init commit

configuration_grok1.py

@@ -0,0 +1,58 @@
+from transformers.configuration_utils import PretrainedConfig
+
+
+class Grok1Config(PretrainedConfig):
+    model_type = "grok-1"
+    keys_to_ignore_at_inference = ["past_key_values"]
+
+    def __init__(
+        self,
+        vocab_size=32000,
+        hidden_size=4096,
+        widening_factor=4.0,
+        num_hidden_layers=32,
+        num_attention_heads=32,
+        num_key_value_heads=32,
+        attn_output_multiplier=1.0,
+        max_attn_value=1.0,
+        max_position_embeddings=4096,
+        rms_norm_eps=1e-5,
+        use_cache=True,
+        pad_token_id=None,
+        bos_token_id=1,
+        eos_token_id=2,
+        tie_word_embeddings=True,
+        num_experts_per_tok=2,
+        num_experts=8,
+        output_router_logits=False,
+        router_aux_loss_coef=0.001,
+        **kwargs
+    ):
+        self.vocab_size = vocab_size
+        self.attn_output_multiplier = attn_output_multiplier
+        self.max_attn_value = max_attn_value
+        self.max_position_embeddings = max_position_embeddings
+        self.hidden_size = hidden_size
+        self.widening_factor = widening_factor
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+
+        # for backward compatibility
+        if num_key_value_heads is None:
+            num_key_value_heads = num_attention_heads
+
+        self.num_key_value_heads = num_key_value_heads
+        self.rms_norm_eps = rms_norm_eps
+        self.use_cache = use_cache
+
+        self.num_experts_per_tok = num_experts_per_tok
+        self.num_experts = num_experts
+        self.output_router_logits = output_router_logits
+        self.router_aux_loss_coef = router_aux_loss_coef
+        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,
+        )

modeling_grok1.py

@@ -0,0 +1,923 @@
+from typing import List, Optional, Tuple, Union
+
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+from transformers.modeling_utils import PreTrainedModel
+from transformers.utils import logging
+
+try:
+    from transformers.modeling_attn_mask_utils import _prepare_4d_causal_attention_mask
+
+    HAS_MASK_UTILS = True
+except ImportError:
+    HAS_MASK_UTILS = False
+
+from .configuration_grok1 import Grok1Config
+from .modeling_grok1_outputs import MoeCausalLMOutputWithPast, MoeModelOutputWithPast
+
+logger = logging.get_logger(__name__)
+
+
+# copied from https://github.com/huggingface/transformers/blob/v4.36.1/src/transformers/models/mixtral/modeling_mixtral.py
+def load_balancing_loss_func(
+    gate_logits: torch.Tensor, num_experts: torch.Tensor = None, top_k=2
+) -> 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 tensors. Shape: [batch_size, seqeunce_length, num_experts].
+        num_experts (`int`, *optional*):
+            Number of experts
+
+    Returns:
+        The auxiliary loss.
+    """
+    if gate_logits is None:
+        return 0
+
+    if isinstance(gate_logits, tuple):
+        # cat along the layers?
+        compute_device = gate_logits[0].device
+        gate_logits = torch.cat(
+            [gate.to(compute_device) for gate in gate_logits], dim=0
+        )
+
+    routing_weights, selected_experts = torch.topk(gate_logits, top_k, dim=-1)
+    routing_weights = routing_weights.softmax(dim=-1)
+
+    # cast the expert indices to int64, otherwise one-hot encoding will fail
+    if selected_experts.dtype != torch.int64:
+        selected_experts = selected_experts.to(torch.int64)
+
+    if len(selected_experts.shape) == 2:
+        selected_experts = selected_experts.unsqueeze(2)
+
+    expert_mask = torch.nn.functional.one_hot(selected_experts, num_experts)
+
+    # For a given token, determine if it was routed to a given expert.
