[Model] Remove transformers attention porting in VITs (#10414)

Signed-off-by: Isotr0py <2037008807@qq.com>

vllm/model_executor/models/blip.py

@@ -4,10 +4,11 @@ from typing import Iterable, Optional, Set, Tuple, Union
 
 import torch
 import torch.nn as nn
+import torch.nn.functional as F
 from PIL import Image
 from transformers import Blip2VisionConfig, BlipVisionConfig
-from transformers.models.blip.modeling_blip import BlipAttention
 
+from vllm.attention.selector import _Backend
 from vllm.config import ModelConfig
 from vllm.distributed import divide, get_tensor_model_parallel_world_size
 from vllm.inputs import DecoderOnlyInputs, token_inputs
@@ -21,11 +22,7 @@ from vllm.multimodal.utils import (cached_get_tokenizer,
                                    repeat_and_pad_placeholder_tokens)
 from vllm.sequence import SequenceData
 
-try:
-    from xformers import ops as xops
-    USE_XFORMERS_OPS = True
-except ImportError:
-    USE_XFORMERS_OPS = False
+from .utils import get_vit_attn_backend
 
 
 def get_blip_patch_grid_length(*, image_size: int, patch_size: int) -> int:
@@ -168,7 +165,7 @@ class BlipVisionEmbeddings(nn.Module):
         return embeddings
 
 
-class BlipParallelAttention(nn.Module):
+class BlipAttention(nn.Module):
     """Multi-headed attention from 'Attention Is All You Need' paper"""
 
     def __init__(
@@ -208,6 +205,12 @@ class BlipParallelAttention(nn.Module):
         self.tp_size = get_tensor_model_parallel_world_size()
         self.num_heads_per_partition = divide(self.num_heads, self.tp_size)
 
+        # Detect attention implementation.
+        self.attn_backend = get_vit_attn_backend(support_fa=False)
+        if self.attn_backend not in {_Backend.TORCH_SDPA, _Backend.XFORMERS}:
+            raise RuntimeError(
+                f"BLIP does not support {self.attn_backend} backend now.")
+
     def _shape(self, tensor: torch.Tensor, seq_len: int, bsz: int):
         return tensor.view(bsz, seq_len, self.num_heads,
                            self.head_dim).transpose(1, 2).contiguous()
@@ -231,11 +234,26 @@ class BlipParallelAttention(nn.Module):
                                          self.num_heads_per_partition,
                                          self.head_dim)
 
+        if self.attn_backend == _Backend.XFORMERS:
+            from xformers import ops as xops
+
             out = xops.memory_efficient_attention_forward(query_states,
                                                           key_states,
                                                           value_states,
                                                           p=self.dropout,
                                                           scale=self.scale)
+        elif self.attn_backend == _Backend.TORCH_SDPA:
+            query_states, key_states, value_states = (x.transpose(1, 2)
+                                                      for x in (query_states,
+                                                                key_states,
+                                                                value_states))
+            out = F.scaled_dot_product_attention(query_states,
+                                                 key_states,
+                                                 value_states,
+                                                 dropout_p=self.dropout,
+                                                 scale=self.scale)
+            out = out.transpose(1, 2)
+
         out = out.view(bsz, tgt_len, -1)
         attn_output, _ = self.projection(out)
 
@@ -285,18 +303,11 @@ class BlipEncoderLayer(nn.Module):
         super().__init__()
 
         # fallback to sdpa attention if tp unavailable
-        num_heads = config.num_attention_heads
-        tp_size = get_tensor_model_parallel_world_size()
-        if USE_XFORMERS_OPS and num_heads % tp_size == 0:
-            self.self_attn = BlipParallelAttention(
+        self.self_attn = BlipAttention(
             config,
             quant_config=quant_config,
             prefix=f"{prefix}.self_attn",
         )
-        else:
-            # Blip doesn't have SDPA attention implemented in transformers
-            # use eager attention instead for cpu backend
-            self.self_attn = BlipAttention(config)
         self.layer_norm1 = nn.LayerNorm(config.hidden_size,
                                         eps=config.layer_norm_eps)
         self.mlp = BlipMLP(config,
@@ -374,11 +385,6 @@ class BlipVisionModel(nn.Module):
         prefix: str = "",
     ) -> None:
         super().__init__()
-
-        tp_size = get_tensor_model_parallel_world_size()
-        num_heads = config.num_attention_heads
-        self.shard_weight = USE_XFORMERS_OPS and num_heads % tp_size == 0
-
         self.config = config
 
