Update modeling_intern_vit.py

modeling_intern_vit.py

@@ -12,123 +12,22 @@ from einops import rearrange
 from timm.models.layers import DropPath
 from torch import nn
 from transformers.activations import ACT2FN
-from transformers.modeling_outputs import BaseModelOutput, BaseModelOutputWithPooling
+from transformers.modeling_outputs import (BaseModelOutput,
+                                           BaseModelOutputWithPooling)
 from transformers.modeling_utils import PreTrainedModel
 from transformers.utils import logging
 
 from .configuration_intern_vit import InternVisionConfig
 
-
 try:
-    from triton_flash_atn import _attention
-
-    from triton_bert_pading import pad_input, unpad_input
-
+    from .flash_attention import FlashAttention
     has_flash_attn = True
 except:
-    print("FlashAttention is not installed.")
+    print('FlashAttention is not installed.')
     has_flash_attn = False
 
-logger = logging.get_logger(__name__)
-
-
-class FlashAttention(nn.Module):
-    """Implement the scaled dot product attention with softmax.
-    Arguments
-    ---------
-        softmax_scale: The temperature to use for the softmax attention.
-                      (default: 1/sqrt(d_keys) where d_keys is computed at
-                      runtime)
-        attention_dropout: The dropout rate to apply to the attention
-                           (default: 0.0)
-    """
-
-    def __init__(
-        self, softmax_scale=None, attention_dropout=0.0, device=None, dtype=None
-    ):
-        super().__init__()
-        self.softmax_scale = softmax_scale
-        self.dropout_p = attention_dropout
 
-    def forward(
-        self,
-        qkv,
-        key_padding_mask=None,
-        causal=False,
-        cu_seqlens=None,
-        max_s=None,
-        need_weights=False,
-    ):
-        """Implements the multihead softmax attention.
-        Arguments
-        ---------
-            qkv: The tensor containing the query, key, and value. (B, S, 3, H, D) if key_padding_mask is None
-                if unpadded: (nnz, 3, h, d)
-            key_padding_mask: a bool tensor of shape (B, S)
-        """
-        assert not need_weights
-        assert qkv.dtype in [torch.float16, torch.bfloat16]
-        assert qkv.is_cuda
-
-        if cu_seqlens is None:
-            batch_size = qkv.shape[0]
-            seqlen = qkv.shape[1]
-            if key_padding_mask is None:
-                qkv = rearrange(qkv, "b s ... -> (b s) ...")
-                max_s = seqlen
-                cu_seqlens = torch.arange(
-                    0,
-                    (batch_size + 1) * seqlen,
-                    step=seqlen,
-                    dtype=torch.int32,
-                    device=qkv.device,
-                )
-                output = _attention.apply(
-                    qkv,
-                    cu_seqlens,
-                    max_s,
-                    self.dropout_p if self.training else 0.0,
-                    sm_scale=self.softmax_scale,
-                    causal=causal,
-                )
-                output = rearrange(output, "(b s) ... -> b s ...", b=batch_size)
-            else:
-                nheads = qkv.shape[-2]
-                x = rearrange(qkv, "b s three h d -> b s (three h d)")
-                x_unpad, indices, cu_seqlens, max_s = unpad_input(x, key_padding_mask)
-                x_unpad = rearrange(
-                    x_unpad, "nnz (three h d) -> nnz three h d", three=3, h=nheads
-                )
-                output_unpad = _attention.apply(
-                    x_unpad,
-                    cu_seqlens,
-                    max_s,
-                    self.dropout_p if self.training else 0.0,
-                    sm_scale=self.softmax_scale,
-                    causal=causal,
-                )
-                output = rearrange(
-                    pad_input(
-                        rearrange(output_unpad, "nnz h d -> nnz (h d)"),
-                        indices,
-                        batch_size,
-                        seqlen,
-                    ),
-                    "b s (h d) -> b s h d",
-                    h=nheads,
-                )
-        else:
-            assert max_s is not None
-            output = _attention.apply(
-                qkv,
-                cu_seqlens,
-                max_s,
-                self.dropout_p if self.training else 0.0,
-                sm_scale=self.softmax_scale,
-                causal=causal,
-            )
-
-        return output, None
+logger = logging.get_logger(__name__)
 
 
 class InternRMSNorm(nn.Module):
@@ -150,25 +49,15 @@ try:
 
