AReaL/realhf/impl/model/utils/padding.py

322 lines
13 KiB
Python

# Modified from flash-attention under BSD-3 license.
# Copyright (c) 2023, Tri Dao.
import torch
import torch.nn.functional as F
from einops import rearrange, repeat
from realhf.base import constants
class IndexFirstAxis(torch.autograd.Function):
@staticmethod
def forward(ctx, input, indices):
ctx.save_for_backward(indices)
assert input.ndim >= 2
ctx.first_axis_dim, other_shape = input.shape[0], input.shape[1:]
second_dim = other_shape.numel()
# TD [2022-03-04] For some reason torch.gather is a bit faster than indexing.
# return input[indices]
return torch.gather(
rearrange(input, "b ... -> b (...)"),
0,
repeat(indices, "z -> z d", d=second_dim),
).reshape(-1, *other_shape)
@staticmethod
def backward(ctx, grad_output):
(indices,) = ctx.saved_tensors
assert grad_output.ndim >= 2
other_shape = grad_output.shape[1:]
grad_output = rearrange(grad_output, "b ... -> b (...)")
grad_input = torch.zeros(
[ctx.first_axis_dim, grad_output.shape[1]],
device=grad_output.device,
dtype=grad_output.dtype,
)
# TD [2022-03-04] For some reason torch.scatter is a bit faster than indexing.
# grad_input[indices] = grad_output
grad_input.scatter_(
0, repeat(indices, "z -> z d", d=grad_output.shape[1]), grad_output
)
return grad_input.reshape(ctx.first_axis_dim, *other_shape), None
def index_first_axis(x: torch.Tensor, indices: torch.LongTensor):
if len(x.shape) == 1:
return x[indices]
else:
return IndexFirstAxis.apply(x, indices)
class IndexPutFirstAxis(torch.autograd.Function):
@staticmethod
def forward(ctx, values, indices, first_axis_dim):
ctx.save_for_backward(indices)
assert indices.ndim == 1
assert values.ndim >= 2
output = torch.zeros(
first_axis_dim,
*values.shape[1:],
device=values.device,
dtype=values.dtype,
)
# TD [2022-03-04] For some reason torch.scatter is a bit faster than indexing.
output[indices] = values
# output.scatter_(0, repeat(indices, "z -> z d", d=values.shape[1]), values)
return output
@staticmethod
def backward(ctx, grad_output):
(indices,) = ctx.saved_tensors
# TD [2022-03-04] For some reason torch.gather is a bit faster than indexing.
grad_values = grad_output[indices]
# grad_values = torch.gather(grad_output, 0, repeat(indices, "z -> z d", d=grad_output.shape[1]))
return grad_values, None, None
def index_put_first_axis(
values: torch.Tensor, indices: torch.LongTensor, first_axis_dim: int
):
if len(values.shape) == 1:
output = torch.zeros(first_axis_dim, device=values.device, dtype=values.dtype)
output[indices] = values
return output
else:
return IndexPutFirstAxis.apply(values, indices, first_axis_dim)
class IndexFirstAxisResidual(torch.autograd.Function):
@staticmethod
def forward(ctx, input, indices):
ctx.save_for_backward(indices)
assert input.ndim >= 2
ctx.first_axis_dim, other_shape = input.shape[0], input.shape[1:]
second_dim = other_shape.numel()
# TD [2022-03-04] For some reason torch.gather is a bit faster than indexing.
output = input[indices]
# We don't want to reshape input (b ... -> b (...)) since it could change the channel_last
# memory format to channel_first. In other words, input might not be contiguous.
# If we don't detach, Pytorch complains about output being a view and is being modified inplace
return output, input.detach()
@staticmethod
def backward(ctx, grad_output, grad_residual):
(indices,) = ctx.saved_tensors
assert grad_output.ndim >= 2
other_shape = grad_output.shape[1:]
assert grad_residual.shape[1:] == other_shape
grad_input = grad_residual
# grad_input[indices] += grad_output
indices = indices.reshape(indices.shape[0], *((1,) * (grad_output.ndim - 1)))
indices = indices.expand_as(grad_output)
grad_input.scatter_add_(0, indices, grad_output)
return grad_input.reshape(ctx.first_axis_dim, *other_shape), None
index_first_axis_residual = IndexFirstAxisResidual.apply
def unpad_input(hidden_states, attention_mask):
"""
Arguments:
hidden_states: (batch, seqlen, ...)
attention_mask: (batch, seqlen), bool / int, 1 means valid and 0 means not valid.
