Skip to content

Update Nemotron 3 Nano perf configs (#2510) #2560

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
ko3n1g merged 1 commit into from
Feb 26, 2026
+45 −28
Merged

Update Nemotron 3 Nano perf configs (#2510) #2560

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
30 changes: 25 additions & 5 deletions scripts/performance/configs/nemotronh/nemotron_3_nano_llm_pretrain.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -31,7 +31,9 @@ def set_nemotron_3_nano_common_configs(cfg: ConfigContainer) -> None:
cfg.ddp.grad_reduce_in_fp32 = False


def nemotron_3_nano_pretrain_config_gb300(precision: str = "bf16", config_variant: str = "v1") -> ConfigContainer:
def nemotron_3_nano_pretrain_config_gb300(
precision: str = "bf16", mock: bool = True, config_variant: str = "v1"
) -> ConfigContainer:
"""GB300, baseline config."""
base_cfg = get_workload_base_config(
model_family_name="nemotronh",
Expand All @@ -47,11 +49,15 @@ def nemotron_3_nano_pretrain_config_gb300(precision: str = "bf16", config_varian
cfg.mixed_precision = precision_config
set_nemotron_3_nano_common_configs(cfg)
set_workload_base_configs(cfg, base_cfg)
if base_cfg.moe_flex_dispatcher_backend is not None:
cfg.model.moe_flex_dispatcher_backend = base_cfg.moe_flex_dispatcher_backend

return cfg


def nemotron_3_nano_pretrain_config_gb200(precision: str = "bf16", config_variant: str = "v1") -> ConfigContainer:
def nemotron_3_nano_pretrain_config_gb200(
precision: str = "bf16", mock: bool = True, config_variant: str = "v1"
) -> ConfigContainer:
"""GB200, baseline config."""
base_cfg = get_workload_base_config(
model_family_name="nemotronh",
Expand All @@ -67,11 +73,15 @@ def nemotron_3_nano_pretrain_config_gb200(precision: str = "bf16", config_varian
cfg.mixed_precision = precision_config
set_nemotron_3_nano_common_configs(cfg)
set_workload_base_configs(cfg, base_cfg)
if base_cfg.moe_flex_dispatcher_backend is not None:
cfg.model.moe_flex_dispatcher_backend = base_cfg.moe_flex_dispatcher_backend

return cfg


def nemotron_3_nano_pretrain_config_b300(precision: str = "bf16", config_variant: str = "v1") -> ConfigContainer:
def nemotron_3_nano_pretrain_config_b300(
precision: str = "bf16", mock: bool = True, config_variant: str = "v1"
) -> ConfigContainer:
"""B300, baseline config."""
base_cfg = get_workload_base_config(
model_family_name="nemotronh",
Expand All @@ -87,11 +97,15 @@ def nemotron_3_nano_pretrain_config_b300(precision: str = "bf16", config_variant
cfg.mixed_precision = precision_config
set_nemotron_3_nano_common_configs(cfg)
set_workload_base_configs(cfg, base_cfg)
if base_cfg.moe_flex_dispatcher_backend is not None:
cfg.model.moe_flex_dispatcher_backend = base_cfg.moe_flex_dispatcher_backend

return cfg


def nemotron_3_nano_pretrain_config_b200(precision: str = "bf16", config_variant: str = "v1") -> ConfigContainer:
def nemotron_3_nano_pretrain_config_b200(
precision: str = "bf16", mock: bool = True, config_variant: str = "v1"
) -> ConfigContainer:
"""B200, baseline config."""
base_cfg = get_workload_base_config(
model_family_name="nemotronh",
Expand All @@ -107,11 +121,15 @@ def nemotron_3_nano_pretrain_config_b200(precision: str = "bf16", config_variant
cfg.mixed_precision = precision_config
set_nemotron_3_nano_common_configs(cfg)
set_workload_base_configs(cfg, base_cfg)
if base_cfg.moe_flex_dispatcher_backend is not None:
cfg.model.moe_flex_dispatcher_backend = base_cfg.moe_flex_dispatcher_backend

return cfg


def nemotron_3_nano_pretrain_config_h100(precision: str = "bf16", config_variant: str = "v1") -> ConfigContainer:
def nemotron_3_nano_pretrain_config_h100(
precision: str = "bf16", mock: bool = True, config_variant: str = "v1"
) -> ConfigContainer:
"""H100, baseline config."""
base_cfg = get_workload_base_config(
model_family_name="nemotronh",
Expand All @@ -127,5 +145,7 @@ def nemotron_3_nano_pretrain_config_h100(precision: str = "bf16", config_variant
cfg.mixed_precision = precision_config
set_nemotron_3_nano_common_configs(cfg)
set_workload_base_configs(cfg, base_cfg)
if base_cfg.moe_flex_dispatcher_backend is not None:
cfg.model.moe_flex_dispatcher_backend = base_cfg.moe_flex_dispatcher_backend

return cfg
43 changes: 20 additions & 23 deletions scripts/performance/configs/nemotronh/nemotron_3_nano_workload_base_configs.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -30,39 +30,36 @@

