[shardformer] update colo attention to support custom mask (#5510)

* [feature] refactor colo attention (#5462)

* [extension] update api

* [feature] add colo attention

* [feature] update sdpa

* [feature] update npu attention

* [feature] update flash-attn

* [test] add flash attn test

* [test] update flash attn test

* [shardformer] update modeling to fit colo attention (#5465)

* [misc] refactor folder structure

* [shardformer] update llama flash-attn

* [shardformer] fix llama policy

* [devops] update tensornvme install

* [test] update llama test

* [shardformer] update colo attn kernel dispatch

* [shardformer] update blip2

* [shardformer] update chatglm

* [shardformer] update gpt2

* [shardformer] update gptj

* [shardformer] update opt

* [shardformer] update vit

* [shardformer] update colo attention mask prep

* [shardformer] update whisper

* [test] fix shardformer tests (#5514)

* [test] fix shardformer tests

* [test] fix shardformer tests

.github/workflows/build_on_pr.yml

@@ -117,7 +117,7 @@ jobs:
           cd TensorNVMe
           conda install cmake
           pip install -r requirements.txt
-          pip install -v .
+          DISABLE_URING=1 pip install -v .
 
       - name: Store TensorNVMe Cache
         run: |
@@ -201,4 +201,4 @@ jobs:
         uses: actions/upload-artifact@v3
         with:
           name: report
-          path: report/
+          path: report/
\ No newline at end of file

.github/workflows/build_on_schedule.yml

@@ -44,7 +44,7 @@ jobs:
           cd TensorNVMe
           conda install cmake
           pip install -r requirements.txt
-          pip install -v .
+          DISABLE_URING=1 pip install -v .
 
       - uses: actions/checkout@v2
         if: steps.check-avai.outputs.avai == 'true'

.github/workflows/compatiblity_test_on_dispatch.yml

@@ -66,7 +66,7 @@ jobs:
           cd TensorNVMe
           apt update && apt install -y cmake
           pip install -r requirements.txt
-          pip install -v .
+          DISABLE_URING=1 pip install -v .
       - uses: actions/checkout@v2
         with:
           ssh-key: ${{ secrets.SSH_KEY_FOR_CI }}

.github/workflows/compatiblity_test_on_pr.yml

@@ -60,7 +60,7 @@ jobs:
           cd TensorNVMe
           apt update && apt install -y cmake
           pip install -r requirements.txt
-          pip install -v .
+          DISABLE_URING=1 pip install -v .
       - uses: actions/checkout@v2
         with:
           ssh-key: ${{ secrets.SSH_KEY_FOR_CI }}

.github/workflows/compatiblity_test_on_schedule.yml

@@ -56,7 +56,7 @@ jobs:
           cd TensorNVMe
           apt update && apt install -y cmake
           pip install -r requirements.txt
-          pip install -v .
+          DISABLE_URING=1 pip install -v .
       - uses: actions/checkout@v2
         with:
           ssh-key: ${{ secrets.SSH_KEY_FOR_CI }}

colossalai/kernel/kernel_loader.py

@@ -6,7 +6,7 @@ from .extensions import (
     CpuAdamX86Extension,
     FlashAttentionDaoCudaExtension,
     FlashAttentionNpuExtension,
-    FlashAttentionXformersCudaExtension,
+    FlashAttentionSdpaCudaExtension,
     FusedOptimizerCudaExtension,
     LayerNormCudaExtension,
     MoeCudaExtension,
@@ -65,9 +65,9 @@ class KernelLoader:
         else:
             usable_exts = []
             for ext in exts:
-                if ext.is_hardware_available():
+                if ext.is_available():
                     # make sure the machine is compatible during kernel loading
-                    ext.assert_hardware_compatible()
+                    ext.assert_compatible()
                     usable_exts.append(ext)
 
         assert len(usable_exts) != 0, f"No usable kernel found for {self.__class__.__name__} on the current machine."
@@ -106,4 +106,20 @@ class ScaledUpperTriangleMaskedSoftmaxLoader(KernelLoader):
 
 
 class FlashAttentionLoader(KernelLoader):
-    REGISTRY = [FlashAttentionNpuExtension, FlashAttentionDaoCudaExtension, FlashAttentionXformersCudaExtension]
+    REGISTRY = [
+        FlashAttentionNpuExtension,
+        FlashAttentionDaoCudaExtension,
+        FlashAttentionSdpaCudaExtension,
+    ]
+
+
+class FlashAttentionWithPaddingMaskLoader(KernelLoader):
+    REGISTRY = [FlashAttentionNpuExtension, FlashAttentionDaoCudaExtension]
+
+
+class FlashAttentionWithCustomMaskLoader(KernelLoader):
+    REGISTRY = [FlashAttentionNpuExtension, FlashAttentionSdpaCudaExtension]
+
+
+class FlashAttentionForFloatAndCustomMaskLoader(KernelLoader):
+    REGISTRY = [FlashAttentionSdpaCudaExtension]

