Hi Still confused on whether it is possible for rotary and relative positional embeddings to be integrated with the fast kernels in pytorch sdpa, allowing for faster training/inference? If so what would be a blue print of how to merge previous architecture that use both to the new sdpa one ?
I would want to integrate sdpa into the ESM model, available on huggingface at https://huggingface.co/docs/transformers/en/model_doc/esm. Here is the section of the code of interest, the forward call of the attention
def forward(
self,
hidden_states: torch.Tensor,
attention_mask: Optional[torch.FloatTensor] = None,
head_mask: Optional[torch.FloatTensor] = None,
encoder_hidden_states: Optional[torch.FloatTensor] = None,
encoder_attention_mask: Optional[torch.FloatTensor] = None,
past_key_value: Optional[Tuple[Tuple[torch.FloatTensor]]] = None,
output_attentions: Optional[bool] = False,
) -> Tuple[torch.Tensor]:
mixed_query_layer = self.query(hidden_states)
# If this is instantiated as a cross-attention module, the keys
# and values come from an encoder; the attention mask needs to be
# such that the encoder's padding tokens are not attended to.
is_cross_attention = encoder_hidden_states is not None
if is_cross_attention and past_key_value is not None:
# reuse k,v, cross_attentions
key_layer = past_key_value[0]
value_layer = past_key_value[1]
attention_mask = encoder_attention_mask
elif is_cross_attention:
key_layer = self.transpose_for_scores(self.key(encoder_hidden_states))
value_layer = self.transpose_for_scores(self.value(encoder_hidden_states))
attention_mask = encoder_attention_mask
elif past_key_value is not None:
key_layer = self.transpose_for_scores(self.key(hidden_states))
value_layer = self.transpose_for_scores(self.value(hidden_states))
key_layer = torch.cat([past_key_value[0], key_layer], dim=2)
value_layer = torch.cat([past_key_value[1], value_layer], dim=2)
else:
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)
# Matt: Our BERT model (which this code was derived from) scales attention logits down by sqrt(head_dim).
# ESM scales the query down by the same factor instead. Modulo numerical stability these are equivalent,
# but not when rotary embeddings get involved. Therefore, we scale the query here to match the original
# ESM code and fix rotary embeddings.
query_layer = query_layer * self.attention_head_size**-0.5
if self.is_decoder:
# if cross_attention save Tuple(torch.Tensor, torch.Tensor) of all cross attention key/value_states.
# Further calls to cross_attention layer can then reuse all cross-attention
# key/value_states (first "if" case)
# if uni-directional self-attention (decoder) save Tuple(torch.Tensor, torch.Tensor) of
# all previous decoder key/value_states. Further calls to uni-directional self-attention
# can concat previous decoder key/value_states to current projected key/value_states (third "elif" case)
# if encoder bi-directional self-attention `past_key_value` is always `None`
past_key_value = (key_layer, value_layer)
if self.position_embedding_type == "rotary":
query_layer, key_layer = self.rotary_embeddings(query_layer, key_layer)
# Take the dot product between "query" and "key" to get the raw attention scores.
attention_scores = torch.matmul(query_layer, key_layer.transpose(-1, -2))
if self.position_embedding_type == "relative_key" or self.position_embedding_type == "relative_key_query":
seq_length = hidden_states.size()[1]
position_ids_l = torch.arange(seq_length, dtype=torch.long, device=hidden_states.device).view(-1, 1)
position_ids_r = torch.arange(seq_length, dtype=torch.long, device=hidden_states.device).view(1, -1)
distance = position_ids_l - position_ids_r
positional_embedding = self.distance_embedding(distance + self.max_position_embeddings - 1)
positional_embedding = positional_embedding.to(dtype=query_layer.dtype) # fp16 compatibility
if self.position_embedding_type == "relative_key":
relative_position_scores = torch.einsum("bhld,lrd->bhlr", query_layer, positional_embedding)
attention_scores = attention_scores + relative_position_scores
elif self.position_embedding_type == "relative_key_query":
relative_position_scores_query = torch.einsum("bhld,lrd->bhlr", query_layer, positional_embedding)
relative_position_scores_key = torch.einsum("bhrd,lrd->bhlr", key_layer, positional_embedding)
attention_scores = attention_scores + relative_position_scores_query + relative_position_scores_key
if attention_mask is not None:
# Apply the attention mask is (precomputed for all layers in EsmModel forward() function)
attention_scores = attention_scores + attention_mask
# Normalize the attention scores to probabilities.
attention_probs = nn.functional.softmax(attention_scores, dim=-1)
# This is actually dropping out entire tokens to attend to, which might
# seem a bit unusual, but is taken from the original Transformer paper.
attention_probs = self.dropout(attention_probs)
# Mask heads if we want to
if head_mask is not None:
attention_probs = attention_probs * head_mask
context_layer = torch.matmul(attention_probs, value_layer)
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, attention_probs) if output_attentions else (context_layer,)
if self.is_decoder:
outputs = outputs + (past_key_value,)
return outputs
How would you change the code to replace with the pytorch function sdpa at https://pytorch.org/docs/master/generated/torch.nn.functional.scaled_dot_product_attention.html#torch.nn.functional.scaled_dot_product_attention, specifically, how/could you change the relative positional encoding into a attention mask that could be passed on to the function? Scouring online it seems possible, meaning the scaled_dot_product_attention function can expect an arbitrary mask but not sure how this would be achieved in this case.