     InternRMSNorm = FusedRMSNorm  # noqa
 
-    logger.info(
-        "Discovered apex.normalization.FusedRMSNorm - will use it instead of InternRMSNorm"
-    )
+    logger.info('Discovered apex.normalization.FusedRMSNorm - will use it instead of InternRMSNorm')
 except ImportError:
     # using the normal InternRMSNorm
     pass
 except Exception:
-    logger.warning(
-        "discovered apex but it failed to load, falling back to InternRMSNorm"
-    )
+    logger.warning('discovered apex but it failed to load, falling back to InternRMSNorm')
     pass
 
 
-NORM2FN = {
-    "rms_norm": InternRMSNorm,
-    "layer_norm": nn.LayerNorm,
-}
-
-
 class InternVisionEmbeddings(nn.Module):
     def __init__(self, config: InternVisionConfig):
         super().__init__()
@@ -182,55 +71,22 @@ class InternVisionEmbeddings(nn.Module):
         )
 
         self.patch_embedding = nn.Conv2d(
-            in_channels=3,
-            out_channels=self.embed_dim,
-            kernel_size=self.patch_size,
-            stride=self.patch_size,
+            in_channels=3, out_channels=self.embed_dim, kernel_size=self.patch_size, stride=self.patch_size
         )
 
         self.num_patches = (self.image_size // self.patch_size) ** 2
         self.num_positions = self.num_patches + 1
 
-        self.position_embedding = nn.Parameter(
-            torch.randn(1, self.num_positions, self.embed_dim)
-        )
-
-    def _get_pos_embed(self, pos_embed, H, W):
-        target_dtype = pos_embed.dtype
-        pos_embed = (
-            pos_embed.float()
-            .reshape(
-                1,
-                self.image_size // self.patch_size,
-                self.image_size // self.patch_size,
-                -1,
-            )
-            .permute(0, 3, 1, 2)
-        )
-        pos_embed = (
-            F.interpolate(pos_embed, size=(H, W), mode="bicubic", align_corners=False)
-            .reshape(1, -1, H * W)
-            .permute(0, 2, 1)
-            .to(target_dtype)
-        )
-        return pos_embed
+        self.position_embedding = nn.Parameter(torch.randn(1, self.num_positions, self.embed_dim))
 
     def forward(self, pixel_values: torch.FloatTensor) -> torch.Tensor:
+        batch_size = pixel_values.shape[0]
         target_dtype = self.patch_embedding.weight.dtype
-        # shape = [*, channel, width, height]
-        patch_embeds = self.patch_embedding(pixel_values)
-        batch_size, _, height, width = patch_embeds.shape
+        patch_embeds = self.patch_embedding(pixel_values)  # shape = [*, width, grid, grid]
         patch_embeds = patch_embeds.flatten(2).transpose(1, 2)
         class_embeds = self.class_embedding.expand(batch_size, 1, -1).to(target_dtype)
         embeddings = torch.cat([class_embeds, patch_embeds], dim=1)
-        position_embedding = torch.cat(
-            [
-                self.position_embedding[:, :1, :],
-                self._get_pos_embed(self.position_embedding[:, 1:, :], height, width),
-            ],
-            dim=1,
-        )
-        embeddings = embeddings + position_embedding.to(target_dtype)
+        embeddings = embeddings + self.position_embedding.to(target_dtype)
         return embeddings
 