+    expert_mask = torch.max(expert_mask, axis=-2).values
+
+    # cast to float32 otherwise mean will fail
+    expert_mask = expert_mask.to(torch.float32)
+    tokens_per_group_and_expert = torch.mean(expert_mask, axis=-2)
+
+    router_prob_per_group_and_expert = torch.mean(routing_weights, axis=-1)
+    return torch.mean(
+        tokens_per_group_and_expert * router_prob_per_group_and_expert.unsqueeze(-1)
+    ) * (num_experts**2)
+
+
+class RMSNorm(nn.Module):
+    def __init__(
+        self,
+        hidden_size: int,
+        eps: float = 1e-5,
+        create_scale: bool = True,
+    ) -> None:
+        super().__init__()
+        self.variance_epsilon = eps
+        if create_scale:
+            self.scale = nn.Parameter(torch.zeros(hidden_size))
+        else:
+            self.scale = 1.0
+
+    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
+        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)
+        hidden_states = self.scale * hidden_states
+        return hidden_states.to(input_dtype)
+
+
+class RotaryEmbedding(nn.Module):
+    def __init__(
+        self, dim: int, max_position_embeddings: int = 2048, base: int = 10000
+    ) -> None:
+        super().__init__()
+        assert dim % 2 == 0
+        self.dim = dim
+        self.max_position_embeddings = max_position_embeddings
+        self.base = base
+        inv_freq = 1.0 / (
+            self.base ** (torch.arange(0, self.dim, 2).float() / self.dim)
+        )
+        self.register_buffer("inv_freq", inv_freq, persistent=False)
+
+        self._set_cos_sin_cache(
+            seq_len=max_position_embeddings,
+            device=self.inv_freq.device,
+            dtype=torch.get_default_dtype(),
+        )
+
+    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=self.inv_freq.dtype
+        )
+
+        freqs = torch.outer(t, self.inv_freq)
+        # Different from paper, but it uses a different permutation in order to obtain the same calculation
+        emb = torch.cat((freqs, freqs), dim=-1)
+        self.register_buffer("cos_cached", emb.cos().to(dtype), persistent=False)
+        self.register_buffer("sin_cached", emb.sin().to(dtype), persistent=False)
+
+    def forward(self, x, seq_len=None):
+        # x: [bs, num_attention_heads, seq_len, head_size]
+        if seq_len > self.max_seq_len_cached:
+            self._set_cos_sin_cache(seq_len=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),
+        )
+
+
+# Copied from transformers.models.llama.modeling_llama.rotate_half
+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)
+
+
+# Copied from transformers.models.llama.modeling_llama.apply_rotary_pos_emb
+def apply_rotary_pos_emb(q, k, cos, sin, position_ids, unsqueeze_dim=1):
+    """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`):
+            The position indices of the tokens corresponding to the query and key tensors. For example, this can be
+            used to pass offsetted position ids when working with a KV-cache.
