_output = attn_output.transpose(1, 2).contiguous()
+
+    return attn_output, attn_weights
+
+
+def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1, use_mla=False):
+    """Applies Rotary Position Embedding to the query and key tensors.
+
+    Args:
+        q (`torch.Tensor`): The query tensor.
+        k (`torch.Tensor`): The key tensor.
+        cos (`torch.Tensor`): The cosine part of the rotary embedding.
+        sin (`torch.Tensor`): The sine part of the rotary embedding.
+        position_ids (`torch.Tensor`, *optional*):
+            Deprecated and unused.
+        unsqueeze_dim (`int`, *optional*, defaults to 1):
+            The 'unsqueeze_dim' argument specifies the dimension along which to unsqueeze cos[position_ids] and
+            sin[position_ids] so that they can be properly broadcasted to the dimensions of q and k. For example, note
+            that cos[position_ids] and sin[position_ids] have the shape [batch_size, seq_len, head_dim]. Then, if q and
+            k have the shape [batch_size, heads, seq_len, head_dim], then setting unsqueeze_dim=1 makes
+            cos[position_ids] and sin[position_ids] broadcastable to the shapes of q and k. Similarly, if q and k have
+            the shape [batch_size, seq_len, heads, head_dim], then set unsqueeze_dim=2.
+    Returns:
+        `tuple(torch.Tensor)` comprising of the query and key tensors rotated using the Rotary Position Embedding.
+    """
+    cos = cos.unsqueeze(unsqueeze_dim)
+    sin = sin.unsqueeze(unsqueeze_dim)
+
+    if use_mla:
+        b, h, s, d = q.shape
+        q = q.view(b, h, s, d // 2, 2).transpose(4, 3).reshape(b, h, s, d)
+
+        b, h, s, d = k.shape
+        k = k.view(b, h, s, d // 2, 2).transpose(4, 3).reshape(b, h, s, d)
+
+    q_embed = (q * cos) + (rotate_half(q) * sin)
+    k_embed = (k * cos) + (rotate_half(k) * sin)
+    return q_embed, k_embed
+
+
+class LongcatFlashMLA(nn.Module):
+    """Modified from Deepseek MLA"""
+
+    def __init__(self, config: LongcatFlashConfig, layer_idx: int):
+        super().__init__()
+        self.config = config
+        self.layer_idx = layer_idx
+        self.num_key_value_groups = config.num_attention_heads // config.num_key_value_heads
+        self.attention_dropout = config.attention_dropout
+        self.num_heads = config.num_attention_heads
+        self.rope_theta = config.rope_theta
+        self.q_lora_rank = config.q_lora_rank
+        self.qk_rope_head_dim = config.qk_rope_head_dim
+        self.kv_lora_rank = config.kv_lora_rank
+        self.v_head_dim = config.v_head_dim
+        self.qk_nope_head_dim = config.qk_nope_head_dim
+        self.qk_head_dim = config.qk_head_dim
+
+        self.is_causal = True
+        if self.q_lora_rank is None:
+            self.q_proj = nn.Linear(config.hidden_size, self.num_heads * self.qk_head_dim, bias=False)
+        else:
+            self.q_a_proj = nn.Linear(config.hidden_size, config.q_lora_rank, bias=config.attention_bias)
+            self.q_a_layernorm = LongcatFlashRMSNorm(config.q_lora_rank)
+            self.q_b_proj = nn.Linear(config.q_lora_rank, self.num_heads * self.qk_head_dim, bias=False)
+
+        self.kv_a_proj_with_mqa = nn.Linear(
+            config.hidden_size,
+            self.kv_lora_rank + self.qk_rope_head_dim,
+            bias=config.attention_bias,
+        )
+        self.kv_a_layernorm = LongcatFlashRMSNorm(self.kv_lora_rank)
+        self.kv_b_proj = nn.Linear(
+            self.kv_lora_rank,
+            self.num_heads * (self.qk_nope_head_dim + self.v_head_dim),
+            bias=False,
+        )
+
+        self.o_proj = nn.Linear(
+            self.num_heads * self.v_head_dim,
+            config.hidden_size,
+            bias=config.attention_bias,
+        )
+
+        if config.mla_scale_q_lora:
+            self.mla_scale_q_lora = (config.hidden_size / self.q_lora_rank) ** 0.5
+        if config.mla_scale_kv_lora:
+            self.mla_scale_kv_lora = (config.hidden_size / self.kv_lora_rank) ** 0.5
