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

Enable TP+PP support #285

Closed
wants to merge 5 commits into from
Closed

Enable TP+PP support #285

Show file tree
Hide file tree
Changes from 2 commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
2b96e7c
Enable TP+PP support
wconstab Apr 29, 2024
c181377
Update
wconstab May 1, 2024
ce618b8
Update
wconstab May 2, 2024
0fd62f1
Update
wconstab May 2, 2024
50e2d2a
Update
wconstab May 2, 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
3 changes: 2 additions & 1 deletion torchtitan/parallelisms/parallelize_llama.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -205,7 +205,8 @@ def parallelize_llama(model, world_mesh, parallel_dims, job_config: JobConfig):
)

# Apply tensor + sequence parallelism to every transformer block
for layer_id, transformer_block in enumerate(model.layers):
for layer_name, transformer_block in model.layers.named_children():
# for layer_id, transformer_block in enumerate(model.layers):
layer_plan = {
"attention": PrepareModuleInput(
input_layouts=(Shard(1), None),
Expand Down
2 changes: 1 addition & 1 deletion train_configs/debug_model.toml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -35,7 +35,7 @@ warmup_steps = 2 # lr scheduler warm up, normally 20% of the train steps
max_norm = 1.0 # grad norm clipping
steps = 10
data_parallel_degree = -1
tensor_parallel_degree = 1
tensor_parallel_degree = 2
pipeline_parallel_degree = 2
fp8_linear = ""
compile = false
Expand Down
Loading