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

Jump to bottom

[LLM] Reconstruct fused transformer layers #7186

Merged
merged 5 commits into from
Oct 12, 2023
Merged

[LLM] Reconstruct fused transformer layers #7186

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
4a981fa
reconstruct fused_transformer_layers
RichardWooSJTU Oct 10, 2023
89ce638
delete origin class
RichardWooSJTU Oct 10, 2023
c6ac7d5
merge develop
RichardWooSJTU Oct 10, 2023
86283ad
code refine
RichardWooSJTU Oct 12, 2023
c3dcf12
merge develop
RichardWooSJTU Oct 12, 2023
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
8 changes: 6 additions & 2 deletions paddlenlp/experimental/transformers/bloom/modeling.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -21,7 +21,8 @@
from paddlenlp_ops import get_padding_offset

from paddlenlp.experimental.transformers.fused_transformer_layers import (
FusedMultiTransformer,
FusedMultiTransformerBase,
FusedMultiTransformerConfig,
)
from paddlenlp.experimental.transformers.generation_utils import (
GenerationInferenceModel,
Expand Down Expand Up @@ -112,7 +113,8 @@
ffn1_bias_attrs = [paddle.ParamAttr(name="fusemt.{}.ffn1_bias".format(i)) for i in range(config.n_layer)]
ffn2_weight_attrs = [paddle.ParamAttr(name="fusemt.{}.ffn2_weight".format(i)) for i in range(config.n_layer)]
ffn2_bias_attrs = [paddle.ParamAttr(name="fusemt.{}.ffn2_bias".format(i)) for i in range(config.n_layer)]
self.transformer_block = FusedMultiTransformer(

transformer_config = FusedMultiTransformerConfig(

Check warning on line 117 in paddlenlp/experimental/transformers/bloom/modeling.py

View check run for this annotation

Codecov / codecov/patch

paddlenlp/experimental/transformers/bloom/modeling.py#L117 

Added line #L117 was not covered by tests
self.embed_dim,
self.n_head,
4 * self.embed_dim,
Expand All @@ -133,6 +135,8 @@
ffn2_weight_attrs=ffn2_weight_attrs,
ffn2_bias_attrs=ffn2_bias_attrs,
)

self.transformer_block = FusedMultiTransformerBase(transformer_config)

Check warning on line 139 in paddlenlp/experimental/transformers/bloom/modeling.py

View check run for this annotation

Codecov / codecov/patch

paddlenlp/experimental/transformers/bloom/modeling.py#L139 

Added line #L139 was not covered by tests
self.cache_kvs = []

# Final Layer Norm
Expand Down
7 changes: 5 additions & 2 deletions paddlenlp/experimental/transformers/chatglm/modeling.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -20,7 +20,8 @@
from paddlenlp_ops import get_padding_offset

from paddlenlp.experimental.transformers.fused_transformer_layers import (
FusedMultiTransformer,
FusedMultiTransformerBase,
FusedMultiTransformerConfig,
)
from paddlenlp.experimental.transformers.generation_utils import (
GenerationInferenceModel,
Expand Down Expand Up @@ -183,7 +184,8 @@
]
ffn2_bias_attrs = [paddle.ParamAttr(name="fusemt.{}.ffn2_bias".format(i)) for i in range(config.num_layers)]
alpha = (2 * self.config.num_hidden_layers) ** 0.5
self.transformer_block = FusedMultiTransformer(

transformer_config = FusedMultiTransformerConfig(

Check warning on line 188 in paddlenlp/experimental/transformers/chatglm/modeling.py

View check run for this annotation

Codecov / codecov/patch

paddlenlp/experimental/transformers/chatglm/modeling.py#L188 

Added line #L188 was not covered by tests
config.hidden_size,
config.num_attention_heads,
4 * config.hidden_size,
Expand All @@ -209,6 +211,7 @@
norm_type="layernorm",
use_neox_rotary_style=True,
)
self.transformer_block = FusedMultiTransformerBase(transformer_config)

Check warning on line 214 in paddlenlp/experimental/transformers/chatglm/modeling.py

View check run for this annotation

Codecov / codecov/patch

paddlenlp/experimental/transformers/chatglm/modeling.py#L214 

Added line #L214 was not covered by tests

def remove_padding(self, input_ids, seq_lens_this_time):
cum_offsets_now = paddle.cumsum(paddle.max(seq_lens_this_time) - seq_lens_this_time)
Expand Down
Loading