ray/rllib/models/torch/modules/relative_multi_head_attention.py
2020-06-23 20:42:30 +02:00

133 lines
4.7 KiB
Python

from ray.rllib.utils.framework import try_import_torch
from ray.rllib.models.torch.misc import SlimFC
from ray.rllib.utils.torch_ops import sequence_mask
torch, nn = try_import_torch()
class RelativeMultiHeadAttention(nn.Module):
"""A RelativeMultiHeadAttention layer as described in [3].
Uses segment level recurrence with state reuse.
"""
def __init__(self,
in_dim,
out_dim,
num_heads,
head_dim,
rel_pos_encoder,
input_layernorm=False,
output_activation=None,
**kwargs):
"""Initializes a RelativeMultiHeadAttention nn.Module object.
Args:
in_dim (int):
out_dim (int):
num_heads (int): The number of attention heads to use.
Denoted `H` in [2].
head_dim (int): The dimension of a single(!) attention head
Denoted `D` in [2].
rel_pos_encoder (:
input_layernorm (bool): Whether to prepend a LayerNorm before
everything else. Should be True for building a GTrXL.
output_activation (Optional[tf.nn.activation]): Optional tf.nn
activation function. Should be relu for GTrXL.
**kwargs:
"""
super().__init__(**kwargs)
# No bias or non-linearity.
self._num_heads = num_heads
self._head_dim = head_dim
# 3=Query, key, and value inputs.
self._qkv_layer = SlimFC(
in_size=in_dim, out_size=3 * num_heads * head_dim, use_bias=False)
self._linear_layer = SlimFC(
in_size=num_heads * head_dim,
out_size=out_dim,
use_bias=False,
activation_fn=output_activation)
self._pos_proj = SlimFC(
in_size=in_dim, out_size=num_heads * head_dim, use_bias=False)
self._uvar = torch.zeros(num_heads, head_dim)
self._vvar = torch.zeros(num_heads, head_dim)
nn.init.xavier_uniform_(self._uvar)
nn.init.xavier_uniform_(self._vvar)
self._rel_pos_encoder = rel_pos_encoder
self._input_layernorm = None
if input_layernorm:
self._input_layernorm = torch.nn.LayerNorm(in_dim)
def forward(self, inputs, memory=None):
T = list(inputs.size())[1] # length of segment (time)
H = self._num_heads # number of attention heads
d = self._head_dim # attention head dimension
# Add previous memory chunk (as const, w/o gradient) to input.
# Tau (number of (prev) time slices in each memory chunk).
Tau = list(memory.shape)[1] if memory is not None else 0
if memory is not None:
memory.requires_grad_(False)
inputs = torch.cat((memory, inputs), dim=1)
# Apply the Layer-Norm.
if self._input_layernorm is not None:
inputs = self._input_layernorm(inputs)
qkv = self._qkv_layer(inputs)
queries, keys, values = torch.chunk(input=qkv, chunks=3, dim=-1)
# Cut out Tau memory timesteps from query.
queries = queries[:, -T:]
queries = torch.reshape(queries, [-1, T, H, d])
keys = torch.reshape(keys, [-1, T + Tau, H, d])
values = torch.reshape(values, [-1, T + Tau, H, d])
R = self._pos_proj(self._rel_pos_encoder)
R = torch.reshape(R, [T + Tau, H, d])
# b=batch
# i and j=time indices (i=max-timesteps (inputs); j=Tau memory space)
# h=head
# d=head-dim (over which we will reduce-sum)
score = torch.einsum("bihd,bjhd->bijh", queries + self._uvar, keys)
pos_score = torch.einsum("bihd,jhd->bijh", queries + self._vvar, R)
score = score + self.rel_shift(pos_score)
score = score / d**0.5
# causal mask of the same length as the sequence
mask = sequence_mask(
torch.arange(Tau + 1, T + Tau + 1), dtype=score.dtype)
mask = mask[None, :, :, None]
masked_score = score * mask + 1e30 * (mask.to(torch.float32) - 1.)
wmat = nn.functional.softmax(masked_score, dim=2)
out = torch.einsum("bijh,bjhd->bihd", wmat, values)
shape = list(out.shape)[:2] + [H * d]
out = torch.reshape(out, shape)
return self._linear_layer(out)
@staticmethod
def rel_shift(x):
# Transposed version of the shift approach described in [3].
# https://github.com/kimiyoung/transformer-xl/blob/
# 44781ed21dbaec88b280f74d9ae2877f52b492a5/tf/model.py#L31
x_size = list(x.shape)
x = torch.nn.functional.pad(x, (0, 0, 1, 0, 0, 0, 0, 0))
x = torch.reshape(x, [x_size[0], x_size[2] + 1, x_size[1], x_size[3]])
x = x[:, 1:, :, :]
x = torch.reshape(x, x_size)
return x