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

【AutoParallel】Update rotary_emb in auto_parallel #8475

Closed
wants to merge 4 commits into from
Closed

【AutoParallel】Update rotary_emb in auto_parallel #8475

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
11e5e65
update rotary_emb in auto_parallel
heavyrain-lzy May 21, 2024
4a964b2
update rotay_emb
heavyrain-lzy May 22, 2024
86b7d19
update rotay_emb
heavyrain-lzy May 22, 2024
89fd4b6
add case for auto_parallel
heavyrain-lzy May 22, 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
4 changes: 2 additions & 2 deletions paddlenlp/transformers/llama/modeling_auto.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -448,7 +448,7 @@
assert past_key_value is None, "fuse rotary not support cache kv for now"
batch_size, seq_length, num_heads, head_dim = query_states.shape
_, kv_seq_len, num_key_value_heads, _ = key_states.shape
cos, sin = self.rotary_emb(value_states, seq_len=kv_seq_len)
cos, sin, _ = self.rotary_emb(value_states, seq_len=kv_seq_len)

Check warning on line 451 in paddlenlp/transformers/llama/modeling_auto.py

View check run for this annotation

Codecov / codecov/patch

paddlenlp/transformers/llama/modeling_auto.py#L451 

Added line #L451 was not covered by tests

paddle_version = float(paddle.__version__[:3])
if ((paddle_version != 0.0) and (paddle_version <= 2.6)) and (num_heads != num_key_value_heads):
Expand Down Expand Up @@ -481,7 +481,7 @@
use_neox_rotary_style=False,
)
else:
cos, sin = self.rotary_emb(value_states, seq_len=kv_seq_len)
cos, sin, _ = self.rotary_emb(value_states, seq_len=kv_seq_len)

Check warning on line 484 in paddlenlp/transformers/llama/modeling_auto.py

View check run for this annotation

Codecov / codecov/patch

paddlenlp/transformers/llama/modeling_auto.py#L484 

Added line #L484 was not covered by tests
# hack here, because elementwise infer spmd not support broadcast now
query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin, position_ids)

Expand Down
4 changes: 2 additions & 2 deletions paddlenlp/transformers/llama/modeling_auto_static.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -421,7 +421,7 @@
if self.config.rope:
if self.use_fused_rope:
assert past_key_value is None, "fuse rotary not support cache kv for now"
cos, sin = self.rotary_emb(value_states, seq_len=kv_seq_len)
cos, sin, _ = self.rotary_emb(value_states, seq_len=kv_seq_len)

Check warning on line 424 in paddlenlp/transformers/llama/modeling_auto_static.py

View check run for this annotation

Codecov / codecov/patch

paddlenlp/transformers/llama/modeling_auto_static.py#L424 

Added line #L424 was not covered by tests
query_states, key_states, _ = fused_rotary_position_embedding(
query_states,
key_states,
Expand All @@ -432,7 +432,7 @@
use_neox_rotary_style=False,
)
else:
cos, sin = self.rotary_emb(value_states, seq_len=kv_seq_len)
cos, sin, _ = self.rotary_emb(value_states, seq_len=kv_seq_len)

Check warning on line 435 in paddlenlp/transformers/llama/modeling_auto_static.py

View check run for this annotation

Codecov / codecov/patch

paddlenlp/transformers/llama/modeling_auto_static.py#L435 

Added line #L435 was not covered by tests
query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin, position_ids)

# [bs, seq_len, num_head, head_dim]
Expand Down
1 change: 1 addition & 0 deletions scripts/distribute/run_ci.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -30,6 +30,7 @@ target_lists_for_llama=(
"paddlenlp/trainer/auto_trainer.py"
"paddlenlp/transformers/llama/modeling_auto_static.py"
"paddlenlp/transformers/llama/modeling_auto.py"
"paddlenlp/transformers/llama/modeling.py"
"scripts/distribute"
)

Expand Down
Loading