colossalai/nn/layer/colo_attention.py

-import enum
-import math
-import warnings
-from dataclasses import dataclass
-from typing import Iterable, Optional, Tuple
-
-import torch
-import torch.nn.functional as F
-from einops import rearrange
-
-from colossalai.accelerator import get_accelerator
-from colossalai.kernel.kernel_loader import FlashAttentionLoader
-
-
-@dataclass
-class SeqLenInfo:
-    seqlens: Iterable[int] = None
-    indices: torch.Tensor = None
-    max_seqlen: int = None
-    cu_seqlens: torch.Tensor = None
-
-    @staticmethod
-    def materialize(
-        attn_mask: torch.Tensor = None, size: Tuple[int] = None, device=get_accelerator().get_current_device()
-    ):
-        if attn_mask is not None:
-            indices = torch.nonzero(attn_mask.flatten(), as_tuple=False).flatten().to(device)
-            seqlens = attn_mask.sum(dim=-1, dtype=torch.int32).flatten()
-        else:
-            batch_size, tgt_len = size[0], size[1]
-            indices = torch.arange(batch_size * tgt_len, dtype=torch.long, device=device)
-            seqlens = torch.LongTensor([tgt_len] * batch_size, device=device)
-        max_seqlen = max(seqlens)
-        cu_seqlens = F.pad(torch.cumsum(seqlens, dim=0, dtype=torch.int32), (1, 0)).to(device)
-        return SeqLenInfo(seqlens.tolist(), indices, max_seqlen, cu_seqlens)
-
-
-class AttnMaskType(enum.Enum):
-    padding = 1
-    causal = 2
-    paddedcausal = 3
-
-
-class Unpad(torch.autograd.Function):
-    """
-    Adapted from
-    https://github.com/HazyResearch/flash-attention/blob/main/flash_attn/bert_padding.py
-    """
-
-    @staticmethod
-    def forward(ctx, tensor: torch.Tensor, indices: torch.Tensor):
-        ctx.save_for_backward(indices)
-        # [b, s, ...]
-        assert tensor.ndim >= 3
-        ctx.bsz = tensor.shape[0]
-        out = rearrange(tensor, "b s ... -> (b s) ...")
-        ctx.shape = out.shape
-        # [ntokens, ...]
-        return out[indices]
-
-    @staticmethod
-    def backward(ctx, grad_output):
-        (indices,) = ctx.saved_tensors
-        # [ntokens, ...]
-        grad = torch.zeros(ctx.shape, dtype=grad_output.dtype, device=grad_output.device)
-        grad[indices] = grad_output
-        grad = rearrange(grad, "(b s) ... -> b s ...", b=ctx.bsz)
-        # [b, s, ...]
-        return grad, None
-
-
-class Repad(torch.autograd.Function):
-    """
-    Adapted from
-    https://github.com/HazyResearch/flash-attention/blob/main/flash_attn/bert_padding.py
-    """
-
-    @staticmethod
-    def forward(ctx, tensor: torch.Tensor, indices: torch.Tensor, batch_size: int, seq_len: int):
-        ctx.save_for_backward(indices)
-        # [ntokens, ...]
-        tensor = tensor
-        out = torch.zeros((batch_size * seq_len, *tensor.shape[1:]), dtype=tensor.dtype, device=tensor.device)
-        # [b*s, ...]
-        out[indices] = tensor
-        return out
-
-    @staticmethod
-    def backward(ctx, grad_output):
-        (indices,) = ctx.saved_tensors
-        # [b*s, ...]
-        grad = grad_output[indices]
-        # [ntokens, ...]
-        return grad, None, None, None
-
-
-class ColoAttention(torch.nn.Module):
-    def __init__(self, embed_dim: int, num_heads: int, dropout: float = 0.0, scale=None):
-        super().__init__()
-        assert (
-            embed_dim % num_heads == 0
-        ), f"the embed dim ({embed_dim}) is not divisible by the number of attention heads ({num_heads})."
-        if scale is not None:
-            self.scale = scale
-        else:
-            self.scale = 1 / math.sqrt(embed_dim // num_heads)
-        self.dropout = dropout
-
-        self.attn = FlashAttentionLoader().load()
-
-    @staticmethod
-    def unpad(tensor: torch.Tensor, indices: torch.Tensor) -> torch.Tensor:
-        return Unpad.apply(tensor, indices)
-
-    @staticmethod
-    def repad(tensor: torch.Tensor, indices: torch.Tensor, batch_size: int, seq_len: int) -> torch.Tensor:
-        return Repad.apply(tensor, indices, batch_size, seq_len)
-
-    def forward(
-        self,
-        query: torch.Tensor,
-        key: torch.Tensor,
-        value: torch.Tensor,
-        attn_mask: Optional[torch.Tensor] = None,
-        origin_attn_mask: Optional[torch.Tensor] = None,
-        attn_mask_type: Optional[AttnMaskType] = None,
-        bias: Optional[torch.Tensor] = None,
-    ):
-        """
-        ColoAttention
-
-        Args:
-            q: (batch, q_seqlen, nheads, headdim)
-            k: (batch, kv_seqlen, nheads, headdim)
-            v: (batch, kv_seqlen, nheads, headdim)
-            origin_attn_mask: (nheads, q_seqlen, kv_seqlen)
-            bias: will not be used
-        Return:
-            attn_out: (batch, q_seqlen, nheads, headdim).
-        """
-        # if flash attention is not applicable, switch to memory effcient attention
-        if self.attn.__name__ == "flash_attention" and (
-            query.dtype not in [torch.float16, torch.bfloat16] or bias != None
-        ):
-            warnings.warn(
-                f"flash-attn expects fp16 or bf16 but got {query.dtype}, switching to xformers' implementation."
-            )
-            self.attn = FlashAttentionLoader().load(ext_name="flash_attention_xformers_cuda")
-
-        padded = attn_mask_type is not None and attn_mask_type.value % 2 == 1
-        causal = attn_mask_type is not None and attn_mask_type.value > 1
-
-        batch_size, tgt_len, src_len = query.shape[0], query.shape[1], key.shape[1]
-        # unpad
-        seq_len_info_q = None
-        seq_len_info_kv = None
-        if padded:
-            # bert style, unpad process
-            assert (
-                attn_mask is not None
-            ), f"attention mask {attn_mask} is not valid for attention mask type {attn_mask_type}."
-            assert attn_mask.dim() == 2, (
-                "attention mask is supposed to have shape (batch_size, seq_len), "
-                + f"but got {attn_mask.dim()} dimensions."
-            )
-
-            # bert style
-            if tgt_len == src_len:
-                seq_len_info_q = SeqLenInfo.materialize(attn_mask=attn_mask, device=query.device)
-                if batch_size > 1:
-                    query, key, value = self.unpad(
-                        torch.stack([query, key, value], dim=2), seq_len_info_q.indices
-                    ).unbind(dim=1)
-                else:
-                    query, key, value = torch.stack([query, key, value], dim=2).squeeze(0).unbind(dim=1)
-                seq_len_info_kv = seq_len_info_q
-            else:
-                seq_len_info_q = SeqLenInfo.materialize(size=(batch_size, tgt_len), device=query.device)
-                seq_len_info_kv = SeqLenInfo.materialize(attn_mask=attn_mask, device=query.device)
-                if batch_size > 1:
-                    query = rearrange(query, "b s ... -> c (b s) ...", c=1)
-                    key, value = self.unpad(torch.stack([query, key, value], dim=2), seq_len_info_kv.indices).unbind(
-                        dim=1
-                    )
-                else:
-                    query, key, value = torch.stack([query, key, value], dim=2).squeeze(0).unbind(dim=1)
-
-        out = self.attn(
-            query,
-            key,
-            value,
-            seq_len_info_q=seq_len_info_q,
-            seq_len_info_kv=seq_len_info_kv,
-            origin_attn_mask=origin_attn_mask,
-            dropout_p=self.dropout,
-            scale=self.scale,
-            causal=causal,
-            padded=padded,
-        )
-
-        # repad
-        if padded:
-            if batch_size > 1:
-                out = self.repad(out, seq_len_info_q.indices, batch_size, tgt_len)
-            out = rearrange(out, "(b s) h d -> b s h d", b=batch_size)
-
-        if len(out.shape) == 4:
-            out = rearrange(out, "b s h d -> b s (h d)")
-        return out

colossalai/shardformer/layer/__init__.py

+from .attn import AttnMaskType, ColoAttention
 from .dropout import DropoutForParallelInput, DropoutForReplicatedInput
 from .embedding import Embedding1D, VocabParallelEmbedding1D
 from .linear import Linear1D_Col, Linear1D_Row
@@ -23,4 +24,6 @@ __all__ = [
     "FusedRMSNorm",
     "FusedLinear1D_Col",
     "ParallelModule",
+    "AttnMaskType",
+    "ColoAttention",
 ]

