def get_normalized_probs(
self,
net_output: Tuple[Tensor, Dict[str, List[Optional[Tensor]]]],
log_probs: bool,
sample: Optional[Dict[str, Tensor]],
):
"""Get normalized probabilities (or log probs) from a net's output."""
if hasattr(self,
"adaptive_softmax") and self.adaptive_softmax is not None:
if sample is not None:
assert "target" in sample
target = sample["target"]
else:
target = None
out = self.adaptive_softmax.get_log_prob(net_output[0],
target=target)
return out.exp_() if not log_probs else out
logits = net_output[0]
if log_probs:
return utils.log_softmax(logits,
dim=-1,
onnx_trace=self.onnx_trace)
else:
return utils.softmax(logits, dim=-1, onnx_trace=self.onnx_trace)
def forward(
self,
query,
key,
value,
attn_mask: Optional[Tensor] = None,
mask_qkv=None,
segment_labels=None,
) -> Tuple[Tensor, Optional[Tensor]]:
bsz, tgt_len, embed_dim = query.size()
# query -> Q/K/V
q = self.q_proj(query)
q *= self.scaling
k = self.k_proj(query)
v = self.v_proj(query)
# multi-head [B, T, C] -> [B, H, T, dim]
q = (q.contiguous().view(bsz, tgt_len, self.num_heads,
self.head_dim).transpose(2, 1))
k = (k.contiguous().view(bsz, -1, self.num_heads,
self.head_dim).transpose(2, 1))
v = (v.contiguous().view(bsz, -1, self.num_heads,
self.head_dim).transpose(2, 1))
src_len = k.size(2)
# Q*K
attn_weights = torch.matmul(q, k.transpose(-2, -1))
# attn_weights = UnilmMultiheadAttention.apply_sparse_mask(attn_weights, tgt_len, src_len, bsz)
if attn_mask is not None:
attn_mask = attn_mask.unsqueeze(1)
attn_weights += attn_mask
attn_weights_float = utils.softmax(attn_weights,
dim=-1,
onnx_trace=self.onnx_trace)
attn_weights = attn_weights_float.type_as(attn_weights)
attn_probs = F.dropout(
attn_weights,
p=self.dropout,
training=self.training,
)
attn = torch.matmul(attn_probs, v)
attn = attn.permute(0, 2, 1, 3).contiguous().view(bsz, src_len, -1)
attn = self.out_proj(attn)
return attn, attn_weights
def forward(
self,
query,
key: Optional[Tensor],
value: Optional[Tensor],
key_padding_mask: Optional[Tensor] = None,
incremental_state: Optional[Dict[str, Dict[str,
Optional[Tensor]]]] = None,
need_weights: bool = True,
static_kv: bool = False,
attn_mask: Optional[Tensor] = None,
before_softmax: bool = False,
need_head_weights: bool = False,
) -> Tuple[Tensor, Optional[Tensor]]:
"""Input shape: Time x Batch x Channel
Args:
key_padding_mask (ByteTensor, optional): mask to exclude
keys that are pads, of shape `(batch, src_len)`, where
padding elements are indicated by 1s.
need_weights (bool, optional): return the attention weights,
averaged over heads (default: False).
attn_mask (ByteTensor, optional): typically used to
implement causal attention, where the mask prevents the
attention from looking forward in time (default: None).
before_softmax (bool, optional): return the raw attention
weights and values before the attention softmax.
need_head_weights (bool, optional): return the attention
weights for each head. Implies *need_weights*. Default:
return the average attention weights over all heads.