         self.embeddings = BlipVisionEmbeddings(config)
@@ -422,7 +428,7 @@ class BlipVisionModel(nn.Module):
             ("qkv_proj", "q_proj", "q"),
             ("qkv_proj", "k_proj", "k"),
             ("qkv_proj", "v_proj", "v"),
-        ] if self.shard_weight else []
+        ]
         params_dict = dict(self.named_parameters())
         loaded_params: Set[str] = set()
         layer_count = len(self.encoder.layers)

vllm/model_executor/models/clip.py

@@ -5,10 +5,11 @@ from typing import Iterable, List, Optional, Set, Tuple, Union
 import numpy as np
 import torch
 import torch.nn as nn
+import torch.nn.functional as F
 from PIL import Image
 from transformers import CLIPVisionConfig
-from transformers.models.clip.modeling_clip import CLIPSdpaAttention
 
+from vllm.attention.selector import _Backend
 from vllm.config import ModelConfig
 from vllm.distributed import divide, get_tensor_model_parallel_world_size
 from vllm.inputs import DecoderOnlyInputs, token_inputs
@@ -23,11 +24,7 @@ from vllm.multimodal.utils import (cached_get_tokenizer,
                                    repeat_and_pad_placeholder_tokens)
 from vllm.sequence import SequenceData
 
-try:
-    from xformers import ops as xops
-    USE_XFORMERS_OPS = True
-except ImportError:
-    USE_XFORMERS_OPS = False
+from .utils import get_vit_attn_backend
 
 
 def get_clip_patch_grid_length(*, image_size: int, patch_size: int) -> int:
@@ -197,7 +194,7 @@ class CLIPVisionEmbeddings(nn.Module):
         return embeddings
 
 
-class CLIPParallelAttention(nn.Module):
+class CLIPAttention(nn.Module):
     """Multi-headed attention from 'Attention Is All You Need' paper"""
 
     def __init__(
@@ -237,6 +234,12 @@ class CLIPParallelAttention(nn.Module):
         self.tp_size = get_tensor_model_parallel_world_size()
         self.num_heads_per_partition = divide(self.num_heads, self.tp_size)
 
+        # Detect attention implementation.
+        self.attn_backend = get_vit_attn_backend(support_fa=False)
+        if self.attn_backend not in {_Backend.TORCH_SDPA, _Backend.XFORMERS}:
+            raise RuntimeError(
+                f"CLIP does not support {self.attn_backend} backend now.")
+
     def _shape(self, tensor: torch.Tensor, seq_len: int, bsz: int):
         return tensor.view(bsz, seq_len, self.num_heads,
                            self.head_dim).transpose(1, 2).contiguous()
@@ -261,11 +264,26 @@ class CLIPParallelAttention(nn.Module):
                                          self.num_heads_per_partition,
                                          self.head_dim)
 
+        if self.attn_backend == _Backend.XFORMERS:
+            from xformers import ops as xops
+
             out = xops.memory_efficient_attention_forward(query_states,
                                                           key_states,
                                                           value_states,
                                                           p=self.dropout,
                                                           scale=self.scale)
+        elif self.attn_backend == _Backend.TORCH_SDPA:
+            query_states, key_states, value_states = (x.transpose(1, 2)
+                                                      for x in (query_states,
+                                                                key_states,
+                                                                value_states))
+            out = F.scaled_dot_product_attention(query_states,
+                                                 key_states,
+                                                 value_states,
+                                                 dropout_p=self.dropout,
+                                                 scale=self.scale)
+            out = out.transpose(1, 2)
+
         out = out.view(bsz, tgt_len, -1)
         attn_output, _ = self.out_proj(out)
 
