Skip to content

Cherry pick Fix callbacks in DSV3 script (14350) into r2.4.0 #14370

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

Merged
chtruong814 merged 1 commit into from
Aug 4, 2025
Merged

Cherry pick Fix callbacks in DSV3 script (14350) into r2.4.0 #14370

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
2 changes: 1 addition & 1 deletion scripts/performance/executors.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -71,7 +71,7 @@ def slurm_executor(
custom_bash_cmds = [] if custom_bash_cmds is None else custom_bash_cmds
err_msgs = []
mounts = []
srun_args = custom_srun_args.copy() + ["--mpi=pmix"]
srun_args = custom_srun_args.copy() + ["--mpi=pmix", "--no-container-mount-home"]

if log_dir != get_nemorun_home():
err_msgs.append(f"\nRun `export NEMORUN_HOME={log_dir}` in your shell environment and rerun this script.")
Expand Down
3 changes: 3 additions & 0 deletions scripts/performance/llm/pretrain_deepseek_v3.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -19,6 +19,7 @@

from nemo.collections.llm.recipes.deepseek_v3 import pretrain_recipe
from nemo.collections.nlp.modules.common.tokenizer_utils import get_nmt_tokenizer
from nemo.lightning.pytorch.callbacks.deepep import DeepEPCallback
from nemo.lightning.pytorch.callbacks.megatron_enable_experimental_callback import MegatronEnableExperimentalCallback
from nemo.lightning.pytorch.callbacks.moe_token_drop import MegatronTokenDropCallback
from nemo.lightning.run.plugins import MemoryProfilePlugin, NsysPlugin, PerfEnvPlugin
Expand Down Expand Up @@ -73,6 +74,8 @@ def override_recipe_configs(
recipe.model.config.moe_token_dispatcher_type = "flex"
recipe.model.config.moe_enable_deepep = True
recipe.model.config.moe_shared_expert_overlap = False # not supported for deepEP
# use force load balance for reducing variance in benchmarking
recipe.model.config.moe_router_force_load_balancing = True
else:
recipe.model.config.moe_token_dispatcher_type = "alltoall"
recipe.model.config.moe_enable_deepep = False
Expand Down
4 changes: 2 additions & 2 deletions scripts/performance/recommended_model_configs/model_configs_h100.csv
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -30,8 +30,8 @@ pre_train,nemotronh,8b,h100,fp8,8,8192,1,1,1,1,1,1,128,,1,0,0,0,,,,
pre_train,nemotronh,47b,h100,fp8,64,8192,8,1,1,1,1,1,192,,1,0,0,0,,,,
pre_train,nemotronh,56b,h100,fp8,64,8192,8,1,1,1,1,1,192,,1,0,0,0,,,,
pre_train,nemotronh,56b,h100,fp8,256,8192,8,1,1,1,1,1,768,,1,0,0,0,,,,
pre_train,deepseek,v3,h100,bf16,1024.0,4096.0,2.0,16.0,1.0,64.0,1.0,1.0,8192.0,1.0,0,0,0,0,mla_up_proj/mlp,,,
pre_train,deepseek,v3,h100,fp8,1024.0,4096.0,2.0,16.0,1.0,64.0,1.0,1.0,8192.0,1.0,0,0,0,0,mla_up_proj/mlp,,,
pre_train,deepseek,v3,h100,bf16,1024.0,4096.0,2.0,16.0,1.0,64.0,1.0,1.0,8192.0,1.0,0,0,0,0,mla_up_proj/mlp/moe,,,
pre_train,deepseek,v3,h100,fp8,1024.0,4096.0,2.0,16.0,1.0,64.0,1.0,1.0,8192.0,1.0,0,0,0,0,mla_up_proj/mlp/moe,,,
pre_train,llama4,e16,h100,bf16,256.0,8192.0,4.0,1.0,1.0,16.0,1.0,1.0,1024.0,4.0,1,0,0,0,,,,
pre_train,llama4,e128,h100,bf16,512.0,8192.0,4.0,1.0,1.0,128.0,1.0,1.0,1024.0,4.0,1,0,0,0,,,,
pre_train,vlm_llama4,e16,h100,bf16,256.0,8192.0,4.0,1.0,1.0,16.0,1.0,1.0,1024.0,4.0,0,0,0,0,,,,
Expand Down
Loading