BASE_NEMOTRON_3_NANO_CONFIG = WorkloadBaseConfig(
num_gpus=8,
global_batch_size=3072,
global_batch_size=512,
micro_batch_size=2,
tensor_model_parallel_size=4,
tensor_model_parallel_size=1,
expert_tensor_parallel_size=1,
expert_model_parallel_size=8,
moe_flex_dispatcher_backend="hybridep",
)

NEMOTRON_3_NANO_PRETRAIN_CONFIG_GB300_BF16_V1 = replace(
BASE_NEMOTRON_3_NANO_CONFIG,
tensor_model_parallel_size=1,
)
NEMOTRON_3_NANO_PRETRAIN_CONFIG_GB300_FP8_MX_V1 = NEMOTRON_3_NANO_PRETRAIN_CONFIG_GB300_BF16_V1
NEMOTRON_3_NANO_PRETRAIN_CONFIG_GB300_BF16_V1 = BASE_NEMOTRON_3_NANO_CONFIG
NEMOTRON_3_NANO_PRETRAIN_CONFIG_GB300_FP8_MX_V1 = BASE_NEMOTRON_3_NANO_CONFIG

NEMOTRON_3_NANO_PRETRAIN_CONFIG_GB200_BF16_V1 = replace(
BASE_NEMOTRON_3_NANO_CONFIG,
tensor_model_parallel_size=1,
)
NEMOTRON_3_NANO_PRETRAIN_CONFIG_GB200_FP8_MX_V1 = NEMOTRON_3_NANO_PRETRAIN_CONFIG_GB200_BF16_V1
NEMOTRON_3_NANO_PRETRAIN_CONFIG_GB200_BF16_V1 = BASE_NEMOTRON_3_NANO_CONFIG
NEMOTRON_3_NANO_PRETRAIN_CONFIG_GB200_FP8_MX_V1 = BASE_NEMOTRON_3_NANO_CONFIG

NEMOTRON_3_NANO_PRETRAIN_CONFIG_B300_BF16_V1 = replace(
BASE_NEMOTRON_3_NANO_CONFIG,
tensor_model_parallel_size=1,
)
NEMOTRON_3_NANO_PRETRAIN_CONFIG_B300_FP8_MX_V1 = NEMOTRON_3_NANO_PRETRAIN_CONFIG_B300_BF16_V1
NEMOTRON_3_NANO_PRETRAIN_CONFIG_B300_BF16_V1 = BASE_NEMOTRON_3_NANO_CONFIG
NEMOTRON_3_NANO_PRETRAIN_CONFIG_B300_FP8_MX_V1 = BASE_NEMOTRON_3_NANO_CONFIG

NEMOTRON_3_NANO_PRETRAIN_CONFIG_B200_BF16_V1 = BASE_NEMOTRON_3_NANO_CONFIG
NEMOTRON_3_NANO_PRETRAIN_CONFIG_B200_FP8_MX_V1 = BASE_NEMOTRON_3_NANO_CONFIG

NEMOTRON_3_NANO_PRETRAIN_CONFIG_B200_BF16_V1 = replace(

_NEMOTRON_3_NANO_PRETRAIN_CONFIG_H100 = replace(
BASE_NEMOTRON_3_NANO_CONFIG,
tensor_model_parallel_size=1,
num_gpus=16,
global_batch_size=1024,
micro_batch_size=1,
recompute_modules=["moe", "layernorm"],
)
NEMOTRON_3_NANO_PRETRAIN_CONFIG_B200_FP8_MX_V1 = NEMOTRON_3_NANO_PRETRAIN_CONFIG_B200_BF16_V1

NEMOTRON_3_NANO_PRETRAIN_CONFIG_H100_BF16_V1 = BASE_NEMOTRON_3_NANO_CONFIG
NEMOTRON_3_NANO_PRETRAIN_CONFIG_H100_FP8_CS_V1 = BASE_NEMOTRON_3_NANO_CONFIG
NEMOTRON_3_NANO_PRETRAIN_CONFIG_H100_BF16_V1 = _NEMOTRON_3_NANO_PRETRAIN_CONFIG_H100
NEMOTRON_3_NANO_PRETRAIN_CONFIG_H100_FP8_CS_V1 = _NEMOTRON_3_NANO_PRETRAIN_CONFIG_H100

__all__ = [
"NEMOTRON_3_NANO_PRETRAIN_CONFIG_GB300_BF16_V1",
Expand Down
Loading