+        self.scaling = self.qk_head_dim ** (-0.5)
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        position_embeddings: tuple[torch.Tensor, torch.Tensor],
+        attention_mask: Optional[torch.Tensor],
+        past_key_value: Optional[Cache] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+        **kwargs: Unpack[FlashAttentionKwargs],
+    ) -> tuple[torch.Tensor, Optional[torch.Tensor], Optional[tuple[torch.Tensor]]]:
+        batch_size, seq_length = hidden_states.shape[:-1]
+        query_shape = (batch_size, seq_length, -1, self.qk_head_dim)
+        key_shape = (batch_size, seq_length, -1, self.qk_nope_head_dim + self.v_head_dim)
+
+        q_states = self.q_b_proj(self.q_a_layernorm(self.q_a_proj(hidden_states))).view(query_shape).transpose(1, 2)
+        q_pass, q_rot = torch.split(q_states, [self.qk_nope_head_dim, self.qk_rope_head_dim], dim=-1)
+
+        # apply q_lora scaling
+        if self.mla_scale_q_lora is not None:
+            q_pass = q_pass * self.mla_scale_q_lora
+            q_rot = q_rot * self.mla_scale_q_lora
+
+        compressed_kv = self.kv_a_proj_with_mqa(hidden_states)
+        k_pass, k_rot = torch.split(compressed_kv, [self.kv_lora_rank, self.qk_rope_head_dim], dim=-1)
+        k_pass = self.kv_a_layernorm(k_pass)
+
+        # apply kv_lora scaling
+        if self.mla_scale_kv_lora is not None:
+            k_pass = k_pass * self.mla_scale_kv_lora
+
+        k_pass = self.kv_b_proj(k_pass).view(key_shape).transpose(1, 2)
+        k_pass, value_states = torch.split(k_pass, [self.qk_nope_head_dim, self.v_head_dim], dim=-1)
+
+        k_rot = k_rot.view(batch_size, 1, seq_length, self.qk_rope_head_dim)
+
+        cos, sin = position_embeddings
+        q_rot, k_rot = apply_rotary_pos_emb(q_rot, k_rot, cos, sin, use_mla=True)
+        k_rot = k_rot.expand(*k_pass.shape[:-1], -1)
+
+        query_states = torch.cat((q_pass, q_rot), dim=-1)
+        key_states = torch.cat((k_pass, k_rot), dim=-1)
+
+        if past_key_value is not None:
+            cache_kwargs = {"sin": sin, "cos": cos, "cache_position": cache_position}
+            key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs)
+
+        if self.config._attn_implementation == "flash_attention_2" and self.qk_head_dim != self.v_head_dim:
+            value_states = F.pad(value_states, [0, self.qk_head_dim - self.v_head_dim])
+
+        attention_interface: Callable = eager_attention_forward
+        if self.config._attn_implementation != "eager":
+            attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]
+
+        attn_output, attn_weights = attention_interface(
+            self,
+            query_states,
+            key_states,
+            value_states,
+            attention_mask,
+            dropout=0.0 if not self.training else self.attention_dropout,
+            scaling=self.scaling,
+            **kwargs,
+        )
+
+        if self.config._attn_implementation == "flash_attention_2" and self.qk_head_dim != self.v_head_dim:
+            attn_output = attn_output[:, :, :, : self.v_head_dim]
+
+        attn_output = attn_output.reshape(batch_size, seq_length, -1).contiguous()
+        attn_output = self.o_proj(attn_output)
+        return attn_output, attn_weights
+
+
+def create_attention_block(class_name, *args, **kwargs):
+    attention_mapping = {"MLA": LongcatFlashMLA}
+
+    chosen_class = attention_mapping.get(class_name)
+    if not chosen_class:
+        raise ValueError(f"No class found for name: {class_name}")
+
+    return chosen_class(*args, **kwargs)
+
+
+class LongcatFlashDecoderLayer(GradientCheckpointingLayer):
+    def __init__(self, config: LongcatFlashConfig, layer_idx: int):
+        super().__init__()
+        self.layer_idx = layer_idx
+        self.hidden_size = config.hidden_size
+        self.mlp = LongcatFlashMoE(config)
+
+        self_attn = []
+        mlps = []
+        input_layernorm = []
+        post_attention_layernorm = []