colossalai/shardformer/layer/attn.py

+from enum import Enum
+from typing import Callable, Dict, Optional, Tuple
+
+import torch
+import torch.nn.functional as F
+
+from colossalai.kernel.kernel_loader import (
+    FlashAttentionForFloatAndCustomMaskLoader,
+    FlashAttentionLoader,
+    FlashAttentionWithCustomMaskLoader,
+    FlashAttentionWithPaddingMaskLoader,
+    KernelLoader,
+)
+
+__all__ = [
+    "AttnMaskType",
+    "ColoAttention",
+]
+
+
+class AttnMaskType(Enum):
+    CUSTOM = 0
+    PADDED = 1
+    CAUSAL = 2
+    PADDED_CAUSAL = 3
+
+
+def invert_mask(mask: torch.Tensor) -> torch.Tensor:
+    """Invert the mask tensor.
+
+    Args:
+        mask (torch.Tensor): Mask tensor. Shape should be [B, 1, Sq, Skv]
+
+    Returns:
+        torch.Tensor: Inverted mask tensor.
+    """
+    inverted_mask = 1.0 - mask
+    return inverted_mask.masked_fill(inverted_mask.bool(), torch.finfo(mask.dtype).min)
+
+
+# adapted from https://github.com/Dao-AILab/flash-attention/blob/main/flash_attn/bert_padding.py
+def get_pad_info(padding_mask: torch.Tensor) -> Tuple[int, torch.Tensor, torch.Tensor]:
+    """Get padding information from padding mask.
+
+    Args:
+        padding_mask (torch.Tensor): Padding mask tensor. Shape should be [B, S]
+
+    Returns:
+        Tuple[int, torch.Tensor, torch.Tensor]: Tuple of (max_seq_len, cu_seqlens, indices)
+    """
+    seqlens_in_batch = padding_mask.sum(dim=-1, dtype=torch.int32)
+    indices = torch.nonzero(padding_mask.flatten(), as_tuple=False).flatten()
+    max_seqlen_in_batch = seqlens_in_batch.max().item()
+    cu_seqlens = F.pad(torch.cumsum(seqlens_in_batch, dim=0, dtype=torch.torch.int32), (1, 0))
+    return max_seqlen_in_batch, cu_seqlens, indices
+
+
+class ColoAttention:
+    _kernel_dispatch_map: Optional[Dict[torch.dtype, Dict[Optional[AttnMaskType], Callable]]] = None
+
+    @staticmethod
+    def _init_kernels_dispatch():
+        if ColoAttention._kernel_dispatch_map is None:
+            # fp16/bf16
+            half_dispatch_map = {
+                None: FlashAttentionLoader(),
+                AttnMaskType.CUSTOM: FlashAttentionWithCustomMaskLoader(),
+                AttnMaskType.PADDED: FlashAttentionWithPaddingMaskLoader(),
+                AttnMaskType.CAUSAL: FlashAttentionLoader(),
+                AttnMaskType.PADDED_CAUSAL: FlashAttentionWithPaddingMaskLoader(),
+            }
+            # fp32
+            float_dispatch_map = {
+                None: FlashAttentionForFloatAndCustomMaskLoader(),
+                AttnMaskType.CUSTOM: FlashAttentionForFloatAndCustomMaskLoader(),
+                AttnMaskType.CAUSAL: FlashAttentionForFloatAndCustomMaskLoader(),
+            }
+            ColoAttention._kernel_dispatch_map = {
+                torch.float16: half_dispatch_map,
+                torch.bfloat16: half_dispatch_map,
+                torch.float32: float_dispatch_map,
+            }
+
+    @staticmethod
+    def _dispatch_kernel(dtype: torch.dtype, mask_type: Optional[AttnMaskType]) -> Callable:
+        ColoAttention._init_kernels_dispatch()
+        if (
+            dtype not in ColoAttention._kernel_dispatch_map
+            or mask_type not in ColoAttention._kernel_dispatch_map[dtype]
+        ):
+            raise ValueError(
+                "FlashAttention kernel is not available for dtype {} and mask_type {}".format(dtype, mask_type)
+            )
+        # lazy load
+        if isinstance(ColoAttention._kernel_dispatch_map[dtype][mask_type], KernelLoader):
+            ColoAttention._kernel_dispatch_map[dtype][mask_type] = ColoAttention._kernel_dispatch_map[dtype][
+                mask_type
+            ].load()
+        return ColoAttention._kernel_dispatch_map[dtype][mask_type]
+
+    @staticmethod
+    def prepare_attn_kwargs(
+        shape_4d: Tuple[int],
+        dtype: torch.dtype,
+        device: torch.device,
+        q_padding_mask: Optional[torch.Tensor] = None,
+        kv_padding_mask: Optional[torch.Tensor] = None,
+        is_causal: bool = False,
+    ) -> Dict[str, torch.Tensor]:
+        """Return a dictionary of keyword arguments for attention function. It supports 4 mask type.
+        1. custom mask: no padding mask and is_causal=False, return {}, users should handle attention mask by themselves.
+        2. padded mask: recv padding mask and is_causal=False, return {attention_mask, attention_mask_type, cu_seqlens_q, cu_seqlens_kv, max_seqlen_q, max_seqlen_kv, q_indices, kv_indices}.
+        3. causal mask: no padding mask and is_causal=True, return {attention_mask, attention_mask_type}.
+        4. padded causal mask: recv padding mask and is_causal=True, return {attention_mask, attention_mask_type, cu_seqlens_q, cu_seqlens_kv, max_seqlen_q, max_seqlen_kv, q_indices, kv_indices}.
+
+        Args:
+            shape_4d (Tuple[int]): Should be (B, 1, Sq, Skv)
+            dtype (torch.dtype): Dtype of attention mask, generally should be ``hidden_states.dtype``
+            device (torch.device): Device of attention mask, generally should be ``hidden_states.device``
+            q_padding_mask (Optional[torch.Tensor], optional): Padding mask of query. It should be a long tensor or int tensor.
+                The shape should be [B, Sq]. ``1`` means valid token, and ``0`` means padding token. Defaults to None.
+            kv_padding_mask (Optional[torch.Tensor], optional): Padding mask of key and value. It should be a long tensor or int tensor.
+                The shape should be [B, Skv]. ``1`` means valid token, and ``0`` means padding token.
+                If it's None and ``q_padding_mask`` is not None, it will be set to ``q_padding_mask``. Defaults to None.
+            is_causal (bool, optional): Whether to use causal attention mask. Defaults to False.
+
+        Returns:
+            Dict[str, torch.Tensor]: Dictionary of keyword arguments for attention function.
+        """
+        if q_padding_mask is None and not is_causal:
+            return {}
+        assert len(shape_4d) == 4 and shape_4d[1] == 1
+        b, _, s_q, s_kv = shape_4d
+        outputs = {}
+        if (q_padding_mask is None or q_padding_mask.bool().all()) and (
+            kv_padding_mask is None or kv_padding_mask.bool().all()
+        ):
+            # no padding
+            assert is_causal
+            outputs["attention_mask_type"] = AttnMaskType.CAUSAL
+            attention_mask = torch.ones(s_q, s_kv, dtype=dtype, device=device).tril(diagonal=0).expand(b, s_q, s_kv)
+        else:
+            if kv_padding_mask is None:
+                # self attention
+                kv_padding_mask = q_padding_mask
+            assert q_padding_mask.shape == (b, s_q) and kv_padding_mask.shape == (
+                b,
+                s_kv,
+            ), f"q_padding_mask shape {q_padding_mask.shape} and kv_padding_mask shape {kv_padding_mask.shape} should be the same. ({shape_4d})"
+            attention_mask = torch.einsum("bi,bj->bij", q_padding_mask, kv_padding_mask).to(dtype=dtype, device=device)
+            max_seqlen_q, cu_seqlens_q, q_indices = get_pad_info(q_padding_mask)
+            max_seqlen_kv, cu_seqlens_kv, kv_indices = get_pad_info(kv_padding_mask)
+            outputs.update(
+                {
+                    "cu_seqlens_q": cu_seqlens_q,
+                    "cu_seqlens_kv": cu_seqlens_kv,
+                    "max_seqlen_q": max_seqlen_q,
+                    "max_seqlen_kv": max_seqlen_kv,
+                    "q_indices": q_indices,
+                    "kv_indices": kv_indices,
+                }
+            )
+            if is_causal:
+                outputs["attention_mask_type"] = AttnMaskType.PADDED_CAUSAL
+                attention_mask = attention_mask * attention_mask.new_ones(s_q, s_kv).tril(diagonal=0)
+            else:
+                outputs["attention_mask_type"] = AttnMaskType.PADDED
+        attention_mask = invert_mask(attention_mask).unsqueeze(1)
+        outputs["attention_mask"] = attention_mask
+        return outputs
+
+    @staticmethod
+    def attention(
+        q: torch.Tensor,
+        k: torch.Tensor,
+        v: torch.Tensor,
+        attention_mask: Optional[torch.Tensor] = None,
+        attention_mask_type: AttnMaskType = AttnMaskType.CUSTOM,
+        cu_seqlens_q: Optional[torch.Tensor] = None,
+        cu_seqlens_kv: Optional[torch.Tensor] = None,
+        max_seqlen_q: Optional[int] = None,
+        max_seqlen_kv: Optional[int] = None,
+        q_indices: Optional[torch.Tensor] = None,
+        kv_indices: Optional[torch.Tensor] = None,
+        dropout_p: float = 0.0,
+        scale: Optional[float] = None,
+    ) -> torch.Tensor:
+        """Flash Attention function. It supports 4 mask type.
+        1. custom mask: recv attention_mask
+        2. padded mask: recv attention_mask, attention_mask_type, cu_seqlens_q, cu_seqlens_kv, max_seqlen_q, max_seqlen_kv, indices
+        3. causal mask: recv attention_mask, attention_mask_type
+        4. padded causal mask: recv attention_mask, attention_mask_type, cu_seqlens_q, cu_seqlens_kv, max_seqlen_q, max_seqlen_kv, indices
+
+        Args:
+            q (torch.Tensor): Query tensor. Shape should be [B, N, Sq, D]
+            k (torch.Tensor): Key tensor. Shape should be [B, N, Skv, D]
+            v (torch.Tensor): Value tensor. Shape should be [B, N, Skv, D]
+            attention_mask (Optional[torch.Tensor], optional): Attention mask tensor. Shape should be [B, 1, Sq, Skv]. Defaults to None.
+            attention_mask_type (AttnMaskType, optional): Attention mask type. Defaults to AttnMaskType.CUSTOM.
+            cu_seqlens_q (Optional[torch.Tensor], optional): The cumulative sequence lengths
+                of the sequences in the batch, used to index into q.
+                Shape should be [B+1]. Defaults to None.
+            cu_seqlens_kv (Optional[torch.Tensor], optional): The cumulative sequence lengths
+                of the sequences in the batch, used to index into kv.
+                Shape should be [B+1]. Defaults to None.
+            max_seqlen_q (Optional[int], optional): Maximum query sequence length in the batch. Defaults to None.
+            max_seqlen_kv (Optional[int], optional): Maximum key/value sequence length in the batch. Defaults to None.
+            indices (Optional[torch.Tensor], optional): The indices of non-masked tokens from the flattened input sequence.
+                Shape should be [NUM_TOKENS]. Defaults to None.
+            dropout_p (float, optional): Dropout probability. Defaults to 0.0.
+            scale (Optional[float], optional): Scaling factor applied prior to softmax. Defaults to None.
+
+        Returns:
+            torch.Tensor: Output tensor. Shape should be [B, N, Sq, D]
+        """
+        # known issue: sdpa does not support attention mask which contains whole row of masked tokens, which leads to nan
+        # this case is usaul when padding mask is used and self attention is performed
+        # thus, we don't use sdpa when padding mask is used
+        # sanity check
+        if attention_mask is not None:
+            assert torch.is_floating_point(attention_mask), "attention_mask should be a floating point tensor."
+            if attention_mask_type in (AttnMaskType.CUSTOM, AttnMaskType.CAUSAL):
+                assert (
+                    cu_seqlens_q is None
+                    and cu_seqlens_kv is None
+                    and max_seqlen_q is None
+                    and max_seqlen_kv is None
+                    and q_indices is None
+                    and kv_indices is None
+                )
+                if attention_mask_type == AttnMaskType.CUSTOM:
+                    assert not torch.all(attention_mask != 0, dim=-1).any()
+            elif attention_mask_type in (
+                AttnMaskType.PADDED,
+                AttnMaskType.PADDED_CAUSAL,
+            ):
+                assert (
+                    cu_seqlens_q is not None
+                    and cu_seqlens_kv is not None
+                    and max_seqlen_q is not None
+                    and max_seqlen_kv is not None
+                    and q_indices is not None
+                    and kv_indices is not None
+                )
+        else:
+            # if attention_mask is None, attention_mask_type should be the default value
+            assert attention_mask_type == AttnMaskType.CUSTOM
+        # kernel dispatch
+        mask_type = attention_mask_type if attention_mask is not None else None
+        attn_func = ColoAttention._dispatch_kernel(q.dtype, mask_type)
+        is_causal = attention_mask is not None and attention_mask_type in (
+            AttnMaskType.CAUSAL,
+            AttnMaskType.PADDED_CAUSAL,
+        )
+        return attn_func(
+            q,
+            k,
+            v,
+            dropout_p=dropout_p,
+            scale=scale,
+            attention_mask=attention_mask,
+            is_causal=is_causal,
+            cu_seqlens_q=cu_seqlens_q,
+            cu_seqlens_kv=cu_seqlens_kv,
+            max_seqlen_q=max_seqlen_q,
+            max_seqlen_kv=max_seqlen_kv,
+            q_indices=q_indices,
+            kv_indices=kv_indices,
+        )