@@ -311,17 +329,11 @@ class CLIPEncoderLayer(nn.Module):
         prefix: str = "",
     ) -> None:
         super().__init__()
-
-        num_heads = config.num_attention_heads
-        tp_size = get_tensor_model_parallel_world_size()
-        if USE_XFORMERS_OPS and num_heads % tp_size == 0:
-            self.self_attn = CLIPParallelAttention(
+        self.self_attn = CLIPAttention(
             config,
             quant_config=quant_config,
             prefix=f"{prefix}.self_attn",
         )
-        else:
-            self.self_attn = CLIPSdpaAttention(config)
         self.layer_norm1 = nn.LayerNorm(config.hidden_size,
                                         eps=config.layer_norm_eps)
         self.mlp = CLIPMLP(config,
@@ -461,11 +473,6 @@ class CLIPVisionModel(nn.Module):
         prefix: str = "",
     ) -> None:
         super().__init__()
-
-        tp_size = get_tensor_model_parallel_world_size()
-        num_heads = config.num_attention_heads
-        self.shard_weight = USE_XFORMERS_OPS and num_heads % tp_size == 0
-
         self.vision_model = CLIPVisionTransformer(
             config=config,
             quant_config=quant_config,
@@ -490,7 +497,7 @@ class CLIPVisionModel(nn.Module):
             ("qkv_proj", "q_proj", "q"),
             ("qkv_proj", "k_proj", "k"),
             ("qkv_proj", "v_proj", "v"),
-        ] if self.shard_weight else []
+        ]
         params_dict = dict(self.named_parameters())
         loaded_params: Set[str] = set()
         layer_count = len(self.vision_model.encoder.layers)

vllm/model_executor/models/intern_vit.py

@@ -12,6 +12,7 @@ import torch.nn as nn
 import torch.nn.functional as F
 from transformers import PretrainedConfig
 
+from vllm.attention.selector import _Backend
 from vllm.distributed import (divide, get_tensor_model_parallel_rank,
                               get_tensor_model_parallel_world_size,
                               split_tensor_along_last_dim,
@@ -24,11 +25,7 @@ from vllm.model_executor.layers.linear import (ColumnParallelLinear,
 from vllm.model_executor.layers.quantization import QuantizationConfig
 from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 
-try:
-    from xformers import ops as xops
-    USE_XFORMERS_OPS = True
-except ImportError:
-    USE_XFORMERS_OPS = False
+from .utils import get_vit_attn_backend
 
 NORM2FN = {
     'rms_norm': RMSNorm,
@@ -186,6 +183,11 @@ class InternParallelAttention(nn.Module):
             prefix=f"{prefix}.proj",
         )
 
+        self.attn_backend = get_vit_attn_backend(support_fa=False)
+        if self.attn_backend not in {_Backend.TORCH_SDPA, _Backend.XFORMERS}:
+            raise RuntimeError(
+                f"InternViT does not support {self.attn_backend} backend now.")
+
     def _apply_qk_norm(self, q: torch.Tensor, k: torch.Tensor):
         if self.tp_size > 1:
             q = tensor_model_parallel_all_gather(q.contiguous())
@@ -211,11 +213,21 @@ class InternParallelAttention(nn.Module):
         k = k.view(B, N, self.num_heads_per_partition, self.head_dim)
         v = v.view(B, N, self.num_heads_per_partition, self.head_dim)
 
-        x = xops.memory_efficient_attention_forward(q, k, v, scale=self.scale)
-        x = x.view(B, N, -1)
+        if self.attn_backend == _Backend.XFORMERS:
+            from xformers import ops as xops
 