+        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[position_ids].unsqueeze(unsqueeze_dim)
+    sin = sin[position_ids].unsqueeze(unsqueeze_dim)
+    q_embed = (q * cos) + (rotate_half(q) * sin)
+    k_embed = (k * cos) + (rotate_half(k) * sin)
+    return q_embed, k_embed
+
+
+class MultiHeadAttention(nn.Module):
+    def __init__(
+        self,
+        hidden_size: int,
+        num_heads: int,
+        num_key_value_heads: Optional[int] = None,
+        max_position_embeddings: int = 2048,
+        attn_output_multiplier: float = 1.0,
+        max_attn_val: float = 30.0,
+    ):
+        super().__init__()
+        self.hidden_size = hidden_size
+        self.num_heads = num_heads
+        self.head_dim = hidden_size // num_heads
+        if num_key_value_heads is None:
+            num_key_value_heads = num_heads
+        self.num_key_value_heads = num_key_value_heads
+        self.attn_output_multiplier = attn_output_multiplier
+        self.max_attn_val = max_attn_val
+
+        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(hidden_size, self.num_heads * self.head_dim, bias=False)
+        self.k_proj = nn.Linear(
+            hidden_size, self.num_key_value_heads * self.head_dim, bias=False
+        )
+        self.v_proj = nn.Linear(
+            hidden_size, self.num_key_value_heads * self.head_dim, bias=False
+        )
+        self.o_proj = nn.Linear(self.num_heads * self.head_dim, hidden_size, bias=False)
+
+        self.rotary_emb = RotaryEmbedding(
+            self.head_dim,
+            max_position_embeddings=max_position_embeddings,
+        )
+
+    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,
+        **kwargs,
+    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
+        bsz, q_len, _ = hidden_states.size()
+
+        query_states = self.q_proj(hidden_states)
+        key_states = self.k_proj(hidden_states)
+        value_states = self.v_proj(hidden_states)
+
+        query_states = query_states.view(
+            bsz, q_len, self.num_heads, self.head_dim
+        ).transpose(1, 2)
+        key_states = key_states.view(
+            bsz, q_len, self.num_key_value_heads, self.head_dim
+        ).transpose(1, 2)
+        value_states = value_states.view(
+            bsz, q_len, 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
+
+        # TODO: repeat kv
+
+        attn_weights = torch.matmul(query_states, key_states.transpose(2, 3)).to(
+            torch.float
+        )
+        attn_weights = attn_weights * self.attn_output_multiplier
+        attn_weights = self.max_attn_val * F.tanh(attn_weights / self.max_attn_val)
+
+        if attn_weights.size() != (bsz, self.num_heads, q_len, kv_seq_len):
+            raise ValueError(
+                f"Attention weights should be of size {(bsz, self.num_heads, q_len, kv_seq_len)}, but is"
+                f" {attn_weights.size()}"
+            )
+
+        if attention_mask is not None:
+            if attention_mask.size() != (bsz, 1, q_len, kv_seq_len):
+                raise ValueError(
+                    f"Attention mask should be of size {(bsz, 1, q_len, kv_seq_len)}, but is {attention_mask.size()}"
+                )
+
+            attn_weights = attn_weights + attention_mask
+
+        attn_weights = F.softmax(attn_weights, dim=-1).to(query_states.dtype)
+        attn_output = torch.matmul(attn_weights, value_states)
+
+        if attn_output.size() != (bsz, self.num_heads, q_len, self.head_dim):
+            raise ValueError(
+                f"`attn_output` should be of size {(bsz, self.num_heads, q_len, self.head_dim)}, but is"
+                f" {attn_output.size()}"
+            )
+
+        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 MoeMLP(nn.Module):
+    def __init__(
+        self,
+        hidden_dim: int,
+        ffn_dim: int,
+    ) -> None:
+        super().__init__()
+        self.linear_v = nn.Linear(hidden_dim, ffn_dim, bias=False)
+        self.linear_1 = nn.Linear(ffn_dim, hidden_dim, bias=False)
+        self.linear = nn.Linear(hidden_dim, ffn_dim, bias=False)
+        self.act_fn = nn.GELU()
+
+    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
+        current_hidden_states = self.act_fn(self.linear(hidden_states)) * self.linear_v(
+            hidden_states
+        )
+        current_hidden_states = self.linear_1(current_hidden_states)
+        return current_hidden_states
+
+
+class MoeBlock(nn.Module):
+    def __init__(
+        self,
+        hidden_dim: int,
+        ffn_dim: int,
+        num_experts: int,
+        top_k: int,
+    ) -> None:
+        super().__init__()
+        self.num_experts = num_experts
+        self.top_k = top_k
+        self.gate = nn.Linear(hidden_dim, num_experts, bias=False)