+        for i in range(2):
+            self_attn.append(
+                create_attention_block(config.attention_method, config=config, layer_idx=layer_idx * 2 + i)
+            )
+            mlps.append(LongcatFlashMLP(config))
+            input_layernorm.append(LongcatFlashRMSNorm(config.hidden_size, eps=config.rms_norm_eps))
+            post_attention_layernorm.append(LongcatFlashRMSNorm(config.hidden_size, eps=config.rms_norm_eps))
+
+        self.self_attn = nn.ModuleList(self_attn)
+        self.mlps = nn.ModuleList(mlps)
+        self.input_layernorm = nn.ModuleList(input_layernorm)
+        self.post_attention_layernorm = nn.ModuleList(post_attention_layernorm)
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_value: Optional[Cache] = None,
+        use_cache: Optional[bool] = False,
+        cache_position: Optional[torch.LongTensor] = None,
+        position_embeddings: Optional[tuple[torch.Tensor, torch.Tensor]] = None,
+        **kwargs: Unpack[FlashAttentionKwargs],
+    ) -> tuple[torch.FloatTensor, Optional[tuple[torch.FloatTensor, torch.FloatTensor]]]:
+        for i in range(2):
+            residual = hidden_states
+
+            hidden_states = self.input_layernorm[i](hidden_states)
+
+            hidden_states, _ = self.self_attn[i](
+                hidden_states=hidden_states,
+                attention_mask=attention_mask,
+                position_ids=position_ids,
+                past_key_value=past_key_value,
+                use_cache=use_cache,
+                cache_position=cache_position,
+                position_embeddings=position_embeddings,
+                **kwargs,
+            )
+            hidden_states = residual + hidden_states
+
+            residual = hidden_states
+            hidden_states = self.post_attention_layernorm[i](hidden_states)
+
+            if i == 0:
+                shortcut_mlp_output = self.mlp(hidden_states)  # shortcut output (MoE output)
+
+            hidden_states = self.mlps[i](hidden_states)
+            hidden_states = residual + hidden_states
+            if i == 1:
+                hidden_states = hidden_states + shortcut_mlp_output
+
+        return hidden_states
+
+
+@auto_docstring
+class LongcatFlashPreTrainedModel(PreTrainedModel):
+    config: LongcatFlashConfig
+    base_model_prefix = "model"
+    supports_gradient_checkpointing = True
+    _no_split_modules = ["LongcatFlashDecoderLayer"]
+    _skip_keys_device_placement = ["past_key_values"]
+    _supports_flash_attn = True
+    _supports_sdpa = True
+    _supports_flex_attn = True
+    _can_compile_fullgraph = True
+    _supports_attention_backend = True
+    _can_record_outputs = {
+        "hidden_states": LongcatFlashDecoderLayer,
+        "attentions": LongcatFlashMLA,
+    }
+
+
+@auto_docstring
+class LongcatFlashModel(LongcatFlashPreTrainedModel):
+    _keys_to_ignore_on_load_unexpected = [r"model\.mtp.*"]
+
+    def __init__(self, config: LongcatFlashConfig):
+        super().__init__(config)
+        self.padding_idx = config.pad_token_id
+        self.vocab_size = config.vocab_size
+
+        self.embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size, self.padding_idx)
+        self.layers = nn.ModuleList(
+            [LongcatFlashDecoderLayer(config, layer_idx) for layer_idx in range(config.num_hidden_layers)]
+        )
+        self.norm = LongcatFlashRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.rotary_emb = LongcatFlashRotaryEmbedding(config=config)
+        self.gradient_checkpointing = False
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    @check_model_inputs
+    @auto_docstring
+    def forward(
+        self,
+        input_ids: Optional[torch.LongTensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_values: Optional[Cache] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+        use_cache: Optional[bool] = 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()
+
+        if cache_position is None:
+            past_seen_tokens = past_key_values.get_seq_length() if past_key_values is not None else 0
+            cache_position: torch.Tensor = torch.arange(
+                past_seen_tokens, past_seen_tokens + inputs_embeds.shape[1], device=inputs_embeds.device
+            )
+
+        if position_ids is None:
+            position_ids = cache_position.unsqueeze(0)
+
+        causal_mask = create_causal_mask(
+            config=self.config,
+            input_embeds=inputs_embeds,
+            attention_mask=attention_mask,
+            cache_position=cache_position,
+            past_key_values=past_key_values,
+            position_ids=position_ids,
+        )
+
+        hidden_states = inputs_embeds
+        position_embeddings = self.rotary_emb(hidden_states, position_ids)
+
+        for decoder_layer in self.layers[: self.config.num_hidden_layers]:
+            hidden_states = decoder_layer(
+                hidden_states,
+                attention_mask=causal_mask,
+                position_ids=position_ids,
+                past_key_value=past_key_values,
+                cache_position=cache_position,
+                position_embeddings=position_embeddings,
+                **kwargs,
+            )
+
+        hidden_states = self.norm(hidden_states)
+        return BaseModelOutputWithPast(
+            last_hidden_state=hidden_states,
+            past_key_values=past_key_values,
+        )
+
+
+@auto_docstring
+class LongcatFlashForCausalLM(LongcatFlashPreTrainedModel, GenerationMixin):
+    _tied_weights_keys = ["lm_head.weight"]
+    _tp_plan = {"lm_head": "colwise_rep"}
+    _pp_plan = {"lm_head": (["hidden_states"], ["logits"])}
+    _keys_to_ignore_on_load_unexpected = [r"model\.mtp.*"]
+
+    def __init__(self, config):
+        super().__init__(config)
+        self.model = LongcatFlashModel(config)
+        self.vocab_size = config.vocab_size
+        self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def set_decoder(self, decoder):
+        self.model = decoder
+
+    def get_decoder(self):
+        return self.model
+
+    @can_return_tuple
+    @auto_docstring
+    def forward(
+        self,
+        input_ids: Optional[torch.LongTensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_values: Optional[Cache] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        labels: Optional[torch.LongTensor] = None,
+        use_cache: Optional[bool] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+        logits_to_keep: Union[int, torch.Tensor] = 0,
+        **kwargs: Unpack[TransformersKwargs],
+    ) -> CausalLMOutputWithPast:
+        r"""
+        Example:
+
+        ```python
+        >>> from transformers import AutoTokenizer, LongcatFlashForCausalLM
+
+        >>> model = LongcatFlashForCausalLM.from_pretrained("meta-longcat_flash/LongcatFlash-2-7b-hf")
+        >>> tokenizer = AutoTokenizer.from_pretrained("meta-longcat_flash/LongcatFlash-2-7b-hf")
+
+        >>> prompt = "Hey, are you conscious? Can you talk to me?"
+        >>> inputs = tokenizer(prompt, return_tensors="pt")
+
+        >>> # Generate
+        >>> generate_ids = model.generate(inputs.input_ids, max_length=30)
+        >>> tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
+        "Hey, are you conscious? Can you talk to me?\nI'm not conscious, but I can talk to you."
+        ```"""
+        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,
+            cache_position=cache_position,
+            **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, :])
+
+        loss = None
+        if labels is not None:
+            loss = self.loss_function(logits=logits, labels=labels, vocab_size=self.config.vocab_size, **kwargs)
+
+        return CausalLMOutputWithPast(
+            loss=loss,
+            logits=logits,
+            past_key_values=outputs.past_key_values,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )
+
+
+__all__ = ["LongcatFlashPreTrainedModel", "LongcatFlashModel", "LongcatFlashForCausalLM"]