File size: 13,290 Bytes
357c94c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
import os
import torch
import datetime
import torch.distributed as dist
from typing import Any, Tuple
from torch import Tensor
from flash_attn.flash_attn_interface import flash_attn_varlen_func


class COMM_INFO:
    def __init__(self):
        self.group = None
        self.sp_size = 1
        self.global_rank = 0
        self.rank_within_group = 0
        self.group_id = 0


nccl_info = COMM_INFO()
_SEQUENCE_PARALLEL_STATE = False


def get_cu_seqlens(text_mask, img_len):
    """Calculate cu_seqlens_q, cu_seqlens_kv using text_mask and img_len

    Args:
        text_mask (torch.Tensor): the mask of text
        img_len (int): the length of image

    Returns:
        torch.Tensor: the calculated cu_seqlens for flash attention
    """
    batch_size = text_mask.shape[0]
    text_len = text_mask.sum(dim=1)
    max_len = text_mask.shape[1] + img_len

    cu_seqlens = torch.zeros([2 * batch_size + 1], dtype=torch.int32, device="cuda")

    for i in range(batch_size):
        s = text_len[i] + img_len
        s1 = i * max_len + s
        s2 = (i + 1) * max_len
        cu_seqlens[2 * i + 1] = s1
        cu_seqlens[2 * i + 2] = s2

    return cu_seqlens

def initialize_sequence_parallel_state(sequence_parallel_size):
    global _SEQUENCE_PARALLEL_STATE
    if sequence_parallel_size > 1:
        _SEQUENCE_PARALLEL_STATE = True
        initialize_sequence_parallel_group(sequence_parallel_size)
    else:
        nccl_info.sp_size = 1
        nccl_info.global_rank = int(os.getenv("RANK", "0"))
        nccl_info.rank_within_group = 0
        nccl_info.group_id = int(os.getenv("RANK", "0"))

def get_sequence_parallel_state():
    return _SEQUENCE_PARALLEL_STATE

def initialize_sequence_parallel_group(sequence_parallel_size):
    """Initialize the sequence parallel group."""
    rank = int(os.getenv("RANK", "0"))
    world_size = int(os.getenv("WORLD_SIZE", "1"))
    assert (
        world_size % sequence_parallel_size == 0
    ), "world_size must be divisible by sequence_parallel_size, but got world_size: {}, sequence_parallel_size: {}".format(
        world_size, sequence_parallel_size)
    nccl_info.sp_size = sequence_parallel_size
    nccl_info.global_rank = rank
    num_sequence_parallel_groups: int = world_size // sequence_parallel_size
    for i in range(num_sequence_parallel_groups):
        ranks = range(i * sequence_parallel_size, (i + 1) * sequence_parallel_size)
        group = dist.new_group(ranks)
        if rank in ranks:
            nccl_info.group = group
            nccl_info.rank_within_group = rank - i * sequence_parallel_size
            nccl_info.group_id = i

def initialize_distributed(seed):
    local_rank = int(os.getenv("RANK", 0))
    world_size = int(os.getenv("WORLD_SIZE", 1))
    torch.cuda.set_device(local_rank)
    dist.init_process_group(backend="nccl", init_method="env://", timeout=datetime.timedelta(seconds=2**31-1), world_size=world_size, rank=local_rank)
    torch.manual_seed(seed)
    torch.cuda.manual_seed_all(seed)
    initialize_sequence_parallel_state(world_size)

def _all_to_all_4D(input: torch.tensor, scatter_idx: int = 2, gather_idx: int = 1, group=None) -> torch.tensor:
    """
    all-to-all for QKV

    Args:
        input (torch.tensor): a tensor sharded along dim scatter dim
        scatter_idx (int): default 1
        gather_idx (int): default 2
        group : torch process group

    Returns:
        torch.tensor: resharded tensor (bs, seqlen/P, hc, hs)
    """
    assert (input.dim() == 4), f"input must be 4D tensor, got {input.dim()} and shape {input.shape}"

    seq_world_size = dist.get_world_size(group)
    if scatter_idx == 2 and gather_idx == 1:
        # input (torch.tensor): a tensor sharded along dim 1 (bs, seqlen/P, hc, hs) output: (bs, seqlen, hc/P, hs)
        bs, shard_seqlen, hc, hs = input.shape
        seqlen = shard_seqlen * seq_world_size
        shard_hc = hc // seq_world_size