colossalai/shardformer/modeling/blip2.py

@@ -3,6 +3,8 @@ from typing import Optional, Tuple
 import torch
 import torch.nn as nn
 
+from colossalai.shardformer.layer import ColoAttention
+
 
 def forward_fn():
     def forward(
@@ -62,8 +64,6 @@ def forward_fn():
 def get_blip2_flash_attention_forward():
     from transformers.models.blip_2.modeling_blip_2 import Blip2Attention
 
-    from colossalai.nn.layer.colo_attention import ColoAttention
-
     def forward(
         self: Blip2Attention,
         hidden_states: torch.Tensor,
@@ -71,16 +71,25 @@ def get_blip2_flash_attention_forward():
         output_attentions: Optional[bool] = False,
     ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
         """Input shape: Batch x Time x Channel"""
-
+        assert head_mask is None, "head_mask is not supported in FlashAttention"
         bsz, tgt_len, embed_dim = hidden_states.size()
         mixed_qkv = self.qkv(hidden_states)
-        mixed_qkv = mixed_qkv.reshape(bsz, tgt_len, 3, self.num_heads, -1).permute(2, 0, 1, 3, 4)
-        query_states, key_states, value_states = mixed_qkv[0], mixed_qkv[1], mixed_qkv[2]
+        mixed_qkv = mixed_qkv.reshape(bsz, tgt_len, 3, self.num_heads, self.head_dim).permute(2, 0, 3, 1, 4)
+        query_states, key_states, value_states = (
+            mixed_qkv[0],
+            mixed_qkv[1],
+            mixed_qkv[2],
+        )
 
-        attention = ColoAttention(
-            embed_dim=self.embed_dim, num_heads=self.num_heads, dropout=self.dropout.p, scale=self.scale
+        dropout_p = self.dropout.p if self.training else 0.0
+        context_layer = ColoAttention.attention(
+            query_states,
+            key_states,
+            value_states,
+            dropout_p=dropout_p,
+            scale=self.scale,
         )
-        context_layer = attention(query_states, key_states, value_states)
+        context_layer = context_layer.permute(0, 2, 1, 3).reshape(bsz, tgt_len, self.embed_dim)
 
         output = self.projection(context_layer)
         outputs = (output, None)
@@ -93,7 +102,11 @@ def get_blip2_flash_attention_forward():
 def get_jit_fused_blip2_QFormer_self_output_forward():
     from transformers.models.blip_2.modeling_blip_2 import Blip2QFormerSelfOutput
 
-    def forward(self: Blip2QFormerSelfOutput, hidden_states: torch.Tensor, input_tensor: torch.Tensor) -> torch.Tensor:
+    def forward(
+        self: Blip2QFormerSelfOutput,
+        hidden_states: torch.Tensor,
+        input_tensor: torch.Tensor,
+    ) -> torch.Tensor:
         hidden_states = self.dense(hidden_states)
         hidden_states = self.dropout_add(hidden_states, input_tensor, self.dropout.p, self.dropout.training)
         hidden_states = self.LayerNorm(hidden_states)
@@ -105,7 +118,11 @@ def get_jit_fused_blip2_QFormer_self_output_forward():
 def get_jit_fused_blip2_QFormer_output_forward():
     from transformers.models.blip_2.modeling_blip_2 import Blip2QFormerOutput
 
