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

attention_mask not needed for training #642

Merged
merged 4 commits into from
Sep 27, 2023
Merged

attention_mask not needed for training #642

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
a2ee893
attention_mask not needed for training
winglian Sep 27, 2023
880e90d
specifically don't use attention mask for phi
winglian Sep 27, 2023
34d913c
use a different check for phi
winglian Sep 27, 2023
d66b8f0
small fixes since phi removed some values from their config
winglian Sep 27, 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: 2 additions & 6 deletions src/axolotl/models/phi/modeling_mixformer_sequential.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -711,12 +711,8 @@ def __init__(
self.resid_dropout = nn.Dropout(config.resid_pdrop)
self.block_idx = block_idx

self.mixer = MHA(config=config, **mixer, layer_idx=block_idx)
mlp_cls = mlp.pop("mlp_cls")
if mlp_cls == "fused_mlp":
self.mlp = FusedMLP(config=config, **mlp)
else:
self.mlp = MLP(config=config, **mlp)
self.mixer = MHA(config, layer_idx=block_idx)
self.mlp = MLP(config)

def forward(
self,
Expand Down
2 changes: 1 addition & 1 deletion src/axolotl/utils/data.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -76,7 +76,7 @@ def prepare_dataset(cfg, tokenizer):

with zero_first(is_main_process()):
train_dataset, eval_dataset = process_datasets_for_packing(
cfg, train_dataset, eval_dataset
cfg, train_dataset, eval_dataset, tokenizer
)
if cfg.max_steps:
total_num_steps = min(
Expand Down
9 changes: 8 additions & 1 deletion src/axolotl/utils/trainer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -397,7 +397,7 @@ def disable_datasets_caching():
set_caching_enabled(True)


def process_datasets_for_packing(cfg, train_dataset, eval_dataset):
def process_datasets_for_packing(cfg, train_dataset, eval_dataset, tokenizer):
drop_long = partial(drop_long_seq, sequence_len=cfg.sequence_len)
with zero_first(is_main_process()):
train_dataset = train_dataset.filter(drop_long, num_proc=os.cpu_count())
Expand All @@ -414,6 +414,13 @@ def process_datasets_for_packing(cfg, train_dataset, eval_dataset):
eval_dataset = eval_dataset.map(
add_position_ids, num_proc=os.cpu_count()
)

# Phi doesn't want the attention_mask feature when training
if "CodeGenTokenizer" in tokenizer.__class__.__name__:
train_dataset = train_dataset.remove_columns("attention_mask")
if eval_dataset:
eval_dataset = eval_dataset.remove_columns("attention_mask")

return train_dataset, eval_dataset


Expand Down