+        self.experts = nn.ModuleList(
+            [MoeMLP(hidden_dim, ffn_dim) for _ in range(num_experts)]
+        )
+
+    def forward(self, hidden_states: torch.Tensor) -> Tuple[torch.Tensor]:
+        batch_size, sequence_length, hidden_dim = hidden_states.shape
+        hidden_states = hidden_states.view(-1, hidden_dim)
+        # router_logits: (batch * sequence_length, n_experts)
+        router_logits = self.gate(hidden_states)
+
+        routing_weights = F.softmax(router_logits, dim=1, dtype=torch.float)
+        routing_weights, selected_experts = torch.topk(
+            routing_weights, self.top_k, dim=-1
+        )
+        # we cast back to the input dtype
+        routing_weights = routing_weights.to(hidden_states.dtype)
+
+        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_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
+
+            # in torch it is faster to index using lists than torch tensors
+            top_x_list = top_x.tolist()
+            idx_list = idx.tolist()
+
+            # 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_list].reshape(-1, hidden_dim)
+            current_hidden_states = (
+                expert_layer(current_state)
+                * routing_weights[top_x_list, idx_list, None]
+            )
+
+            # 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 DecoderLayer(nn.Module):
+    def __init__(
+        self,
+        hidden_size: int,
+        num_heads: int,
+        num_key_value_heads: int,
+        num_experts: int,
+        top_k: int,
+        widening_factor: float = 4.0,
+        max_position_embeddings: int = 2048,
+        attn_output_multiplier: float = 1.0,
+        max_attn_val: float = 30.0,
+        rms_norm_eps: float = 1e-5,
+    ) -> None:
+        super().__init__()
+        self.attn = MultiHeadAttention(
+            hidden_size,
+            num_heads,
+            num_key_value_heads,
+            max_position_embeddings=max_position_embeddings,
+            attn_output_multiplier=attn_output_multiplier,
+            max_attn_val=max_attn_val,
+        )
+        ffn_dim = int(hidden_size * widening_factor)
+        self.moe_block = MoeBlock(hidden_size, ffn_dim, num_experts, top_k)
+        self.pre_attn_norm = RMSNorm(hidden_size, eps=rms_norm_eps)
+        self.post_attn_norm = RMSNorm(hidden_size, eps=rms_norm_eps)
+        self.pre_moe_norm = RMSNorm(hidden_size, eps=rms_norm_eps)
+        self.post_moe_norm = RMSNorm(hidden_size, eps=rms_norm_eps)
+
+    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: Optional[bool] = False,
+        output_router_logits: Optional[bool] = False,
+        use_cache: Optional[bool] = False,
+        **kwargs,
+    ) -> Tuple[
+        torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]
+    ]:
+        residual = hidden_states
+        hidden_states = self.pre_attn_norm(hidden_states)
+        hidden_states, attention_weights, present_key_value = self.attn(
+            hidden_states,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_value=past_key_value,
+            output_attentions=output_attentions,
+            use_cache=use_cache,
+        )
+        hidden_states = self.post_attn_norm(hidden_states)
+        hidden_states = residual + hidden_states
+
+        residual = hidden_states
+        hidden_states = self.pre_moe_norm(hidden_states)
+        hidden_states, router_logits = self.moe_block(hidden_states)
+        hidden_states = self.post_moe_norm(hidden_states)
+        hidden_states = residual + hidden_states
+
+        outputs = (hidden_states,)
+        if output_attentions:
+            outputs += (attention_weights,)
+        if use_cache:
+            outputs += (present_key_value,)
+        if output_router_logits:
+            outputs += (router_logits,)
+        return outputs
+
+
+class Grok1PretrainedModel(PreTrainedModel):
+    config_class = Grok1Config
+    base_model_prefix = "model"
+    supports_gradient_checkpointing = True
+    _no_split_modules = ["DecoderLayer"]
+    _skip_keys_device_placement = "past_key_values"
+    _supports_flash_attn_2 = False
+    _supports_cache_class = False
+
+    def _init_weights(self, module) -> None:
+        if isinstance(module, nn.Linear):
+            module.weight.data.zero_()
+            if module.bias is not None:
+                module.bias.data.zero_()
+        elif isinstance(module, nn.Embedding):
+            module.weight.data.zero_()
+
+
+# Copied from transformers.models.bart.modeling_bart._make_causal_mask
+def _make_causal_mask(
+    input_ids_shape: torch.Size,
+    dtype: torch.dtype,
+    device: torch.device,
+    past_key_values_length: int = 0,
+):
+    """
+    Make causal mask used for bi-directional self-attention.