-    def forward(self: Blip2QFormerOutput, hidden_states: torch.Tensor, input_tensor: torch.Tensor) -> torch.Tensor:
+    def forward(
+        self: Blip2QFormerOutput,
+        hidden_states: torch.Tensor,
+        input_tensor: torch.Tensor,
+    ) -> torch.Tensor:
         hidden_states = self.dense(hidden_states)
         hidden_states = self.dropout_add(hidden_states, input_tensor, self.dropout.p, self.dropout.training)
         hidden_states = self.LayerNorm(hidden_states)

colossalai/shardformer/modeling/chatglm2.py

 """ PyTorch ChatGLM model. """
+
 from typing import List, Optional, Tuple
 
 import torch
@@ -9,63 +10,49 @@ from transformers.utils import logging
 
 from colossalai.pipeline.stage_manager import PipelineStageManager
 from colossalai.shardformer import ShardConfig
+from colossalai.shardformer.layer import AttnMaskType, ColoAttention
 from colossalai.shardformer.layer._operation import gather_forward_split_backward, split_forward_gather_backward
 from colossalai.shardformer.modeling.chatglm2_6b.modeling_chatglm import ChatGLMForConditionalGeneration, ChatGLMModel
 
 
 def get_flash_core_attention_forward():
-    from colossalai.nn.layer.colo_attention import AttnMaskType, ColoAttention
-
     from .chatglm2_6b.modeling_chatglm import CoreAttention
 
     def forward(self: CoreAttention, query_layer, key_layer, value_layer, attention_mask):
-        pytorch_major_version = int(torch.__version__.split(".")[0])
-        if pytorch_major_version >= 2:
-            query_layer, key_layer, value_layer = [k.permute(1, 2, 0, 3) for k in [query_layer, key_layer, value_layer]]
-            if attention_mask is None and query_layer.shape[2] == key_layer.shape[2]:
-                context_layer = torch.nn.functional.scaled_dot_product_attention(
-                    query_layer, key_layer, value_layer, is_causal=True
-                )
-            else:
-                if attention_mask is not None:
-                    attention_mask = ~attention_mask
-                context_layer = torch.nn.functional.scaled_dot_product_attention(
-                    query_layer, key_layer, value_layer, attention_mask
-                )
-            context_layer = context_layer.permute(2, 0, 1, 3)
-            new_context_layer_shape = context_layer.size()[:-2] + (self.hidden_size_per_partition,)
-            context_layer = context_layer.reshape(*new_context_layer_shape)
-        else:
-            # Raw attention scores
-            query_layer = query_layer.permute(1, 0, 2, 3).contiguous()
-            key_layer = key_layer.permute(1, 0, 2, 3).contiguous()
-            value_layer = value_layer.permute(1, 0, 2, 3).contiguous()
-
-            scale = 1.0 / self.norm_factor
-            if self.coeff is not None:
-                scale = scale * self.coeff
-
-            flash_attention_mask = None
-            attn_mask_type = None
-            if attention_mask is None:
-                attn_mask_type = AttnMaskType.causal
-            else:
-                flash_attention_mask = ~(attention_mask[:, :, -1].squeeze(1).to(torch.bool)).contiguous()
-                if not torch.all(flash_attention_mask):
-                    attn_mask_type = AttnMaskType.paddedcausal
-
-            attention = ColoAttention(
-                embed_dim=self.hidden_size_per_partition,
-                num_heads=self.num_attention_heads_per_partition,
-                dropout=self.attention_dropout.p,
-                scale=scale,
+        query_layer, key_layer, value_layer = [k.permute(1, 2, 0, 3) for k in [query_layer, key_layer, value_layer]]
+        if attention_mask is None and query_layer.shape[2] == key_layer.shape[2]:
+            attention_mask_type = AttnMaskType.CAUSAL
+            attn_bias = torch.zeros(
+                query_layer.shape[0],
+                1,
+                query_layer.shape[2],
+                key_layer.shape[2],
+                dtype=query_layer.dtype,
+                device=query_layer.device,
             )
-            context_layer = attention(
-                query_layer, key_layer, value_layer, attn_mask=flash_attention_mask, attn_mask_type=attn_mask_type
+            temp_mask = (
+                torch.ones(query_layer.shape[2], key_layer.shape[2], dtype=torch.bool, device=query_layer.device)
+                .tril(diagonal=0)
+                .expand(query_layer.shape[0], 1, -1, -1)
             )
-
-            context_layer = context_layer.permute(1, 0, -1).contiguous()
-
+            attn_bias.masked_fill_(temp_mask.logical_not(), torch.finfo(query_layer.dtype).min)
+        else:
+            attention_mask_type = AttnMaskType.CUSTOM
+            if attention_mask is not None:
+                attn_bias = torch.zeros_like(attention_mask, dtype=query_layer.dtype)
+                attn_bias.masked_fill_(attention_mask, torch.finfo(query_layer.dtype).min)
+        dropout_p = self.attention_dropout.p if self.training else 0.0
+        context_layer = ColoAttention.attention(
+            query_layer,
+            key_layer,
+            value_layer,
+            attention_mask=attn_bias,
+            attention_mask_type=attention_mask_type,
+            dropout_p=dropout_p,
+        )
+        context_layer = context_layer.permute(2, 0, 1, 3)
+        new_context_layer_shape = context_layer.size()[:-2] + (self.hidden_size_per_partition,)
+        context_layer = context_layer.reshape(*new_context_layer_shape)
         return context_layer
 
     return forward
@@ -169,11 +156,17 @@ class ChatGLMPipelineForwards:
         if self.pre_seq_len is not None:
             if past_key_values is None:
                 past_key_values = self.get_prompt(
-                    batch_size=batch_size, device=input_ids.device, dtype=inputs_embeds.dtype
+                    batch_size=batch_size,
+                    device=input_ids.device,
+                    dtype=inputs_embeds.dtype,
                 )
             if attention_mask is not None:
                 attention_mask = torch.cat(
-                    [attention_mask.new_ones((batch_size, self.pre_seq_len)), attention_mask], dim=-1
+                    [
+                        attention_mask.new_ones((batch_size, self.pre_seq_len)),
+                        attention_mask,
+                    ],
+                    dim=-1,
                 )
         if full_attention_mask is None:
             if (attention_mask is not None and not attention_mask.all()) or (past_key_values and seq_length != 1):
@@ -200,7 +193,9 @@ class ChatGLMPipelineForwards:
 
         if shard_config.enable_sequence_parallelism:
             hidden_states = split_forward_gather_backward(
-                hidden_states, dim=0, process_group=shard_config.tensor_parallel_process_group
+                hidden_states,
+                dim=0,
+                process_group=shard_config.tensor_parallel_process_group,
             )
         for idx in range(start_idx, end_idx):
             layer = self.encoder._get_layer(idx)
@@ -208,7 +203,12 @@ class ChatGLMPipelineForwards:
                 all_hidden_states = all_hidden_states + (hidden_states,)
             if self.encoder.gradient_checkpointing and self.encoder.training:
                 layer_ret = torch.utils.checkpoint.checkpoint(
-                    layer, hidden_states, attention_mask, rotary_pos_emb, past_key_values[idx], use_cache
+                    layer,
+                    hidden_states,
+                    attention_mask,
+                    rotary_pos_emb,
+                    past_key_values[idx],
+                    use_cache,
                 )
             else:
                 layer_ret = layer(
@@ -224,7 +224,9 @@ class ChatGLMPipelineForwards:
 