Return:
hidden_states: (total_nnz, ...), where total_nnz = number of tokens in selected in attention_mask.
cu_seqlens: (batch + 1), the cumulative sequence lengths, used to index into hidden_states.
max_seqlen_in_batch: int
"""
seqlens_in_batch = attention_mask.sum(dim=-1, dtype=torch.int32)
indices = torch.nonzero(attention_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)
)
# TD [2022-03-04] We don't want to index with a bool mask, because Pytorch will expand the
# bool mask, then call nonzero to get the indices, then index with those. The indices is @dim
# times larger than it needs to be, wasting memory. It's faster and more memory-efficient to
# index with integer indices. Moreover, torch's index is a bit slower than it needs to be,
# so we write custom forward and backward to make it a bit faster.
return (
index_first_axis(rearrange(hidden_states, "b s ... -> (b s) ..."), indices),
indices,
cu_seqlens,
max_seqlen_in_batch,
)
def unpad_input_for_concatenated_sequences(hidden_states, attention_mask_in_length):
"""
Supports concatenating short samples in one sequence.
The attention_mask_in_length is utilized to mask other short samples.
It helps efficient training of variant lengths-based samples
(e.g., the supervised fine-tuning task in large language model).
The motivation for this function is explained
[here](https://github.com/Dao-AILab/flash-attention/issues/432#issuecomment-1668822286).
For example, if batch = 3 and seqlen = 6, the attention_mask_in_length is:
```
[
[2, 3, 0, 0, 0, 0],
[3, 2, 0, 0, 0, 0],
[6, 0, 0, 0, 0, 0]
]
```
, which refers to the 3D-attention mask:
```
[
[
[1, 0, 0, 0, 0, 0],
[1, 1, 0, 0, 0, 0],
[0, 0, 1, 0, 0, 0],
[0, 0, 1, 1, 0, 0],
[0, 0, 1, 1, 1, 0],
[0, 0, 0, 0, 0, 1]
],
[
[1, 0, 0, 0, 0, 0],
[1, 1, 0, 0, 0, 0],
[1, 1, 1, 0, 0, 0],
[0, 0, 0, 1, 0, 0],
[0, 0, 0, 1, 1, 0],
[0, 0, 0, 0, 0, 1]
],
[
[1, 0, 0, 0, 0, 0],
[1, 1, 0, 0, 0, 0],
[1, 1, 1, 0, 0, 0],
[1, 1, 1, 1, 0, 0],
[1, 1, 1, 1, 1, 0],
[1, 1, 1, 1, 1, 1]
]
]
```.
Arguments:
hidden_states: (batch, seqlen, ...)
attention_mask_in_length: (batch, seqlen), int, a nonzero number (e.g., 1, 2, 3, etc.) means length of concatenated sequence in b-th batch, and 0 means none.
Return:
hidden_states: (total_nnz, ...), where total_nnz = number of tokens in selected in attention_mask.
cu_seqlens: (batch + 1), the cumulative sequence lengths, used to index into hidden_states.
max_seqlen_in_batch: int
"""
length = attention_mask_in_length.sum(dim=-1)
seqlen = attention_mask_in_length.size(-1)
attention_mask_2d = torch.arange(
seqlen, device=length.device, dtype=length.dtype
).expand(len(length), seqlen) < length.unsqueeze(1)
real_indices_idx = torch.nonzero(
attention_mask_in_length.flatten(), as_tuple=False
).flatten()
seqlens_in_batch = attention_mask_in_length.flatten()[real_indices_idx]
indices = torch.nonzero(attention_mask_2d.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)
)
# TD [2022-03-04] We don't want to index with a bool mask, because Pytorch will expand the
# bool mask, then call nonzero to get the indices, then index with those. The indices is @dim
# times larger than it needs to be, wasting memory. It's faster and more memory-efficient to
# index with integer indices. Moreover, torch's index is a bit slower than it needs to be,
# so we write custom forward and backward to make it a bit faster.
return (
index_first_axis(rearrange(hidden_states, "b s ... -> (b s) ..."), indices),
indices,
cu_seqlens,
max_seqlen_in_batch,
)
def pad_input(hidden_states, indices, batch, seqlen):
"""
Arguments:
hidden_states: (total_nnz, ...), where total_nnz = number of tokens in selected in attention_mask.
indices: (total_nnz)
Return:
hidden_states: (batch, seqlen, ...)