+    """
+    bsz, tgt_len = input_ids_shape
+    mask = torch.full((tgt_len, tgt_len), torch.finfo(dtype).min, device=device)
+    mask_cond = torch.arange(mask.size(-1), device=device)
+    mask.masked_fill_(mask_cond < (mask_cond + 1).view(mask.size(-1), 1), 0)
+    mask = mask.to(dtype)
+
+    if past_key_values_length > 0:
+        mask = torch.cat(
+            [
+                torch.zeros(
+                    tgt_len, past_key_values_length, dtype=dtype, device=device
+                ),
+                mask,
+            ],
+            dim=-1,
+        )
+    return mask[None, None, :, :].expand(
+        bsz, 1, tgt_len, tgt_len + past_key_values_length
+    )
+
+
+# Copied from transformers.models.bart.modeling_bart._expand_mask
+def _expand_mask(mask: torch.Tensor, dtype: torch.dtype, tgt_len: Optional[int] = None):
+    """
+    Expands attention_mask from `[bsz, seq_len]` to `[bsz, 1, tgt_seq_len, src_seq_len]`.
+    """
+    bsz, src_len = mask.size()
+    tgt_len = tgt_len if tgt_len is not None else src_len
+
+    expanded_mask = mask[:, None, None, :].expand(bsz, 1, tgt_len, src_len).to(dtype)
+
+    inverted_mask = 1.0 - expanded_mask
+
+    return inverted_mask.masked_fill(
+        inverted_mask.to(torch.bool), torch.finfo(dtype).min
+    )
+
+
+class Grok1Model(Grok1PretrainedModel):
+    def __init__(self, config: Grok1Config) -> None:
+        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(
+            [
+                DecoderLayer(
+                    hidden_size=config.hidden_size,
+                    num_heads=config.num_attention_heads,
+                    num_key_value_heads=config.num_key_value_heads,
+                    num_experts=config.num_experts,
+                    top_k=config.num_experts_per_tok,
+                    widening_factor=config.widening_factor,
+                    max_position_embeddings=config.max_position_embeddings,
+                    attn_output_multiplier=config.attn_output_multiplier,
+                    max_attn_val=config.max_attn_value,
+                    rms_norm_eps=config.rms_norm_eps,
+                )
+                for layer_idx in range(config.num_hidden_layers)
+            ]
+        )
+        self.norm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.gradient_checkpointing = False
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.embed_tokens
+
+    def set_input_embeddings(self, value):
+        self.embed_tokens = value
+
+    # Copied from transformers.models.bart.modeling_bart.BartDecoder._prepare_decoder_attention_mask
+    def _prepare_decoder_attention_mask(
+        self, attention_mask, input_shape, inputs_embeds, past_key_values_length
+    ):
+        # create causal mask
+        # [bsz, seq_len] -> [bsz, 1, tgt_seq_len, src_seq_len]
+        combined_attention_mask = None
+        if input_shape[-1] > 1:
+            combined_attention_mask = _make_causal_mask(
+                input_shape,
+                inputs_embeds.dtype,
+                device=inputs_embeds.device,
+                past_key_values_length=past_key_values_length,
+            )
+
+        if attention_mask is not None:
+            # [bsz, seq_len] -> [bsz, 1, tgt_seq_len, src_seq_len]
+            expanded_attn_mask = _expand_mask(
+                attention_mask, inputs_embeds.dtype, tgt_len=input_shape[-1]
+            ).to(inputs_embeds.device)
+            combined_attention_mask = (
+                expanded_attn_mask
+                if combined_attention_mask is None
+                else expanded_attn_mask + combined_attention_mask
+            )
+
+        return combined_attention_mask
+
+    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_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 input_ids and inputs_embeds at the same time"
+            )
+        elif input_ids is not None:
+            batch_size, seq_length = input_ids.shape[:2]
+        elif inputs_embeds is not None:
+            batch_size, seq_length = inputs_embeds.shape[:2]
+        else:
+            raise ValueError("You have to specify either input_ids or inputs_embeds")
+
+        seq_length_with_past = seq_length
+        past_key_values_length = 0
+        if past_key_values is not None:
+            past_key_values_length = past_key_values[0][0].shape[2]
+            seq_length_with_past = seq_length_with_past + past_key_values_length
+
+        if position_ids is None:
+            device = input_ids.device if input_ids is not None else inputs_embeds.device
+            position_ids = torch.arange(
+                past_key_values_length,
+                seq_length + past_key_values_length,
+                dtype=torch.long,
+                device=device,
+            )
+            position_ids = position_ids.unsqueeze(0)
+
+        if inputs_embeds is None:
+            inputs_embeds = self.embed_tokens(input_ids)
+
+        if HAS_MASK_UTILS:
+            # 4d mask is passed through the layers
+            attention_mask = _prepare_4d_causal_attention_mask(
+                attention_mask,
+                (batch_size, seq_length),
+                inputs_embeds,
+                past_key_values_length,
+            )
+        else:
+            if attention_mask is None:
+                attention_mask = torch.ones(
+                    (batch_size, seq_length_with_past),
+                    dtype=torch.bool,
+                    device=inputs_embeds.device,
+                )
+            attention_mask = self._prepare_decoder_attention_mask(
+                attention_mask,
+                (batch_size, seq_length),
+                inputs_embeds,
+                past_key_values_length,
+            )
+
+        # embed positions
+        hidden_states = inputs_embeds
+
+        if self.gradient_checkpointing and self.training:
+            if use_cache:
+                logger.warning_once(
+                    "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
+                )
+                use_cache = False
+
+        # decoder layers
+        all_hidden_states = () if output_hidden_states else None
+        all_self_attns = () if output_attentions else None
+        all_router_logits = () if output_router_logits else None
+        next_decoder_cache = () if use_cache else None
+
+        for idx, decoder_layer in enumerate(self.layers):
+            if output_hidden_states:
+                all_hidden_states += (hidden_states,)
+
+            past_key_value = (
+                past_key_values[idx] if past_key_values is not None else None
+            )
+
+            if self.gradient_checkpointing and self.training:
+
+                def create_custom_forward(module):
+                    def custom_forward(*inputs):
+                        # None for past_key_value
+                        return module(*inputs, past_key_value, output_attentions)
+
+                    return custom_forward
+
+                layer_outputs = torch.utils.checkpoint.checkpoint(
+                    create_custom_forward(decoder_layer),
+                    hidden_states,
+                    attention_mask,
+                    position_ids,
+                )
+            else:
+                layer_outputs = decoder_layer(
+                    hidden_states,
+                    attention_mask=attention_mask,
+                    position_ids=position_ids,
+                    past_key_value=past_key_value,
+                    output_attentions=output_attentions,
+                    use_cache=use_cache,
+                )
+
+            hidden_states = layer_outputs[0]
+
+            if use_cache:
+                next_decoder_cache += (layer_outputs[2 if output_attentions else 1],)
+
+            if output_attentions:
+                all_self_attns += (layer_outputs[1],)
+
+            if output_router_logits:
+                all_router_logits += (layer_outputs[-1],)
+
+        hidden_states = self.norm(hidden_states)
+
+        # add hidden states from the last decoder layer
+        if output_hidden_states:
+            all_hidden_states += (hidden_states,)
+        next_cache = next_decoder_cache if use_cache else None
+
+        if not return_dict:
+            return tuple(
+                v
+                for v in [
+                    hidden_states,
+                    next_cache,
+                    all_hidden_states,
+                    all_self_attns,
+                    all_router_logits,
+                ]
+                if v is not None
+            )
+        return MoeModelOutputWithPast(
+            last_hidden_state=hidden_states,
+            past_key_values=next_cache,
+            hidden_states=all_hidden_states,
+            attentions=all_self_attns,
+            router_logits=all_router_logits,
+        )
+
+
+class Grok1ModelForCausalLM(Grok1PretrainedModel):
+    _tied_weights_keys = ["lm_head.weight"]
+