-        x, _ = self.proj(x)
-        return x
+            out = xops.memory_efficient_attention_forward(q,
+                                                          k,
+                                                          v,
+                                                          scale=self.scale)
+        elif self.attn_backend == _Backend.TORCH_SDPA:
+            q, k, v = (x.transpose(1, 2) for x in (q, k, v))
+            out = F.scaled_dot_product_attention(q, k, v, scale=self.scale)
+            out = out.transpose(1, 2)
+
+        out = out.view(B, N, -1)
+        out, _ = self.proj(out)
+        return out
 
 
 class InternSdpaAttention(nn.Module):
@@ -362,7 +374,7 @@ class InternVisionEncoderLayer(nn.Module):
         tp_size = get_tensor_model_parallel_world_size()
         num_heads = config.num_attention_heads
 
-        if USE_XFORMERS_OPS and (num_heads + num_dummy_heads) % tp_size == 0:
+        if (num_heads + num_dummy_heads) % tp_size == 0:
             return InternParallelAttention(config,
                                            quant_config=quant_config,
                                            num_dummy_heads=num_dummy_heads,

vllm/model_executor/models/molmo.py

@@ -187,7 +187,7 @@ class MultiHeadDotProductAttention(nn.Module):
         )
 
         # Detect attention implementation.
-        self.attn_backend: _Backend = get_vit_attn_backend()
+        self.attn_backend: _Backend = get_vit_attn_backend(support_fa=True)
         if self.attn_backend not in {
                 _Backend.FLASH_ATTN, _Backend.TORCH_SDPA, _Backend.XFORMERS
         }:

vllm/model_executor/models/qwen2_vl.py

@@ -260,7 +260,7 @@ class Qwen2VisionAttention(nn.Module):
                                       prefix=f"{prefix}.proj")
 
         # Detect attention implementation.
-        self.attn_backend: _Backend = get_vit_attn_backend()
+        self.attn_backend: _Backend = get_vit_attn_backend(support_fa=True)
         if self.attn_backend not in {
                 _Backend.FLASH_ATTN, _Backend.TORCH_SDPA, _Backend.XFORMERS
         }:

vllm/model_executor/models/siglip.py

@@ -6,11 +6,12 @@ from typing import Iterable, List, Optional, Set, Tuple, Union
 
 import numpy as np
 import torch
+import torch.nn.functional as F
 from PIL import Image
 from torch import nn
 from transformers import SiglipVisionConfig
-from transformers.models.siglip.modeling_siglip import SiglipSdpaAttention
 
+from vllm.attention.selector import _Backend
 from vllm.config import ModelConfig
 from vllm.distributed import divide, get_tensor_model_parallel_world_size
 from vllm.inputs import DecoderOnlyInputs, token_inputs
@@ -27,11 +28,7 @@ from vllm.multimodal.utils import (cached_get_tokenizer,
                                    repeat_and_pad_placeholder_tokens)
 from vllm.sequence import SequenceData
 
-try:
-    from xformers import ops as xops
-    USE_XFORMERS_OPS = True
-except ImportError:
-    USE_XFORMERS_OPS = False
+from .utils import get_vit_attn_backend
 
 
 def get_siglip_patch_grid_length(*, image_size: int, patch_size: int) -> int:
@@ -254,7 +251,7 @@ class SiglipVisionEmbeddings(nn.Module):
         return embeddings
 
 
-class SiglipParallelAttention(nn.Module):
+class SiglipAttention(nn.Module):
 
     def __init__(
         self,
@@ -293,6 +290,11 @@ class SiglipParallelAttention(nn.Module):
         self.tp_size = get_tensor_model_parallel_world_size()
         self.num_heads_per_partition = divide(self.num_heads, self.tp_size)
 
+        self.attn_backend = get_vit_attn_backend(support_fa=False)
+        if self.attn_backend not in {_Backend.TORCH_SDPA, _Backend.XFORMERS}:
+            raise RuntimeError(
+                f"SIGLIP does not support {self.attn_backend} backend now.")
+
     def forward(
         self,
         hidden_states: torch.Tensor,
@@ -313,11 +315,26 @@ class SiglipParallelAttention(nn.Module):
                                          self.num_heads_per_partition,
                                          self.head_dim)
 