         if shard_config.enable_sequence_parallelism:
             hidden_states = gather_forward_split_backward(
-                hidden_states, dim=0, process_group=shard_config.tensor_parallel_process_group
+                hidden_states,
+                dim=0,
+                process_group=shard_config.tensor_parallel_process_group,
             )
         if output_hidden_states:
             all_hidden_states = all_hidden_states + (hidden_states,)
@@ -234,7 +236,14 @@ class ChatGLMPipelineForwards:
                 hidden_states = self.encoder.final_layernorm(hidden_states)
             if not return_dict:
                 return tuple(
-                    v for v in [hidden_states, presents, all_hidden_states, all_self_attentions] if v is not None
+                    v
+                    for v in [
+                        hidden_states,
+                        presents,
+                        all_hidden_states,
+                        all_self_attentions,
+                    ]
+                    if v is not None
                 )
             return BaseModelOutputWithPast(
                 last_hidden_state=hidden_states,
@@ -368,7 +377,9 @@ def get_chatglm_sequence_parallel_forward_fn(shard_config: ShardConfig):
         # Run encoder.
         # [seq_len, batch_size, hidden_size] -> [seq_len/TP_size, batch_size, hidden_size]
         inputs_embeds = split_forward_gather_backward(
-            inputs_embeds, dim=0, process_group=shard_config.tensor_parallel_process_group
+            inputs_embeds,
+            dim=0,
+            process_group=shard_config.tensor_parallel_process_group,
         )
         hidden_states, presents, all_hidden_states, all_self_attentions = self.encoder(
             inputs_embeds,
@@ -380,7 +391,9 @@ def get_chatglm_sequence_parallel_forward_fn(shard_config: ShardConfig):
         )
 
         hidden_states = gather_forward_split_backward(
-            hidden_states, dim=0, process_group=shard_config.tensor_parallel_process_group
+            hidden_states,
+            dim=0,
+            process_group=shard_config.tensor_parallel_process_group,
         )
 
         if not return_dict:

colossalai/shardformer/modeling/llama.py

@@ -15,7 +15,9 @@ from transformers.utils import logging
 from colossalai.pipeline.stage_manager import PipelineStageManager
 from colossalai.shardformer.shard import ShardConfig
 
-from ..layer import cross_entropy_1d
+
+from ..layer import ColoAttention, cross_entropy_1d
+
 
 try:
     from transformers.models.llama.modeling_llama import _prepare_4d_causal_attention_mask
@@ -105,18 +107,25 @@ class LlamaPipelineForwards:
 
         # embed positions, for the first stage, hidden_states is the input embeddings,
         # for the other stages, hidden_states is the output of the previous stage
-        if attention_mask is None:
-            attention_mask = torch.ones(
-                (batch_size, seq_length_with_past), dtype=torch.bool, device=hidden_states.device
-            )
-        if LATEST_VERSION:
-            attention_mask = _prepare_4d_causal_attention_mask(
-                attention_mask, (batch_size, seq_length), hidden_states, past_key_values_length
+        if shard_config.enable_flash_attention:
+            # in this case, attention_mask is a dict rather than a tensor
+            mask_shape = (batch_size, 1, seq_length_with_past, seq_length_with_past)
+            attention_mask = ColoAttention.prepare_attn_kwargs(
+                mask_shape, hidden_states.dtype, hidden_states.device, q_padding_mask=attention_mask, is_causal=True
             )
         else:
-            attention_mask = self._prepare_decoder_attention_mask(
-                attention_mask, (batch_size, seq_length), hidden_states, past_key_values_length
-            )
+            if attention_mask is None:
+                attention_mask = torch.ones(
+                    (batch_size, seq_length_with_past), dtype=torch.bool, device=hidden_states.device
+                )
+            if LATEST_VERSION:
+                attention_mask = _prepare_4d_causal_attention_mask(
+                    attention_mask, (batch_size, seq_length), hidden_states, past_key_values_length
+                )
+            else:
+                attention_mask = self._prepare_decoder_attention_mask(
+                    attention_mask, (batch_size, seq_length), hidden_states, past_key_values_length
+                )
 
         if self.gradient_checkpointing and self.training:
             if use_cache:
@@ -262,6 +271,7 @@ class LlamaPipelineForwards:
             stage_manager=stage_manager,
             hidden_states=hidden_states,
             stage_index=stage_index,
+            shard_config=shard_config,
         )
         past_key_values = None
 
@@ -352,6 +362,7 @@ class LlamaPipelineForwards:
             stage_manager=stage_manager,
             hidden_states=hidden_states,
             stage_index=stage_index,
+            shard_config=shard_config,
         )
 
         if input_ids is not None:
@@ -420,8 +431,6 @@ class LlamaPipelineForwards:
 def get_llama_flash_attention_forward(shard_config: ShardConfig):
     from transformers.models.llama.modeling_llama import LlamaAttention, apply_rotary_pos_emb
 
-    from colossalai.nn.layer.colo_attention import AttnMaskType, ColoAttention
-
     llama_version = 2
     try:
         from transformers.models.llama.modeling_llama import repeat_kv
@@ -432,7 +441,7 @@ def get_llama_flash_attention_forward(shard_config: ShardConfig):
     def forward(
         self: LlamaAttention,
         hidden_states: torch.Tensor,
-        attention_mask: Optional[torch.Tensor] = None,
+        attention_mask: Optional[dict] = None,
         position_ids: Optional[torch.LongTensor] = None,
         past_key_value: Optional[Tuple[torch.Tensor]] = None,
         output_attentions: bool = False,
@@ -466,31 +475,10 @@ def get_llama_flash_attention_forward(shard_config: ShardConfig):
             key_states = repeat_kv(key_states, self.num_key_value_groups)
             value_states = repeat_kv(value_states, self.num_key_value_groups)
 
-        me_input_shape = (bsz, q_len, self.num_heads, self.head_dim)
-        query_states = query_states.transpose(1, 2).contiguous().view(*me_input_shape)
-        key_states = key_states.transpose(1, 2).contiguous().view(*me_input_shape)
-        value_states = value_states.transpose(1, 2).contiguous().view(*me_input_shape)
-
-        flash_attention_mask = None
-        attn_mask_type = AttnMaskType.causal
-        if not getattr(shard_config, "causal_lm", False) and attention_mask != 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()}"
-                )
-            flash_attention_mask = ~(attention_mask[:, :, -1].squeeze(1).to(torch.bool)).contiguous()
-            attn_mask_type = AttnMaskType.paddedcausal
-
-        if not hasattr(self, "attention"):
-            self.attention = ColoAttention(embed_dim=self.hidden_size, num_heads=self.num_heads)
-        attn_output = self.attention(
-            query_states,
-            key_states,
-            value_states,
-            attn_mask=flash_attention_mask,
-            attn_mask_type=attn_mask_type,
-            origin_attn_mask=attention_mask,
-        )
+        assert isinstance(attention_mask, dict), "Flash Attention Error: attention_mask should be a dict."
+        attn_output = ColoAttention.attention(query_states, key_states, value_states, **attention_mask)
+        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)
 
@@ -499,6 +487,137 @@ def get_llama_flash_attention_forward(shard_config: ShardConfig):
     return forward
 
