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

neuron changes for 1B,3B,8B models #45

Open
wants to merge 2 commits into
base: mainline_upstream_do_not_merge
Choose a base branch
from
+4 −2
Open

neuron changes for 1B,3B,8B models #45

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
0f8a6ab
neuron changes for 1B,3B,8B models
aws-mengchiy Jan 2, 2025
2ce84ed
Delete duplicate stack_cfg config and setdefault input_partition_type
aws-mengchiy Jan 2, 2025
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
6 changes: 4 additions & 2 deletions axlearn/experiments/text/gpt/fuji.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -99,7 +99,10 @@ class Version(enum.Enum):
},
Version.V2: {
"test": 2 * (1024**4), # 2T tokens
"1B": 2 * (1024**4), # 2T tokens
"3B": 2 * (1024**4), # 2T tokens
"7B": 2 * (1024**4), # 2T tokens
"8B": 2 * (1024**4), # 2T tokens
"70B": 2 * (1024**4), # 2T tokens
},
Version.V3: {
Expand Down Expand Up @@ -224,7 +227,6 @@ def get_trainer_kwargs(
learner_kwargs=dict(peak_lr=3e-4, weight_decay=0.1),
max_sequence_length=max_sequence_length,
train_batch_size=train_batch_size,
input_partition_type=None if backend != "neuron" else DataPartitionType.BATCH,
max_step=max_step,
mesh_shape=mesh_shape_from_axes(data=-1, fsdp=8),
mesh_rules=(
Expand Down Expand Up @@ -416,7 +418,6 @@ def get_trainer_kwargs(
learner_kwargs=dict(peak_lr=1.5e-4, weight_decay=0.1),
max_sequence_length=max_sequence_length,
train_batch_size=train_batch_size,
input_partition_type=None if backend != "neuron" else DataPartitionType.BATCH,
max_step=max_step,
mesh_shape=mesh_shape_from_axes(fsdp=-1),
mesh_rules=(
Expand Down Expand Up @@ -464,6 +465,7 @@ def get_trainer_kwargs(
max_step=trainer_kwargs["max_step"],
**trainer_kwargs.pop("learner_kwargs"),
)
trainer_kwargs.setdefault("input_partition_type", None if backend != "neuron" else DataPartitionType.BATCH)
# pylint: enable=use-dict-literal
return trainer_kwargs

Expand Down
Loading