 
@@ -244,17 +100,15 @@ class InternAttention(nn.Module):
         self.num_heads = config.num_attention_heads
         self.use_flash_attn = config.use_flash_attn and has_flash_attn
         if config.use_flash_attn and not has_flash_attn:
-            print(
-                "Warning: Flash Attention is not available, use_flash_attn is set to False."
-            )
+            print('Warning: Flash Attention is not available, use_flash_attn is set to False.')
         self.head_dim = self.embed_dim // self.num_heads
         if self.head_dim * self.num_heads != self.embed_dim:
             raise ValueError(
-                f"embed_dim must be divisible by num_heads (got `embed_dim`: {self.embed_dim} and `num_heads`:"
-                f" {self.num_heads})."
+                f'embed_dim must be divisible by num_heads (got `embed_dim`: {self.embed_dim} and `num_heads`:'
+                f' {self.num_heads}).'
             )
 
-        self.scale = self.head_dim**-0.5
+        self.scale = self.head_dim ** -0.5
         self.qkv = nn.Linear(self.embed_dim, 3 * self.embed_dim, bias=config.qkv_bias)
         self.attn_drop = nn.Dropout(config.attention_dropout)
         self.proj_drop = nn.Dropout(config.dropout)
@@ -271,28 +125,15 @@ class InternAttention(nn.Module):
 
     def _naive_attn(self, x):
         B, N, C = x.shape
-        qkv = (
-            self.qkv(x)
-            .reshape(B, N, 3, self.num_heads, C // self.num_heads)
-            .permute(2, 0, 3, 1, 4)
-        )
-        # make torchscript happy (cannot use tensor as tuple)
-        q, k, v = qkv.unbind(0)
+        qkv = self.qkv(x).reshape(B, N, 3, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4)
+        q, k, v = qkv.unbind(0)  # make torchscript happy (cannot use tensor as tuple)
 
         if self.qk_normalization:
             B_, H_, N_, D_ = q.shape
-            q = (
-                self.q_norm(q.transpose(1, 2).flatten(-2, -1))
-                .view(B_, N_, H_, D_)
-                .transpose(1, 2)
-            )
-            k = (
-                self.k_norm(k.transpose(1, 2).flatten(-2, -1))
-                .view(B_, N_, H_, D_)
-                .transpose(1, 2)
-            )
+            q = self.q_norm(q.transpose(1, 2).flatten(-2, -1)).view(B_, N_, H_, D_).transpose(1, 2)
+            k = self.k_norm(k.transpose(1, 2).flatten(-2, -1)).view(B_, N_, H_, D_).transpose(1, 2)
 
-        attn = (q * self.scale) @ k.transpose(-2, -1)
+        attn = ((q * self.scale) @ k.transpose(-2, -1))
         attn = attn.softmax(dim=-1)
         attn = self.attn_drop(attn)
 
@@ -303,9 +144,7 @@ class InternAttention(nn.Module):
 
     def _flash_attn(self, x, key_padding_mask=None, need_weights=False):
         qkv = self.qkv(x)
-        qkv = rearrange(
-            qkv, "b s (three h d) -> b s three h d", three=3, h=self.num_heads
-        )
+        qkv = rearrange(qkv, 'b s (three h d) -> b s three h d', three=3, h=self.num_heads)
 
         if self.qk_normalization:
             q, k, v = qkv.unbind(2)
@@ -314,21 +153,14 @@ class InternAttention(nn.Module):
             qkv = torch.stack([q, k, v], dim=2)
 
         context, _ = self.inner_attn(
-            qkv,
-            key_padding_mask=key_padding_mask,
-            need_weights=need_weights,
-            causal=False,
+            qkv, key_padding_mask=key_padding_mask, need_weights=need_weights, causal=False
         )
-        outs = self.proj(rearrange(context, "b s h d -> b s (h d)"))
+        outs = self.proj(rearrange(context, 'b s h d -> b s (h d)'))
         outs = self.proj_drop(outs)
         return outs
 
     def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
-        x = (
-            self._naive_attn(hidden_states)
-            if not self.use_flash_attn
-            else self._flash_attn(hidden_states)
-        )
+        x = self._naive_attn(hidden_states) if not self.use_flash_attn else self._flash_attn(hidden_states)
         return x
 