+        if self.attn_backend == _Backend.XFORMERS:
+            from xformers import ops as xops
+
             out = xops.memory_efficient_attention_forward(query_states,
                                                           key_states,
                                                           value_states,
                                                           p=self.dropout,
                                                           scale=self.scale)
+        elif self.attn_backend == _Backend.TORCH_SDPA:
+            query_states, key_states, value_states = (x.transpose(1, 2)
+                                                      for x in (query_states,
+                                                                key_states,
+                                                                value_states))
+            out = F.scaled_dot_product_attention(query_states,
+                                                 key_states,
+                                                 value_states,
+                                                 dropout_p=self.dropout,
+                                                 scale=self.scale)
+            out = out.transpose(1, 2)
+
         out = out.view(batch_size, q_len, -1)
         attn_output, _ = self.out_proj(out)
 
@@ -372,17 +389,11 @@ class SiglipEncoderLayer(nn.Module):
 
         self.embed_dim = config.hidden_size
 
-        num_heads = config.num_attention_heads
-        tp_size = get_tensor_model_parallel_world_size()
-        if USE_XFORMERS_OPS and num_heads % tp_size == 0:
-            self.self_attn = SiglipParallelAttention(
+        self.self_attn = SiglipAttention(
             config,
             quant_config=quant_config,
             prefix=f"{prefix}.self_attn",
         )
-        else:
-            self.self_attn = SiglipSdpaAttention(config)
-
         self.layer_norm1 = nn.LayerNorm(self.embed_dim,
                                         eps=config.layer_norm_eps)
         self.mlp = SiglipMLP(
@@ -569,10 +580,6 @@ class SiglipVisionModel(nn.Module):
     ) -> None:
         super().__init__()
 
-        num_heads = config.num_attention_heads
-        tp_size = get_tensor_model_parallel_world_size()
-        self.shard_weight = USE_XFORMERS_OPS and num_heads % tp_size == 0
-
         self.vision_model = SiglipVisionTransformer(
             config,
             quant_config,
@@ -601,7 +608,7 @@ class SiglipVisionModel(nn.Module):
             ("qkv_proj", "q_proj", "q"),
             ("qkv_proj", "k_proj", "k"),
             ("qkv_proj", "v_proj", "v"),
-        ] if self.shard_weight else []
+        ]
         params_dict = dict(self.named_parameters())
         loaded_params: Set[str] = set()
         layer_count = len(self.vision_model.encoder.layers)

vllm/model_executor/models/utils.py

@@ -587,7 +587,11 @@ class LLMWrapper(nn.Module):
         return llm(*args, **kwargs)
 
 
-def get_vit_attn_backend() -> _Backend:
+def get_vit_attn_backend(support_fa: bool = False) -> _Backend:
+    """
+    Get the available attention backend for Vision Transformer.
+    """
+    # TODO(Isotr0py): Remove `support_fa` after support FA for all ViTs attn.
     selected_backend: Optional[_Backend] = get_global_forced_attn_backend()
     if selected_backend is None:
         backend_by_env_var: Optional[str] = envs.VLLM_ATTENTION_BACKEND
@@ -596,7 +600,7 @@ def get_vit_attn_backend() -> _Backend:
     if selected_backend is None:
         # For Volta and Turing GPUs, use xformers instead.
         device_available = current_platform.has_device_capability(80)
-        if device_available:
+        if device_available and support_fa:
             from transformers.utils import is_flash_attn_2_available
             if is_flash_attn_2_available():
                 selected_backend = _Backend.FLASH_ATTN
@@ -606,7 +610,8 @@ def get_vit_attn_backend() -> _Backend:
                     "so we use xformers backend instead. You can run "
                     "`pip install flash-attn` to use flash-attention backend.")
                 selected_backend = _Backend.XFORMERS
-        elif current_platform.is_cpu():
+        elif current_platform.is_cpu() or current_platform.is_rocm():
+            # ROCM doesn't support xformers
             selected_backend = _Backend.TORCH_SDPA
         else:
             selected_backend = _Backend.XFORMERS