        # transpose groups of heads with the seq-len parallel dimension, so that we can scatter them!
        # (bs, seqlen/P, hc, hs) -reshape-> (bs, seq_len/P, P, hc/P, hs) -transpose(0,2)-> (P, seq_len/P, bs, hc/P, hs)
        input_t = (input.reshape(bs, shard_seqlen, seq_world_size, shard_hc, hs).transpose(0, 2).contiguous())

        output = torch.empty_like(input_t)
        # https://pytorch.org/docs/stable/distributed.html#torch.distributed.all_to_all_single
        # (P, seq_len/P, bs, hc/P, hs) scatter seqlen -all2all-> (P, seq_len/P, bs, hc/P, hs) scatter head
        if seq_world_size > 1:
            dist.all_to_all_single(output, input_t, group=group)
            torch.cuda.synchronize()
        else:
            output = input_t
        # if scattering the seq-dim, transpose the heads back to the original dimension
        output = output.reshape(seqlen, bs, shard_hc, hs)

        # (seq_len, bs, hc/P, hs) -reshape-> (bs, seq_len, hc/P, hs)
        output = output.transpose(0, 1).contiguous().reshape(bs, seqlen, shard_hc, hs)

        return output

    elif scatter_idx == 1 and gather_idx == 2:
        # input (torch.tensor): a tensor sharded along dim 1 (bs, seqlen, hc/P, hs) output: (bs, seqlen/P, hc, hs)
        bs, seqlen, shard_hc, hs = input.shape
        hc = shard_hc * seq_world_size
        shard_seqlen = seqlen // seq_world_size
        seq_world_size = dist.get_world_size(group)

        # transpose groups of heads with the seq-len parallel dimension, so that we can scatter them!
        # (bs, seqlen, hc/P, hs) -reshape-> (bs, P, seq_len/P, hc/P, hs) -transpose(0, 3)-> (hc/P, P, seqlen/P, bs, hs) -transpose(0, 1) -> (P, hc/P, seqlen/P, bs, hs)
        input_t = (input.reshape(bs, seq_world_size, shard_seqlen, shard_hc,
                                 hs).transpose(0,
                                               3).transpose(0,
                                                            1).contiguous().reshape(seq_world_size, shard_hc,
                                                                                    shard_seqlen, bs, hs))

        output = torch.empty_like(input_t)
        # https://pytorch.org/docs/stable/distributed.html#torch.distributed.all_to_all_single
        # (P, bs x hc/P, seqlen/P, hs) scatter seqlen -all2all-> (P, bs x seq_len/P, hc/P, hs) scatter head
        if seq_world_size > 1:
            dist.all_to_all_single(output, input_t, group=group)
            torch.cuda.synchronize()
        else:
            output = input_t

        # if scattering the seq-dim, transpose the heads back to the original dimension
        output = output.reshape(hc, shard_seqlen, bs, hs)

        # (hc, seqlen/N, bs, hs) -tranpose(0,2)-> (bs, seqlen/N, hc, hs)
        output = output.transpose(0, 2).contiguous().reshape(bs, shard_seqlen, hc, hs)

        return output
    else:
        raise RuntimeError("scatter_idx must be 1 or 2 and gather_idx must be 1 or 2")


class SeqAllToAll4D(torch.autograd.Function):
    @staticmethod
    def forward(
        ctx: Any,
        group: dist.ProcessGroup,
        input: Tensor,
        scatter_idx: int,
        gather_idx: int,
    ) -> Tensor:
        ctx.group = group
        ctx.scatter_idx = scatter_idx
        ctx.gather_idx = gather_idx

        return _all_to_all_4D(input, scatter_idx, gather_idx, group=group)

    @staticmethod
    def backward(ctx: Any, *grad_output: Tensor) -> Tuple[None, Tensor, None, None]:
        return (
            None,
            SeqAllToAll4D.apply(ctx.group, *grad_output, ctx.gather_idx, ctx.scatter_idx),
            None,
            None,
        )


def all_to_all_4D(
    input_: torch.Tensor,
    scatter_dim: int = 2,
    gather_dim: int = 1,
):
    return SeqAllToAll4D.apply(nccl_info.group, input_, scatter_dim, gather_dim)


def _all_to_all(
    input_: torch.Tensor,
    world_size: int,
    group: dist.ProcessGroup,
    scatter_dim: int,
    gather_dim: int,
):
    input_list = [t.contiguous() for t in torch.tensor_split(input_, world_size, scatter_dim)]
    output_list = [torch.empty_like(input_list[0]) for _ in range(world_size)]
    dist.all_to_all(output_list, input_list, group=group)
    return torch.cat(output_list, dim=gather_dim).contiguous()


class _AllToAll(torch.autograd.Function):
    """All-to-all communication.