 
@@ -352,41 +184,28 @@ class InternVisionEncoderLayer(nn.Module):
         super().__init__()
         self.embed_dim = config.hidden_size
         self.intermediate_size = config.intermediate_size
-        self.norm_type = config.norm_type
 
         self.attn = InternAttention(config)
         self.mlp = InternMLP(config)
-        self.norm1 = NORM2FN[self.norm_type](self.embed_dim, eps=config.layer_norm_eps)
-        self.norm2 = NORM2FN[self.norm_type](self.embed_dim, eps=config.layer_norm_eps)
+        self.norm1 = InternRMSNorm(self.embed_dim, eps=config.layer_norm_eps)
+        self.norm2 = InternRMSNorm(self.embed_dim, eps=config.layer_norm_eps)
 
         self.ls1 = nn.Parameter(config.initializer_factor * torch.ones(self.embed_dim))
         self.ls2 = nn.Parameter(config.initializer_factor * torch.ones(self.embed_dim))
-        self.drop_path1 = (
-            DropPath(drop_path_rate) if drop_path_rate > 0.0 else nn.Identity()
-        )
-        self.drop_path2 = (
-            DropPath(drop_path_rate) if drop_path_rate > 0.0 else nn.Identity()
-        )
+        self.drop_path1 = DropPath(drop_path_rate) if drop_path_rate > 0. else nn.Identity()
+        self.drop_path2 = DropPath(drop_path_rate) if drop_path_rate > 0. else nn.Identity()
 
     def forward(
-        self,
-        hidden_states: torch.Tensor,
-    ) -> Tuple[
-        torch.FloatTensor,
-        Optional[torch.FloatTensor],
-        Optional[Tuple[torch.FloatTensor]],
-    ]:
+            self,
+            hidden_states: torch.Tensor,
+    ) -> Tuple[torch.FloatTensor, Optional[torch.FloatTensor], Optional[Tuple[torch.FloatTensor]]]:
         """
         Args:
             hidden_states (`Tuple[torch.FloatTensor, Optional[torch.FloatTensor]]`): input to the layer of shape `(batch, seq_len, embed_dim)`
         """
-        hidden_states = hidden_states + self.drop_path1(
-            self.attn(self.norm1(hidden_states)) * self.ls1
-        )
+        hidden_states = hidden_states + self.drop_path1(self.attn(self.norm1(hidden_states)) * self.ls1)
 
-        hidden_states = hidden_states + self.drop_path2(
-            self.mlp(self.norm2(hidden_states)) * self.ls2
-        )
+        hidden_states = hidden_states + self.drop_path2(self.mlp(self.norm2(hidden_states)) * self.ls2)
 
         return hidden_states
 
@@ -405,23 +224,16 @@ class InternVisionEncoder(nn.Module):
         super().__init__()
         self.config = config
         # stochastic depth decay rule
-        dpr = [
-            x.item()
-            for x in torch.linspace(0, config.drop_path_rate, config.num_hidden_layers)
-        ]
-        self.layers = nn.ModuleList(
-            [
-                InternVisionEncoderLayer(config, dpr[idx])
-                for idx in range(config.num_hidden_layers)
-            ]
-        )
+        dpr = [x.item() for x in torch.linspace(0, config.drop_path_rate, config.num_hidden_layers)]
+        self.layers = nn.ModuleList([
+            InternVisionEncoderLayer(config, dpr[idx]) for idx in range(config.num_hidden_layers)])
         self.gradient_checkpointing = True
 
     def forward(
-        self,
-        inputs_embeds,
-        output_hidden_states: Optional[bool] = None,
-        return_dict: Optional[bool] = None,
+            self,
+            inputs_embeds,
+            output_hidden_states: Optional[bool] = None,
+            return_dict: Optional[bool] = None,
     ) -> Union[Tuple, BaseModelOutput]:
         r"""
         Args:
@@ -434,13 +246,9 @@ class InternVisionEncoder(nn.Module):
                 Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
         """
         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
+            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
 