"""
if need_head_weights:
need_weights = True
tgt_len, bsz, embed_dim = query.size()
# src_len = value.size(0)
assert embed_dim == self.embed_dim
assert list(query.size()) == [tgt_len, bsz, embed_dim]
if (self.enable_torch_version and not self.onnx_trace
and incremental_state is None and not static_kv):
assert key is not None and value is not None
return F.multi_head_attention_forward(
query,
key,
value,
self.embed_dim,
self.num_heads,
torch.empty([0]),
torch.cat(
(self.q_proj.bias, self.k_proj.bias, self.v_proj.bias)),
self.bias_k,
self.bias_v,
self.add_zero_attn,
self.dropout,
self.out_proj.weight,
self.out_proj.bias,
self.training,
key_padding_mask,
need_weights,
attn_mask,
use_separate_proj_weight=True,
q_proj_weight=self.q_proj.weight,
k_proj_weight=self.k_proj.weight,
v_proj_weight=self.v_proj.weight,
)
if incremental_state is not None:
saved_state = self._get_input_buffer(incremental_state)
if saved_state is not None and "prev_key" in saved_state:
# previous time steps are cached - no need to recompute
# key and value if they are static
if static_kv:
assert self.encoder_decoder_attention and not self.self_attention
key = value = None
else:
saved_state = None
if self.self_attention:
q = self.q_proj(query)
k = self.k_proj(query)
v = self.v_proj(query)
elif self.encoder_decoder_attention:
# encoder-decoder attention
q = self.q_proj(query)
if key is None:
assert value is None
k = v = None
else:
k = self.k_proj(key)
v = self.v_proj(key)
else:
assert key is not None and value is not None
q = self.q_proj(query)
k = self.k_proj(key)
v = self.v_proj(value)
q *= self.scaling
if self.bias_k is not None:
assert self.bias_v is not None
k = torch.cat([k, self.bias_k.repeat(1, bsz, 1)])
v = torch.cat([v, self.bias_v.repeat(1, bsz, 1)])
if attn_mask is not None:
attn_mask = torch.cat(
[attn_mask,
attn_mask.new_zeros(attn_mask.size(0), 1)],
dim=1)
if key_padding_mask is not None:
key_padding_mask = torch.cat(
[
key_padding_mask,
key_padding_mask.new_zeros(key_padding_mask.size(0),
1),
],
dim=1,
)
q = (q.contiguous().view(tgt_len, bsz * self.num_heads,
self.head_dim).transpose(0, 1))
if k is not None:
k = (k.contiguous().view(-1, bsz * self.num_heads,
self.head_dim).transpose(0, 1))
if v is not None:
v = (v.contiguous().view(-1, bsz * self.num_heads,
self.head_dim).transpose(0, 1))
if saved_state is not None:
# saved states are stored with shape (bsz, num_heads, seq_len, head_dim)
if "prev_key" in saved_state:
_prev_key = saved_state["prev_key"]
assert _prev_key is not None
prev_key = _prev_key.view(bsz * self.num_heads, -1,
self.head_dim)
if static_kv:
k = prev_key
else:
assert k is not None
k = torch.cat([prev_key, k], dim=1)
if "prev_value" in saved_state:
_prev_value = saved_state["prev_value"]
assert _prev_value is not None
prev_value = _prev_value.view(bsz * self.num_heads, -1,
self.head_dim)
if static_kv:
v = prev_value
else:
assert v is not None
v = torch.cat([prev_value, v], dim=1)
prev_key_padding_mask: Optional[Tensor] = None
if "prev_key_padding_mask" in saved_state:
prev_key_padding_mask = saved_state["prev_key_padding_mask"]
assert k is not None and v is not None
key_padding_mask = RelativeMultiheadAttention._append_prev_key_padding_mask(
key_padding_mask=key_padding_mask,
prev_key_padding_mask=prev_key_padding_mask,
batch_size=bsz,
src_len=k.size(1),
static_kv=static_kv,
)
saved_state["prev_key"] = k.view(bsz, self.num_heads, -1,
self.head_dim)
saved_state["prev_value"] = v.view(bsz, self.num_heads, -1,
self.head_dim)
saved_state["prev_key_padding_mask"] = key_padding_mask
# In this branch incremental_state is never None
assert incremental_state is not None
incremental_state = self._set_input_buffer(incremental_state,
saved_state)
assert k is not None
src_len = k.size(1)
