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

Llama #23

Merged
merged 3 commits into from
Jun 4, 2024
Merged

Llama #23

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
2b028cc
support Llama testing
jeffnvidia May 19, 2024
7044db7
change the FIX ME comment
jeffnvidia Jun 3, 2024
e8fd40a
remove cmd_args from gpt.toml and adjust FIXME in Llama.toml
jeffnvidia Jun 4, 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: 3 additions & 0 deletions conf/v0.6/general/test/gpt.toml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,3 @@
name = "gpt"
description = "gpt"
test_template_name = "NeMoLauncher"
13 changes: 13 additions & 0 deletions conf/v0.6/general/test/llama.toml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,13 @@
name = "llama"
description = "Llama2 70b"
test_template_name = "NeMoLauncher"
# FIXME : ~training.model.position_embedding_type was added in the extra_cmd_args in order to fix a bug from NeMo repository (https://github.com/NVIDIA/NeMo).
# the commit that should fix this issue in NeMo is : 5b296e8af832c67d361fdfb80a165db3affaf76a.
# Once the new release of NeMoLauncher includes this commit (check by downloading the corresponding container and look inside /opt for this commit), ~training.model.position_embedding_type should be removed from the extra args
extra_cmd_args = "~training.model.position_embedding_type +training.model.fsdp=True ~training.model.optim.bucket_cap_mb ~training.model.optim.overlap_grad_sync ~training.model.optim.overlap_param_sync ~training.model.optim.contiguous_grad_buffer training.model.virtual_pipeline_model_parallel_size=null training.model.megatron_amp_O2=False training.model.activations_checkpoint_num_layers=null training.model.gradient_accumulation_fusion=False training.model.use_cpu_initialization=True training.model.optim.name=fused_adam training.model.tokenizer.model=TOKENIZER_MODEL training.exp_manager.create_wandb_logger=False"

[cmd_args]
"training" = "llama/llama2_70b"
"training.trainer.max_steps" = "120"
"training.model.global_batch_size" = "256"
"training.model.pipeline_model_parallel_size" = "1"
3 changes: 0 additions & 3 deletions conf/v0.6/general/test/nemo_launcher.toml
View file
Open in desktop

This file was deleted.

10 changes: 4 additions & 6 deletions conf/v0.6/general/test_template/nemo_launcher.toml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -33,7 +33,7 @@ name = "NeMoLauncher"
default = "8"

[cmd_args.training]
values = ["gpt3/40b_improved"]
values = ["gpt3/40b_improved", "llama/llama2_70b"]
default = "gpt3/40b_improved"
[cmd_args.training.exp_manager]
[cmd_args.training.exp_manager.create_checkpoint_callback]
Expand All @@ -42,9 +42,8 @@ name = "NeMoLauncher"

[cmd_args.training.trainer]
[cmd_args.training.trainer.max_steps]
type = "preset"
values = ["100", "500", "1000", "2000"]
default = "100"
type = "int"
default = "400"

[cmd_args.training.trainer.val_check_interval]
type = "preset"
Expand All @@ -62,8 +61,7 @@ name = "NeMoLauncher"

[cmd_args.training.model]
[cmd_args.training.model.global_batch_size]
type = "preset"
values = ["8", "16", "32", "128"]
type = "int"
default = "128"

[cmd_args.training.model.micro_batch_size]
Expand Down