         encoder_states = () if output_hidden_states else None
         hidden_states = inputs_embeds
@@ -450,8 +258,8 @@ class InternVisionEncoder(nn.Module):
                 encoder_states = encoder_states + (hidden_states,)
             if self.gradient_checkpointing and self.training:
                 layer_outputs = torch.utils.checkpoint.checkpoint(
-                    encoder_layer, hidden_states
-                )
+                    encoder_layer,
+                    hidden_states)
             else:
                 layer_outputs = encoder_layer(
                     hidden_states,
@@ -469,9 +277,9 @@ class InternVisionEncoder(nn.Module):
 
 
 class InternVisionModel(PreTrainedModel):
-    main_input_name = "pixel_values"
+    main_input_name = 'pixel_values'
     config_class = InternVisionConfig
-    _no_split_modules = ["InternVisionEncoderLayer"]
+    _no_split_modules = ['InternVisionEncoderLayer']
 
     def __init__(self, config: InternVisionConfig):
         super().__init__(config)
@@ -484,46 +292,30 @@ class InternVisionModel(PreTrainedModel):
         pos_emb = self.embeddings.position_embedding
         _, num_positions, embed_dim = pos_emb.shape
         cls_emb = pos_emb[:, :1, :]
-        pos_emb = (
-            pos_emb[:, 1:, :]
-            .reshape(1, old_size // patch_size, old_size // patch_size, -1)
-            .permute(0, 3, 1, 2)
-        )
-        pos_emb = F.interpolate(
-            pos_emb.float(),
-            size=new_size // patch_size,
-            mode="bicubic",
-            align_corners=False,
-        )
+        pos_emb = pos_emb[:, 1:, :].reshape(1, old_size // patch_size, old_size // patch_size, -1).permute(0, 3, 1, 2)
+        pos_emb = F.interpolate(pos_emb.float(), size=new_size // patch_size, mode='bicubic', align_corners=False)
         pos_emb = pos_emb.to(cls_emb.dtype).reshape(1, embed_dim, -1).permute(0, 2, 1)
         pos_emb = torch.cat([cls_emb, pos_emb], dim=1)
         self.embeddings.position_embedding = nn.Parameter(pos_emb)
-        self.embeddings.image_size = new_size
-        logger.info(
-            "Resized position embeddings from {} to {}".format(old_size, new_size)
-        )
+        logger.info('Resized position embeddings from {} to {}'.format(old_size, new_size))
 
     def get_input_embeddings(self):
         return self.embeddings
 
     def forward(
-        self,
-        pixel_values: Optional[torch.FloatTensor] = None,
-        output_hidden_states: Optional[bool] = None,
-        return_dict: Optional[bool] = None,
-        pixel_embeds: Optional[torch.FloatTensor] = None,
+            self,
+            pixel_values: Optional[torch.FloatTensor] = None,
+            output_hidden_states: Optional[bool] = None,
+            return_dict: Optional[bool] = None,
+            pixel_embeds: Optional[torch.FloatTensor] = None,
     ) -> Union[Tuple, BaseModelOutputWithPooling]:
         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
+            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
 
         if pixel_values is None and pixel_embeds is None:
-            raise ValueError("You have to specify pixel_values or pixel_embeds")
+            raise ValueError('You have to specify pixel_values or pixel_embeds')
 
         if pixel_embeds is not None:
             hidden_states = pixel_embeds
@@ -531,7 +323,7 @@ class InternVisionModel(PreTrainedModel):
             if len(pixel_values.shape) == 4:
                 hidden_states = self.embeddings(pixel_values)
             else:
-                raise ValueError(f"wrong pixel_values size: {pixel_values.shape}")
+                raise ValueError(f'wrong pixel_values size: {pixel_values.shape}')
         encoder_outputs = self.encoder(
             inputs_embeds=hidden_states,
             output_hidden_states=output_hidden_states,