# This is part of a workaround to get around fork/join parallelism
# not supporting Optional types.
if key_padding_mask is not None and key_padding_mask.dim() == 0:
key_padding_mask = None
if key_padding_mask is not None:
assert key_padding_mask.size(0) == bsz
assert key_padding_mask.size(1) == src_len
if self.add_zero_attn:
assert v is not None
src_len += 1
k = torch.cat([k, k.new_zeros((k.size(0), 1) + k.size()[2:])],
dim=1)
v = torch.cat([v, v.new_zeros((v.size(0), 1) + v.size()[2:])],
dim=1)
if attn_mask is not None:
attn_mask = torch.cat(
[attn_mask,
attn_mask.new_zeros(attn_mask.size(0), 1)],
dim=1)
if key_padding_mask is not None:
key_padding_mask = torch.cat(
[
key_padding_mask,
torch.zeros(key_padding_mask.size(0),
1).type_as(key_padding_mask),
],
dim=1,
)
attn_weights = torch.bmm(q, k.transpose(
1, 2)) # [bsz * head, tgt_len, tgt_len]
# relative position
if self.maximum_relative_position and self.self_attention:
keys_length = k.size(1)
relative_pos = self.relative_positions(
keys_length, self.maximum_relative_position).to(k.device)
relative_repr_keys = self.relative_position_keys(relative_pos)
relative_repr_values = self.relative_position_values(relative_pos)
attn_weights += self.matmul_with_relative_representations(
q, relative_repr_keys)
else:
relative_repr_keys = relative_repr_values = None
# # TODO: sparse operation, need to be implemented later
# attn_weights = MultiheadAttention.apply_sparse_mask(attn_weights, tgt_len, src_len, bsz)
assert list(
attn_weights.size()) == [bsz * self.num_heads, tgt_len, src_len]
if attn_mask is not None:
attn_mask = attn_mask.unsqueeze(0)
if self.onnx_trace:
attn_mask = attn_mask.repeat(attn_weights.size(0), 1, 1)
attn_weights += attn_mask
if key_padding_mask is not None:
# don't attend to padding symbols
attn_weights = attn_weights.view(bsz, self.num_heads, tgt_len,
src_len)
attn_weights = attn_weights.masked_fill(
key_padding_mask.unsqueeze(1).unsqueeze(2).to(torch.bool),
float("-inf"))
attn_weights = attn_weights.view(bsz * self.num_heads, tgt_len,
src_len)
if before_softmax:
return attn_weights, v
attn_weights_float = utils.softmax(attn_weights,
dim=-1,
onnx_trace=self.onnx_trace)
# attn_weights = attn_weights_float.type_as(attn_weights) # useless
attn_probs = F.dropout(
attn_weights_float.type_as(attn_weights),
p=self.dropout,
training=self.training,
)
assert v is not None
attn = torch.bmm(attn_probs, v)
# relative position
if relative_repr_values is not None:
attn += self.matmul_with_relative_representations(
attn_probs, relative_repr_values, transpose=False)
assert list(
attn.size()) == [bsz * self.num_heads, tgt_len, self.head_dim]
if self.onnx_trace and attn.size(1) == 1:
# when ONNX tracing a single decoder step (sequence length == 1)
# the transpose is a no-op copy before view, thus unnecessary
attn = attn.contiguous().view(tgt_len, bsz, embed_dim)
else:
attn = attn.transpose(0,
1).contiguous().view(tgt_len, bsz, embed_dim)
attn = self.out_proj(attn)
attn_weights: Optional[Tensor] = None
if need_weights:
attn_weights = attn_weights_float.view(bsz, self.num_heads,
tgt_len,
src_len).transpose(1, 0)
if not need_head_weights:
# average attention weights over heads
attn_weights = attn_weights.mean(dim=0)
return attn, attn_weights
def forward(
self,
query,
key: Optional[Tensor],
value: Optional[Tensor],
key_padding_mask: None,
need_weights: bool = True,
static_kv: bool = False,
attn_mask: Optional[Tensor] = None,
before_softmax: bool = False,
need_head_weights: bool = False,
):
if need_head_weights:
need_weights = True
tgt_len, bsz, embed_dim = query.size()
q = self.q_proj(query)
k = self.k_proj(query)
v = self.v_proj(query)
q = (
q.contiguous()
.view(tgt_len, bsz * self.num_heads, self.head_dim)
.transpose(0, 1)
)
if k is not None:
k = (
k.contiguous()
.view(-1, bsz * self.num_heads, self.head_dim)
.transpose(0, 1)
)
if v is not None:
v = (
v.contiguous()
.view(-1, bsz * self.num_heads, self.head_dim)
.transpose(0, 1)
)
assert k is not None
src_len = k.size(1)
attn_weights = torch.bmm(q, k.transpose(1, 2))
assert list(attn_weights.size()) == [bsz * self.num_heads, tgt_len, src_len]
if before_softmax:
return attn_weights, v
attn_weights_float = utils.softmax(
attn_weights, dim=-1, onnx_trace=self.onnx_trace
)
# attn_weights = attn_weights_float.type_as(attn_weights) # useless
attn_probs = F.dropout(
attn_weights_float.type_as(attn_weights),
p=self.dropout,
training=self.training,
)
assert v is not None
attn = torch.bmm(attn_probs, v)
attn = attn.transpose(0, 1).contiguous().view(tgt_len, bsz, embed_dim)
attn = self.out_proj(attn)
return attn, attn_weights