    Args:
        input_: input matrix
        process_group: communication group
        scatter_dim: scatter dimension
        gather_dim: gather dimension
    """

    @staticmethod
    def forward(ctx, input_, process_group, scatter_dim, gather_dim):
        ctx.process_group = process_group
        ctx.scatter_dim = scatter_dim
        ctx.gather_dim = gather_dim
        ctx.world_size = dist.get_world_size(process_group)
        output = _all_to_all(input_, ctx.world_size, process_group, scatter_dim, gather_dim)
        return output

    @staticmethod
    def backward(ctx, grad_output):
        grad_output = _all_to_all(
            grad_output,
            ctx.world_size,
            ctx.process_group,
            ctx.gather_dim,
            ctx.scatter_dim,
        )
        return (
            grad_output,
            None,
            None,
            None,
        )

def all_to_all(
    input_: torch.Tensor,
    scatter_dim: int = 2,
    gather_dim: int = 1,
):
    return _AllToAll.apply(input_, nccl_info.group, scatter_dim, gather_dim)


class _AllGather(torch.autograd.Function):
    """All-gather communication with autograd support.

    Args:
        input_: input tensor
        dim: dimension along which to concatenate
    """

    @staticmethod
    def forward(ctx, input_, dim):
        ctx.dim = dim
        world_size = nccl_info.sp_size
        group = nccl_info.group
        input_size = list(input_.size())

        ctx.input_size = input_size[dim]

        tensor_list = [torch.empty_like(input_) for _ in range(world_size)]
        input_ = input_.contiguous()
        dist.all_gather(tensor_list, input_, group=group)

        output = torch.cat(tensor_list, dim=dim)
        return output

    @staticmethod
    def backward(ctx, grad_output):
        world_size = nccl_info.sp_size
        rank = nccl_info.rank_within_group
        dim = ctx.dim
        input_size = ctx.input_size

        sizes = [input_size] * world_size

        grad_input_list = torch.split(grad_output, sizes, dim=dim)
        grad_input = grad_input_list[rank]

        return grad_input, None


def all_gather(input_: torch.Tensor, dim: int = 1):
    """Performs an all-gather operation on the input tensor along the specified dimension.

    Args:
        input_ (torch.Tensor): Input tensor of shape [B, H, S, D].
        dim (int, optional): Dimension along which to concatenate. Defaults to 1.

    Returns:
        torch.Tensor: Output tensor after all-gather operation, concatenated along 'dim'.
    """
    return _AllGather.apply(input_, dim)

def parallel_attention(q, k, v, img_q_len, img_kv_len, cu_seqlens_q, cu_seqlens_kv, max_seqlen_q, max_seqlen_kv,):
    """
    img_q_len,img_kv_len: 32256
    text_mask: 2x256
    query: [2, 32256, 24, 128])
    encoder_query: [2, 256, 24, 128]
    """
    query, encoder_query = q
    key, encoder_key = k
    value, encoder_value = v
    rank = torch.distributed.get_rank()
    if get_sequence_parallel_state():
        query = all_to_all_4D(query, scatter_dim=2, gather_dim=1)  # [2, 32256, 24, 128]
        key = all_to_all_4D(key, scatter_dim=2, gather_dim=1)
        value = all_to_all_4D(value, scatter_dim=2, gather_dim=1)
        def shrink_head(encoder_state, dim):
            local_heads = encoder_state.shape[dim] // nccl_info.sp_size
            return encoder_state.narrow(dim, nccl_info.rank_within_group * local_heads, local_heads)
        encoder_query = shrink_head(encoder_query, dim=2)
        encoder_key = shrink_head(encoder_key, dim=2)
        encoder_value = shrink_head(encoder_value, dim=2)

    sequence_length = query.size(1) # 32256
    encoder_sequence_length = encoder_query.size(1)  # 256

    query = torch.cat([query, encoder_query], dim=1)
    key = torch.cat([key, encoder_key], dim=1)
    value = torch.cat([value, encoder_value], dim=1)
    bsz = query.shape[0]
    head = query.shape[-2]
    head_dim = query.shape[-1]
    query, key, value = [
            x.view(x.shape[0] * x.shape[1], *x.shape[2:])
            for x in [query, key, value]
        ]
    hidden_states = flash_attn_varlen_func(
        query,
        key,
        value,
        cu_seqlens_q,
        cu_seqlens_kv,
        max_seqlen_q,
        max_seqlen_kv,
    )
    # B, S, 3, H, D
    hidden_states = hidden_states.view(bsz, max_seqlen_q, head, head_dim).contiguous()
    
    hidden_states, encoder_hidden_states = hidden_states.split_with_sizes((sequence_length, encoder_sequence_length),
                                                                            dim=1)
    if get_sequence_parallel_state():
        hidden_states = all_to_all_4D(hidden_states, scatter_dim=1, gather_dim=2)
        encoder_hidden_states = all_gather(encoder_hidden_states, dim=2).contiguous()
    hidden_states = hidden_states.to(query.dtype)
    encoder_hidden_states = encoder_hidden_states.to(query.dtype)

    attn = torch.cat([hidden_states, encoder_hidden_states], dim=1)

    b, s, _, _= attn.shape
    attn = attn.reshape(b, s, -1)
    return attn, None