 
+def get_llama_model_forward_for_flash_attn(shard_config: ShardConfig):
+    logger = logging.get_logger(__name__)
+    assert shard_config.enable_flash_attention, "Flash Attention is not enabled."
+
+    def forward(
+        self: LlamaModel,
+        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,
+        return_dict: Optional[bool] = None,
+    ) -> Union[Tuple, BaseModelOutputWithPast]:
+        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 decoder_input_ids and decoder_inputs_embeds at the same time")
+        elif input_ids is not None:
+            batch_size, seq_length = input_ids.shape
+        elif inputs_embeds is not None:
+            batch_size, seq_length, _ = inputs_embeds.shape
+        else:
+            raise ValueError("You have to specify either decoder_input_ids or decoder_inputs_embeds")
+
+        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).view(-1, seq_length)
+        else:
+            position_ids = position_ids.view(-1, seq_length).long()
+
+        if inputs_embeds is None:
+            inputs_embeds = self.embed_tokens(input_ids)
+        # embed positions
+        hidden_states = inputs_embeds
+
+        # in this case, attention_mask is a dict rather than a tensor
+        mask_shape = (batch_size, 1, seq_length_with_past, seq_length_with_past)
+        attention_mask = ColoAttention.prepare_attn_kwargs(
+            mask_shape, hidden_states.dtype, hidden_states.device, q_padding_mask=attention_mask, is_causal=True
+        )
+
+        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
+        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],)
+
+        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] if v is not None)
+        return BaseModelOutputWithPast(
+            last_hidden_state=hidden_states,
+            past_key_values=next_cache,
+            hidden_states=all_hidden_states,
+            attentions=all_self_attns,
+        )
+
+    return forward
+
+
 def get_lm_forward_with_dist_cross_entropy(shard_config: ShardConfig):
     from transformers import LlamaForCausalLM
 

colossalai/shardformer/modeling/vit.py

-import math
 from typing import List, Optional, Tuple, Union
 
 import torch
@@ -6,6 +5,7 @@ from transformers.models.vit.modeling_vit import BaseModelOutput, ViTEncoder
 from transformers.utils import logging
 
 from colossalai.pipeline.stage_manager import PipelineStageManager
+from colossalai.shardformer.layer import ColoAttention
 
 
 def _encoder_forward(
@@ -98,7 +98,9 @@ def ViTModel_pipeline_forward(stage_manager: PipelineStageManager, stage_index:
                 pixel_values = pixel_values.to(expected_dtype)
 
             embedding_output = self.embeddings(
-                pixel_values, bool_masked_pos=bool_masked_pos, interpolate_pos_encoding=interpolate_pos_encoding
+                pixel_values,
+                bool_masked_pos=bool_masked_pos,
+                interpolate_pos_encoding=interpolate_pos_encoding,
             )
             hidden_states = embedding_output
         else:
@@ -336,34 +338,27 @@ def ViTForMaskedImageModeling_pipeline_forward(stage_manager: PipelineStageManag
 def get_vit_flash_self_attention_forward():
     from transformers.models.vit.modeling_vit import ViTSelfAttention
 
-    from colossalai.nn.layer.colo_attention import ColoAttention
-
-    def transpose_for_scores(x: torch.Tensor, num_attention_heads, attention_head_size) -> torch.Tensor:
-        new_x_shape = x.size()[:-1] + (num_attention_heads, attention_head_size)
-        x = x.view(new_x_shape)
-        return x
-
     def forward(
         self: ViTSelfAttention,
         hidden_states: torch.Tensor,
         head_mask: Optional[torch.Tensor] = None,
         output_attentions: bool = False,
     ) -> Union[Tuple[torch.Tensor, torch.Tensor], Tuple[torch.Tensor]]:
+        assert head_mask is None, "head_mask is not supported for FlashAttention"
         mixed_query_layer = self.query(hidden_states)
 
-        key_layer = transpose_for_scores(self.key(hidden_states), self.num_attention_heads, self.attention_head_size)
-        value_layer = transpose_for_scores(
-            self.value(hidden_states), self.num_attention_heads, self.attention_head_size
-        )
-        query_layer = transpose_for_scores(mixed_query_layer, self.num_attention_heads, self.attention_head_size)
+        key_layer = self.transpose_for_scores(self.key(hidden_states))
+        value_layer = self.transpose_for_scores(self.value(hidden_states))
+        query_layer = self.transpose_for_scores(mixed_query_layer)
 
-        scale = 1.0 / math.sqrt(self.attention_head_size)
-        attention = ColoAttention(
-            embed_dim=self.all_head_size, num_heads=self.num_attention_heads, dropout=self.dropout.p, scale=scale
-        )
-        context_layer = attention(query_layer, key_layer, value_layer)
+        dropout_p = self.dropout.p if self.training else 0.0
+        context_layer = ColoAttention.attention(query_layer, key_layer, value_layer, dropout_p=dropout_p)
+
+        context_layer = context_layer.permute(0, 2, 1, 3).contiguous()
+        new_context_layer_shape = context_layer.size()[:-2] + (self.all_head_size,)
+        context_layer = context_layer.view(new_context_layer_shape)
 
-        outputs = (context_layer,)
+        outputs = (context_layer, None) if output_attentions else (context_layer,)
 
         return outputs
 

colossalai/shardformer/policies/gpt2.py

@@ -8,6 +8,7 @@ import colossalai.shardformer.layer as col_nn
 from ..modeling.gpt2 import (
     GPT2PipelineForwards,
     get_gpt2_flash_attention_forward,
+    get_gpt_model_forward_for_flash_attn,
     get_lm_forward_with_dist_cross_entropy,
     gpt2_sequence_parallel_forward_fn,
 )
@@ -75,7 +76,11 @@ class GPT2Policy(Policy):
                     SubModuleReplacementDescription(
                         suffix="attn.c_attn",
                         target_module=col_nn.GPT2FusedLinearConv1D_Col,
-                        kwargs={"n_fused": 3, "seq_parallel": use_sequence_parallel, "overlap": overlap},
+                        kwargs={
+                            "n_fused": 3,
+                            "seq_parallel": use_sequence_parallel,
+                            "overlap": overlap,
+                        },
                     ),
                     SubModuleReplacementDescription(
                         suffix="attn.c_proj",
@@ -87,7 +92,11 @@ class GPT2Policy(Policy):
                     SubModuleReplacementDescription(
                         suffix="mlp.c_fc",
                         target_module=col_nn.GPT2FusedLinearConv1D_Col,
-                        kwargs={"n_fused": 1, "seq_parallel": use_sequence_parallel, "overlap": overlap},
+                        kwargs={
+                            "n_fused": 1,
+                            "seq_parallel": use_sequence_parallel,
+                            "overlap": overlap,
+                        },
                     ),
                     SubModuleReplacementDescription(
                         suffix="mlp.c_proj",
@@ -150,6 +159,10 @@ class GPT2Policy(Policy):
                 policy=policy,
                 target_key=GPT2Attention,
             )
+            if not self.shard_config.pipeline_stage_manager:
+                policy[GPT2Model].method_replacement = {
+                    "forward": get_gpt_model_forward_for_flash_attn(self.shard_config)
+                }
 
         if self.shard_config.enable_sequence_parallelism:
             policy[GPT2Model].method_replacement = {"forward": gpt2_sequence_parallel_forward_fn(self.shard_config)}
@@ -223,14 +236,21 @@ class GPT2Policy(Policy):
                 num_stages=stage_manager.num_stages,
             )
             method_replacement = {
-                "forward": partial(new_forward, stage_manager=stage_manager, shard_config=self.shard_config)
+                "forward": partial(
+                    new_forward,
+                    stage_manager=stage_manager,
+                    shard_config=self.shard_config,
+                )
             }
         else:
             layers_per_stage = Policy.distribute_layers(len(module.h), stage_manager.num_stages)
             stage_index = Policy.get_stage_index(layers_per_stage, stage_manager.stage)
             method_replacement = {
                 "forward": partial(
-                    new_forward, stage_manager=stage_manager, stage_index=stage_index, shard_config=self.shard_config
+                    new_forward,
+                    stage_manager=stage_manager,
+                    stage_index=stage_index,
+                    shard_config=self.shard_config,
                 )
             }
         self.append_or_create_method_replacement(description=method_replacement, policy=policy, target_key=model_cls)
@@ -245,7 +265,9 @@ class GPT2ModelPolicy(GPT2Policy):
 
         if self.pipeline_stage_manager is not None:
             self.set_pipeline_forward(
-                model_cls=GPT2Model, new_forward=GPT2PipelineForwards.gpt2_model_forward, policy=policy
+                model_cls=GPT2Model,
+                new_forward=GPT2PipelineForwards.gpt2_model_forward,
+                policy=policy,
             )
         return policy
 