+    def __init__(self, config: Grok1Config):
+        super().__init__(config)
+        self.model = Grok1Model(config)
+        self.vocab_size = config.vocab_size
+        self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
+        self.router_aux_loss_coef = config.router_aux_loss_coef
+        self.num_experts = config.num_experts
+        self.num_experts_per_tok = config.num_experts_per_tok
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.model.embed_tokens
+
+    def set_input_embeddings(self, value):
+        self.model.embed_tokens = value
+
+    def get_output_embeddings(self):
+        return self.lm_head
+
+    def set_output_embeddings(self, new_embeddings):
+        self.lm_head = new_embeddings
+
+    def set_decoder(self, decoder):
+        self.model = decoder
+
+    def get_decoder(self):
+        return self.model
+
+    def forward(
+        self,
+        input_ids: torch.LongTensor = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_values: Optional[List[torch.FloatTensor]] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        labels: Optional[torch.LongTensor] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        output_router_logits: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[Tuple, MoeCausalLMOutputWithPast]:
+        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
+        )
+        return_dict = (
+            return_dict if return_dict is not None else self.config.use_return_dict
+        )
+
+        # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
+        outputs = self.model(
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_values=past_key_values,
+            inputs_embeds=inputs_embeds,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            output_router_logits=output_router_logits,
+            return_dict=return_dict,
+        )
+
+        hidden_states = outputs[0]
+        logits = self.lm_head(hidden_states)
+        logits = logits.float()
+
+        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 = nn.CrossEntropyLoss()
+            shift_logits = shift_logits.view(-1, self.config.vocab_size)
+            shift_labels = shift_labels.view(-1)
+            # Enable model parallelism
+            shift_labels = shift_labels.to(shift_logits.device)
+            loss = loss_fct(shift_logits, shift_labels)
+
+        aux_loss = None
+        if output_router_logits:
+            aux_loss = load_balancing_loss_func(
+                outputs.router_logits if return_dict else outputs[-1],
+                self.num_experts,
+                self.num_experts_per_tok,
+            )
+            if labels is not None:
+                loss += self.router_aux_loss_coef * aux_loss
+
+        if not return_dict:
+            output = (logits,) + outputs[1:]
+            if output_router_logits:
+                output = (aux_loss,) + output
+            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,
+        )
+
+    def prepare_inputs_for_generation(
+        self,
+        input_ids,
+        past_key_values=None,
+        attention_mask=None,
+        inputs_embeds=None,
+        **kwargs,
+    ):
+        if past_key_values:
+            input_ids = input_ids[:, -1:]
+
+        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)
+
+        # if `inputs_embeds` are passed, we only want to use them in the 1st generation step
+        if inputs_embeds is not None and past_key_values is None:
+            model_inputs = {"inputs_embeds": inputs_embeds}
+        else:
+            model_inputs = {"input_ids": input_ids}
+
+        model_inputs.update(
+            {
+                "position_ids": position_ids,
+                "past_key_values": past_key_values,
+                "use_cache": kwargs.get("use_cache"),
+                "attention_mask": attention_mask,
+            }
+        )
+        return model_inputs

modeling_grok1_outputs.py

@@ -0,0 +1,106 @@
+from dataclasses import dataclass
+from typing import Optional, Tuple
+
+import torch
+from transformers.modeling_outputs import ModelOutput
+
+__all__ = [
+    "MoeModelOutputWithPast",
+    "MoeCausalLMOutputWithPast",
+]
+
+try:
+    from transformers.modeling_outputs import (
+        MoeCausalLMOutputWithPast,
+        MoeModelOutputWithPast,
+    )
+except:
+
+    @dataclass
+    class MoeModelOutputWithPast(ModelOutput):
+        """
+        Base class for model's outputs, with potential hidden states and attentions.
