Skip to content

[graph_trainer] Fix missing CP wiring in llama3 and deepseek_v3 parallelize #2808

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

Closed
SherlockNoMad wants to merge 1 commit into from
Closed

[graph_trainer] Fix missing CP wiring in llama3 and deepseek_v3 parallelize #2808

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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: 7 additions & 1 deletion torchtitan/experiments/graph_trainer/deepseek_v3/parallelize.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16,7 +16,7 @@
TrainingConfig,
)
from torchtitan.distributed import ParallelDims

from torchtitan.distributed.context_parallel import apply_cp_to_attention_module
from torchtitan.distributed.tensor_parallel import maybe_enable_async_tp
from torchtitan.experiments.graph_trainer.common_utils import (
annotate_ac_regions,
Expand Down Expand Up @@ -138,6 +138,12 @@ def parallelize_deepseekv3(
pad_multiple=pad_multiple,
)

if parallel_dims.cp_enabled:
apply_cp_to_attention_module(
[block.attention.inner_attention for block in model.layers.values()],
parallel_dims.get_mesh("cp"),
)

if ac_config.mode != "none":
apply_graph_ac(compile_config, ac_config)

Expand Down
9 changes: 9 additions & 0 deletions torchtitan/experiments/graph_trainer/llama3/parallelize.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,6 +15,7 @@
TrainingConfig,
)
from torchtitan.distributed import ParallelDims
from torchtitan.distributed.context_parallel import apply_cp_to_attention_module
from torchtitan.distributed.tensor_parallel import maybe_enable_async_tp
from torchtitan.experiments.graph_trainer.common_utils import (
annotate_ac_regions,
Expand Down Expand Up @@ -101,9 +102,17 @@ def parallelize_llama(
tp_mesh,
enable_loss_parallel=not parallelism.disable_loss_parallel,
enable_float8_tensorwise_tp=enable_float8_tensorwise_tp,
enable_cp=parallel_dims.cp_enabled,
enable_sp=parallelism.enable_sequence_parallel,
)
maybe_enable_async_tp(parallelism, compile_config, tp_mesh)

if parallel_dims.cp_enabled:
apply_cp_to_attention_module(
[block.attention.inner_attention for block in model.layers.values()],
parallel_dims.get_mesh("cp"),
)

if ac_config.mode != "none":
apply_graph_ac(compile_config, ac_config)

Expand Down
Loading