Skip to content

Ko3n1g/chore/reapply 2152 and 2209 #2273

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 2 commits into from
Feb 8, 2026
Merged

Ko3n1g/chore/reapply 2152 and 2209 #2273

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
f2fee27
Reapply "cp: `Update Qwen3 235B A22B MXFP8 GB200/300 recipe and resol…
ko3n1g Feb 8, 2026
595e767
Reapply "cp: `Dsv3 Recipe Update (2152)` into `r0.3.0` (#2186)"
ko3n1g Feb 8, 2026
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/configs/deepseek/deepseek_llm_pretrain.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -95,7 +95,7 @@ def deepseek_v3_pretrain_config_gb200(
pipeline_model_parallel_size=base_cfg.pipeline_model_parallel_size,
virtual_pipeline_model_parallel_size=base_cfg.virtual_pipeline_model_parallel_size,
moe_flex_dispatcher_backend=base_cfg.moe_flex_dispatcher_backend,
layout=None,
layout=base_cfg.pp_layout,
)
set_deepseek_v3_common_configs(cfg)
set_workload_base_configs(cfg, base_cfg)
Expand Down
25 changes: 7 additions & 18 deletions scripts/performance/configs/deepseek/deepseek_workload_base_configs.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -42,28 +42,20 @@
BASE_DEEPSEEK_V3_CONFIG,
num_gpus=256,
global_batch_size=2048,
pipeline_model_parallel_size=4,
virtual_pipeline_model_parallel_size=4,
expert_model_parallel_size=64,
moe_flex_dispatcher_backend="hybridep",
moe_a2a_overlap=False,
cuda_graph_impl="transformer_engine",
cuda_graph_scope=["attn", "moe_router", "moe_preprocess"],
recompute_modules=["moe_act"],
)
DEEPSEEK_V3_PRETRAIN_CONFIG_GB300_BF16_V1 = DEEPSEEK_V3_PRETRAIN_CONFIG_GB300_V1
DEEPSEEK_V3_PRETRAIN_CONFIG_GB300_FP8_CS_V1 = DEEPSEEK_V3_PRETRAIN_CONFIG_GB300_V1
DEEPSEEK_V3_PRETRAIN_CONFIG_GB300_FP8_MX_V1 = DEEPSEEK_V3_PRETRAIN_CONFIG_GB300_V1
DEEPSEEK_V3_PRETRAIN_CONFIG_GB300_NVFP4_V1 = replace(
BASE_DEEPSEEK_V3_CONFIG,
micro_batch_size=2,
pipeline_model_parallel_size=2,
virtual_pipeline_model_parallel_size=8,
pp_layout="Et*4|(t*4|)*14tmL",
expert_model_parallel_size=32,
moe_flex_dispatcher_backend="hybridep",
moe_a2a_overlap=False,
cuda_graph_scope=[],
recompute_modules=["mla_up_proj"],
)
DEEPSEEK_V3_PRETRAIN_CONFIG_GB300_BF16_V1 = DEEPSEEK_V3_PRETRAIN_CONFIG_GB300_V1
DEEPSEEK_V3_PRETRAIN_CONFIG_GB300_FP8_CS_V1 = DEEPSEEK_V3_PRETRAIN_CONFIG_GB300_V1
DEEPSEEK_V3_PRETRAIN_CONFIG_GB300_FP8_MX_V1 = DEEPSEEK_V3_PRETRAIN_CONFIG_GB300_V1
DEEPSEEK_V3_PRETRAIN_CONFIG_GB300_NVFP4_V1 = DEEPSEEK_V3_PRETRAIN_CONFIG_GB300_V1


DEEPSEEK_V3_PRETRAIN_CONFIG_GB200_V1 = replace(
Expand Down Expand Up @@ -142,10 +134,7 @@
DEEPSEEK_V3_PRETRAIN_CONFIG_GB300_BF16_V2 = DEEPSEEK_V3_PRETRAIN_CONFIG_GB300_V2
DEEPSEEK_V3_PRETRAIN_CONFIG_GB300_FP8_CS_V2 = DEEPSEEK_V3_PRETRAIN_CONFIG_GB300_V2
DEEPSEEK_V3_PRETRAIN_CONFIG_GB300_FP8_MX_V2 = DEEPSEEK_V3_PRETRAIN_CONFIG_GB300_V2
DEEPSEEK_V3_PRETRAIN_CONFIG_GB300_NVFP4_V2 = replace(
DEEPSEEK_V3_PRETRAIN_CONFIG_GB300_NVFP4_V1,
global_batch_size=4096,
)
DEEPSEEK_V3_PRETRAIN_CONFIG_GB300_NVFP4_V2 = DEEPSEEK_V3_PRETRAIN_CONFIG_GB300_V2


DEEPSEEK_V3_PRETRAIN_CONFIG_GB200_V2 = replace(
Expand Down
5 changes: 3 additions & 2 deletions scripts/performance/configs/qwen/qwen3_workload_base_configs.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -198,8 +198,8 @@
QWEN3_235B_A22B_PRETRAIN_CONFIG_GB300_FP8_CS_V1,
num_gpus=256,
pipeline_model_parallel_size=4,
virtual_pipeline_model_parallel_size=12,
expert_model_parallel_size=16,
expert_model_parallel_size=32,
cuda_graph_scope=["attn", "moe_router", "moe_preprocess"],
global_batch_size=8192,
)

Expand All @@ -217,6 +217,7 @@
QWEN3_235B_A22B_PRETRAIN_CONFIG_GB200_FP8_CS_V2 = replace(
QWEN3_235B_A22B_PRETRAIN_CONFIG_GB200_FP8_CS_V1,
num_gpus=256,
expert_model_parallel_size=32,
global_batch_size=8192,
)

Expand Down
3 changes: 3 additions & 0 deletions scripts/performance/perf_plugins.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -281,6 +281,9 @@ def _set_model_specific_environment_variables(
if model_family_name == "llama" and model_recipe_name == "llama31_405b" and train_task == "pretrain":
if compute_dtype == "fp8_cs":
del_cudnn_ln = False
if model_family_name == "deepseek":
if compute_dtype == "fp8_mx":
del_cudnn_ln = False
if del_cudnn_ln:
if "NVTE_NORM_FWD_USE_CUDNN" in executor.env_vars:
executor.env_vars.pop("NVTE_NORM_FWD_USE_CUDNN")
Expand Down
Loading