Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

modify block_multihead_attention api #8456

Merged
merged 5 commits into from
May 28, 2024
Merged

modify block_multihead_attention api #8456

Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
336131f
modify block_multihead_attention api
ming1753 May 15, 2024
cfa05a1
Merge branch 'develop' of https://github.com/PaddlePaddle/PaddleNLP i…
ming1753 May 21, 2024
9bec47b
add param to blha
ming1753 May 23, 2024
7634b64
modify fused_transformer_layers
ming1753 May 23, 2024
11c122e
fix bug
ming1753 May 23, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
17 changes: 17 additions & 0 deletions paddlenlp/experimental/transformers/fused_transformer_layers.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -584,6 +584,13 @@ def compute_qkv(self, src, residual_input, i):
qkv_out = self.compute_qkv_linear(ln_out, i)
return qkv_out, residual_input

def compute_max_len(self, seq_lens_encoder, seq_lens_decoder, cum_offsets):
if seq_lens_encoder is None or seq_lens_decoder is None or cum_offsets is None:
return None, None
return paddle.incubate.nn.functional.blha_get_max_len(
seq_lens_encoder, seq_lens_decoder, cum_offsets # cum_offsets.shape[0] used as bsz
)

def compute_fmha(
self,
qkv_out,
Expand Down Expand Up @@ -816,6 +823,12 @@ def forward(

assert self.num_layers == len(self.qkv_weights)

max_enc_len_this_time, max_dec_len_this_time = self.compute_max_len(
kwargs.get("seq_lens_encoder", None), kwargs.get("seq_lens_decoder", None), cum_offsets
)
kwargs["max_enc_len_this_time"] = max_enc_len_this_time
kwargs["max_dec_len_this_time"] = max_dec_len_this_time

residual_input = src
for i in range(self.num_layers):
qkv_out, residual_input = self.compute_qkv(src, residual_input, i)
Expand Down Expand Up @@ -1385,6 +1398,8 @@ def compute_attn(
None, # qkv_bias
None, # out_shifts
None, # out_smooths
kwargs.get("max_enc_len_this_time", None),
kwargs.get("max_dec_len_this_time", None),
rotary_embs,
attn_mask,
kwargs.get("tgt_mask", None),
Expand Down Expand Up @@ -1471,6 +1486,8 @@ def compute_attn(
self.qkv_biases[i] if len(self.qkv_biases) > 0 else None,
self.linear_shifts[i] if len(self.linear_shifts) > 0 else None,
self.linear_smooths[i] if len(self.linear_smooths) > 0 else None,
kwargs.get("max_enc_len_this_time", None),
kwargs.get("max_dec_len_this_time", None),
rotary_embs,
attn_mask,
kwargs.get("tgt_mask", None),
Expand Down
Loading