"""
dim = hidden_states.shape[-1]
# output = torch.zeros((batch * seqlen), dim, device=hidden_states.device, dtype=hidden_states.dtype)
# output[indices] = hidden_states
output = index_put_first_axis(hidden_states, indices, batch * seqlen)
return rearrange(output, "(b s) ... -> b s ...", b=batch)
def pad_sequence_parallel_input(
packed_input_ids: torch.Tensor, cu_seqlens: torch.Tensor, max_seqlen: int
):
"""Sequence parallel requires packed_input_ids has a shape of 1 dimension
[total_seq_len], and total_seq_len should be divisible by
tensor_parallel_world_size. This function is used to pad packed_input_ids to
suitable length with an empty sequence, and return new packed_input_ids,
cu_seqlens and max_seqlen.
Args:
packed_input_ids (torch.Tensor): unpadded packed_input_ids
cu_seqlens (torch.Tensor): unpadded cu_seqlens
max_seqlen (int): unpadded max_seqlen
Returns:
(torch.Tensor, torch.Tensor, int, int): padded (packed_input_ids, cu_seqlens, max_seqlen, pad_size)
"""
tp_world_size = constants.tensor_parallel_world_size()
pad_size = 0
if len(packed_input_ids) % tp_world_size != 0:
pad_size = tp_world_size - len(packed_input_ids) % tp_world_size
packed_input_ids = torch.nn.functional.pad(
packed_input_ids, (0, pad_size), value=1
)
cu_seqlens = torch.nn.functional.pad(
cu_seqlens, (0, 1), value=len(packed_input_ids)
)
max_seqlen = max_seqlen if pad_size < max_seqlen else pad_size
return packed_input_ids, cu_seqlens, max_seqlen, pad_size
def pad_sequence_parallel_generate_input(
packed_input_ids: torch.Tensor, cu_seqlens: torch.Tensor, max_seqlen: int
):
"""Only for pipeline generate input when model+seq parallel is enabled. To
make sure inputs for seq parallel model have a shape with first dimension
divisible by tensor_parallel_world_size, the packed_input_ids should have
length divisible by tensor_parallel_world_size, and contains number of
sequences divisible by tensor_parallel_world_size.
Args:
packed_input_ids (torch.Tensor): unpadded packed_input_ids
cu_seqlens (torch.Tensor): unpadded cu_seqlens
max_seqlen (int): unpadded max_seqlen
Returns:
(torch.Tensor, torch.Tensor, int, int, int): padded (packed_input_ids, cu_seqlens, max_seqlen, pad_size, pad_seq_size)
"""
tp_world_size = constants.tensor_parallel_world_size()
pad_size, pad_seq_size = 0, 0
if (
len(packed_input_ids) % tp_world_size != 0
or (len(cu_seqlens) - 1) % tp_world_size != 0
):
pad_size = tp_world_size - len(packed_input_ids) % tp_world_size
pad_seq_size = tp_world_size - (len(cu_seqlens) - 1) % tp_world_size
if pad_size < pad_seq_size:
pad_size += tp_world_size
pad_cu_seqlens = torch.tensor(list(range(1, pad_seq_size)) + [pad_size]) + len(
packed_input_ids
)
pad_cu_seqlens = pad_cu_seqlens.to(
dtype=cu_seqlens.dtype, device=cu_seqlens.device
)
packed_input_ids = torch.nn.functional.pad(
packed_input_ids, (0, pad_size), value=1
)
cu_seqlens = torch.cat([cu_seqlens, pad_cu_seqlens], dim=0)
max_seqlen = (
max_seqlen
if (pad_size - pad_seq_size + 1) < max_seqlen
else (pad_size - pad_seq_size + 1)
)
return packed_input_ids, cu_seqlens, max_seqlen, pad_size, pad_seq_size