+
+        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.
+            past_key_values (`tuple(tuple(torch.FloatTensor))`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
+                Tuple of `tuple(torch.FloatTensor)` of length `config.n_layers`, with each tuple having 2 tensors of shape
+                `(batch_size, num_heads, sequence_length, embed_size_per_head)`) and optionally if
+                `config.is_encoder_decoder=True` 2 additional tensors of shape `(batch_size, num_heads,
+                encoder_sequence_length, embed_size_per_head)`.
+
+                Contains pre-computed hidden-states (key and values in the self-attention blocks and optionally if
+                `config.is_encoder_decoder=True` in the cross-attention blocks) that can be used (see `past_key_values`
+                input) to speed up sequential decoding.
+            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.
+            attentions (`tuple(torch.FloatTensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
+                Tuple of `torch.FloatTensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
+                sequence_length)`.
+
+                Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
+                heads.
+            router_logits (`tuple(torch.FloatTensor)`, *optional*, returned when `output_router_probs=True` and `config.add_router_probs=True` is passed or when `config.output_router_probs=True`):
+                Tuple of `torch.FloatTensor` (one for each layer) of shape `(batch_size, sequence_length, num_experts)`.
+
+                Raw router logtis (post-softmax) that are computed by MoE routers, these terms are used to compute the auxiliary
+                loss for Mixture of Experts models.
+        """
+
+        last_hidden_state: torch.FloatTensor = None
+        past_key_values: Optional[Tuple[Tuple[torch.FloatTensor]]] = None
+        hidden_states: Optional[Tuple[torch.FloatTensor]] = None
+        attentions: Optional[Tuple[torch.FloatTensor]] = None
+        router_logits: Optional[Tuple[torch.FloatTensor]] = None
+
+    @dataclass
+    class MoeCausalLMOutputWithPast(ModelOutput):
+        """
+        Base class for causal language model (or autoregressive) with mixture of experts 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).
+
+            aux_loss (`torch.FloatTensor`, *optional*, returned when `labels` is provided):
+                aux_loss for the sparse modules.
+
+            router_logits (`tuple(torch.FloatTensor)`, *optional*, returned when `output_router_probs=True` and `config.add_router_probs=True` is passed or when `config.output_router_probs=True`):
+                Tuple of `torch.FloatTensor` (one for each layer) of shape `(batch_size, sequence_length, num_experts)`.
+
+                Raw router logtis (post-softmax) that are computed by MoE routers, these terms are used to compute the auxiliary
+                loss for Mixture of Experts models.
+
+            past_key_values (`tuple(tuple(torch.FloatTensor))`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
+                Tuple of `tuple(torch.FloatTensor)` of length `config.n_layers`, with each tuple having 2 tensors of shape
+                `(batch_size, num_heads, sequence_length, embed_size_per_head)`)
+
+                Contains pre-computed hidden-states (key and values in the self-attention blocks) that can be used (see
+                `past_key_values` input) to speed up sequential decoding.
+            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.
+            attentions (`tuple(torch.FloatTensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
+                Tuple of `torch.FloatTensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
+                sequence_length)`.
+
+                Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
+                heads.
+        """
+
+        loss: Optional[torch.FloatTensor] = None
+        aux_loss: Optional[torch.FloatTensor] = None
+        logits: torch.FloatTensor = None
+        past_key_values: Optional[Tuple[Tuple[torch.FloatTensor]]] = None
+        hidden_states: Optional[Tuple[torch.FloatTensor]] = None
+        attentions: Optional[Tuple[torch.FloatTensor]] = None
+        router_logits: Optional[Tuple[torch.FloatTensor]] = None