@@ -299,7 +321,12 @@ class GPT2LMHeadModelPolicy(GPT2Policy):
         if stage_manager is not None:
             if stage_manager.num_stages > 1 and id(module.transformer.wte.weight) == id(module.lm_head.weight):
                 first_stage, last_stage = 0, stage_manager.num_stages - 1
-                return [{first_stage: module.transformer.wte.weight, last_stage: module.lm_head.weight}]
+                return [
+                    {
+                        first_stage: module.transformer.wte.weight,
+                        last_stage: module.lm_head.weight,
+                    }
+                ]
         return []
 
 
@@ -315,7 +342,9 @@ class GPT2DoubleHeadsModelPolicy(GPT2Policy):
                 GPT2DoubleHeadsModel: ModulePolicyDescription(
                     sub_module_replacement=[
                         SubModuleReplacementDescription(
-                            suffix="lm_head", target_module=col_nn.Linear1D_Col, kwargs={"gather_output": True}
+                            suffix="lm_head",
+                            target_module=col_nn.Linear1D_Col,
+                            kwargs={"gather_output": True},
                         )
                     ]
                 )
@@ -350,7 +379,12 @@ class GPT2DoubleHeadsModelPolicy(GPT2Policy):
         if stage_manager is not None:
             if stage_manager.num_stages > 1 and id(module.transformer.wte.weight) == id(module.lm_head.weight):
                 first_stage, last_stage = 0, stage_manager.num_stages - 1
-                return [{first_stage: module.transformer.wte.weight, last_stage: module.lm_head.weight}]
+                return [
+                    {
+                        first_stage: module.transformer.wte.weight,
+                        last_stage: module.lm_head.weight,
+                    }
+                ]
         return []
 
 
@@ -392,7 +426,10 @@ class GPT2ForTokenClassificationPolicy(GPT2Policy):
             addon_module = {
                 GPT2ForTokenClassification: ModulePolicyDescription(
                     sub_module_replacement=[
-                        SubModuleReplacementDescription(suffix="dropout", target_module=col_nn.DropoutForParallelInput)
+                        SubModuleReplacementDescription(
+                            suffix="dropout",
+                            target_module=col_nn.DropoutForParallelInput,
+                        )
                     ]
                 )
             }

colossalai/shardformer/policies/gptj.py

@@ -6,7 +6,11 @@ from torch import Tensor, nn
 
 import colossalai.shardformer.layer as col_nn
 
-from ..modeling.gptj import GPTJPipelineForwards, get_gptj_flash_attention_forward
+from ..modeling.gptj import (
+    GPTJPipelineForwards,
+    get_gptj_flash_attention_forward,
+    gptj_model_forward_for_flash_attention,
+)
 from .base_policy import ModulePolicyDescription, Policy, SubModuleReplacementDescription
 
 __all__ = [
@@ -71,17 +75,26 @@ class GPTJPolicy(Policy):
                     SubModuleReplacementDescription(
                         suffix="attn.k_proj",
                         target_module=col_nn.Linear1D_Col,
-                        kwargs={"seq_parallel": use_sequence_parallel, "overlap": overlap},
+                        kwargs={
+                            "seq_parallel": use_sequence_parallel,
+                            "overlap": overlap,
+                        },
                     ),
                     SubModuleReplacementDescription(
                         suffix="attn.q_proj",
                         target_module=col_nn.Linear1D_Col,
-                        kwargs={"seq_parallel": use_sequence_parallel, "overlap": overlap},
+                        kwargs={
+                            "seq_parallel": use_sequence_parallel,
+                            "overlap": overlap,
+                        },
                     ),
                     SubModuleReplacementDescription(
                         suffix="attn.v_proj",
                         target_module=col_nn.Linear1D_Col,
-                        kwargs={"seq_parallel": use_sequence_parallel, "overlap": overlap},
+                        kwargs={
+                            "seq_parallel": use_sequence_parallel,
+                            "overlap": overlap,
+                        },
                     ),
                     SubModuleReplacementDescription(
                         suffix="attn.out_proj",
@@ -143,6 +156,12 @@ class GPTJPolicy(Policy):
                 policy=policy,
                 target_key=GPTJAttention,
             )
+            if not self.shard_config.pipeline_stage_manager:
+                self.append_or_create_method_replacement(
+                    description={"forward": gptj_model_forward_for_flash_attention(self.shard_config)},
+                    policy=policy,
+                    target_key=GPTJModel,
+                )
 
         return policy
 
@@ -185,7 +204,10 @@ class GPTJPolicy(Policy):
         stage_index = Policy.get_stage_index(layers_per_stage, stage_manager.stage)
         method_replacement = {
             "forward": partial(
-                new_forward, stage_manager=stage_manager, stage_index=stage_index, shard_config=self.shard_config
+                new_forward,
+                stage_manager=stage_manager,
+                stage_index=stage_index,
+                shard_config=self.shard_config,
             )
         }
         self.append_or_create_method_replacement(description=method_replacement, policy=policy, target_key=model_cls)
@@ -203,7 +225,9 @@ class GPTJModelPolicy(GPTJPolicy):
 
         if self.pipeline_stage_manager is not None:
             self.set_pipeline_forward(
-                model_cls=GPTJModel, new_forward=GPTJPipelineForwards.gptj_model_forward, policy=policy
+                model_cls=GPTJModel,
+                new_forward=GPTJPipelineForwards.gptj_model_forward,
+                policy=policy,
             )
         return policy
 
@@ -230,7 +254,9 @@ class GPTJForCausalLMPolicy(GPTJPolicy):
                 GPTJForCausalLM: ModulePolicyDescription(
                     sub_module_replacement=[
                         SubModuleReplacementDescription(
-                            suffix="lm_head", target_module=col_nn.Linear1D_Col, kwargs={"gather_output": True}
+                            suffix="lm_head",
+                            target_module=col_nn.Linear1D_Col,
+                            kwargs={"gather_output": True},
                         )
                     ]
                 )
@@ -239,7 +265,9 @@ class GPTJForCausalLMPolicy(GPTJPolicy):
 
         if self.pipeline_stage_manager is not None:
             self.set_pipeline_forward(
-                model_cls=GPTJForCausalLM, new_forward=GPTJPipelineForwards.gptj_causallm_model_forward, policy=policy
+                model_cls=GPTJForCausalLM,
+                new_forward=GPTJPipelineForwards.gptj_causallm_model_forward,
+                policy=policy,
             )
         return policy
 
@@ -256,7 +284,12 @@ class GPTJForCausalLMPolicy(GPTJPolicy):
         if stage_manager is not None:
             if stage_manager.num_stages > 1 and id(module.transformer.wte.weight) == id(module.lm_head.weight):
                 first_stage, last_stage = 0, stage_manager.num_stages - 1
-                return [{first_stage: module.transformer.wte.weight, last_stage: module.lm_head.weight}]
+                return [
+                    {
+                        first_stage: module.transformer.wte.weight,
+                        last_stage: module.lm_head.weight,
+                    }
+                ]
         return []
 
 

colossalai/shardformer/policies/llama.py

@@ -11,6 +11,7 @@ from colossalai.shardformer.layer import FusedRMSNorm, Linear1D_Col, Linear1D_Ro
 from ..modeling.llama import (
     LlamaPipelineForwards,
     get_llama_flash_attention_forward,
+    get_llama_model_forward_for_flash_attn,
     get_lm_forward_with_dist_cross_entropy,
 )
 from .base_policy import ModulePolicyDescription, Policy, SubModuleReplacementDescription
@@ -135,6 +136,15 @@ class LlamaPolicy(Policy):
                 policy=policy,
                 target_key=LlamaAttention,
             )
+            if self.pipeline_stage_manager is None:
+                # replace llama model forward method
+                self.append_or_create_method_replacement(
+                    description={
+                        "forward": get_llama_model_forward_for_flash_attn(self.shard_config),
+                    },
+                    policy=policy,
+                    target_key=LlamaModel,
+                )
 
         return policy