Skip to content

remove deprecated MP params #2511

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 25, 2026
Merged

remove deprecated MP params #2511

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
d36f69b
chore(beep boop 🤖): Bump `uv.lock` (main) (2026-02-24)
dimapihtar Feb 24, 2026
f2ccbb6
remove deprecated mp args
dimapihtar Feb 24, 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 3rdparty/Megatron-LM
Open in desktop
Submodule Megatron-LM updated 22 files
+1 −1 .github/copy-pr-bot.yaml
+0 −1 examples/gpt3/gpt_config.yaml
+0 −20 examples/multimodal/energon_util.py
+25 −1 megatron/core/inference/batch_dimensions_utils.py
+0 −8 megatron/core/model_parallel_config.py
+1 −9 megatron/core/models/bert/bert_model.py
+6 −5 megatron/core/models/multimodal/llava_model.py
+14 −2 megatron/core/optimizer/distrib_optimizer.py
+0 −32 megatron/core/ssm/mamba_mixer.py
+0 −22 megatron/core/tensor_parallel/layers.py
+8 −1 megatron/training/arguments.py
+0 −9 megatron/training/yaml_arguments.py
+2 −1 tests/unit_tests/inference/engines/test_dynamic_engine.py
+0 −3 tests/unit_tests/models/test_mamba_moe_model.py
+0 −2 tests/unit_tests/ssm/test_mamba_context_parallel.py
+1 −1 tests/unit_tests/ssm/test_mamba_mixer.py
+130 −0 tests/unit_tests/test_optimizer.py
+0 −1 tests/unit_tests/transformer/moe/test_upcycling.py
+0 −2 tests/unit_tests/transformer/test_multi_token_prediction.py
+1 −2 tools/checkpoint/saver_base.py
+1 −2 tools/checkpoint/saver_legacy.py
+1 −2 tools/checkpoint/saver_llava.py
1 change: 0 additions & 1 deletion src/megatron/bridge/models/deepseek/deepseek_v2_bridge.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -69,7 +69,6 @@ def provider_bridge(self, hf_pretrained: PreTrainedCausalLM) -> MLAModelProvider
provider.cross_entropy_loss_fusion = True
provider.masked_softmax_fusion = True
provider.persist_layer_norm = True
provider.async_tensor_model_parallel_allreduce = True
provider.gradient_accumulation_fusion = True

provider.hidden_dropout = 0.0
Expand Down
1 change: 0 additions & 1 deletion src/megatron/bridge/models/deepseek/deepseek_v3_bridge.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -74,7 +74,6 @@ def provider_bridge(self, hf_pretrained: PreTrainedCausalLM) -> MLAModelProvider
provider.cross_entropy_loss_fusion = True
provider.masked_softmax_fusion = True
provider.persist_layer_norm = True
provider.async_tensor_model_parallel_allreduce = True
provider.gradient_accumulation_fusion = True

provider.hidden_dropout = 0.0
Expand Down
1 change: 0 additions & 1 deletion src/megatron/bridge/models/kimi/kimi_provider.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -109,7 +109,6 @@ class KimiK2Provider(MLATransformerConfig, GPTModelProvider):
layernorm_epsilon: float = 1e-6
bf16: bool = True
params_dtype: torch.dtype = torch.bfloat16
async_tensor_model_parallel_allreduce: bool = True
attention_softmax_in_fp32: bool = False
persist_layer_norm: bool = True
num_layers_in_first_pipeline_stage: Optional[int] = None
Expand Down
1 change: 0 additions & 1 deletion src/megatron/bridge/models/model_provider.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -466,7 +466,6 @@ class ModelParallelKwargs(TypedDict, total=False):
context_parallel_size: int
expert_model_parallel_size: int
expert_tensor_parallel_size: int
moe_extended_tp: bool
sequence_parallel: bool
virtual_pipeline_model_parallel_size: int | None
hierarchical_context_parallel_sizes: list[int] | None
Expand Down
1 change: 0 additions & 1 deletion src/megatron/bridge/models/qwen_vl/qwen3_vl_provider.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -244,7 +244,6 @@ class Qwen3VLMoEModelProvider(GPTModelProvider):
bias_dropout_fusion: bool = True
masked_softmax_fusion: bool = False # Don't fuse masked softmax (Qwen specific)
deallocate_pipeline_outputs: bool = True
async_tensor_model_parallel_allreduce: bool = True
distribute_saved_activations: bool = False
cp_comm_type: str = "p2p"

Expand Down
2 changes: 0 additions & 2 deletions src/megatron/bridge/recipes/nemotronh/nemotron_3_nano.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -160,7 +160,6 @@ def nemotron_3_nano_pretrain_config() -> ConfigContainer:

cfg.model.init_method_std = 0.0173
cfg.model.apply_rope_fusion = False
cfg.model.async_tensor_model_parallel_allreduce = True
cfg.model.gradient_accumulation_fusion = True
cfg.model.use_fused_weighted_squared_relu = True

Expand Down Expand Up @@ -331,7 +330,6 @@ def _nemotron_3_nano_finetune_common(
expert_tensor_parallel_size=expert_tensor_parallelism,
expert_model_parallel_size=expert_model_parallelism,
apply_rope_fusion=False,
async_tensor_model_parallel_allreduce=True,
attention_backend="fused",
gradient_accumulation_fusion=True,
init_method_std=0.0173,
Expand Down
1 change: 0 additions & 1 deletion src/megatron/bridge/training/model_load_save.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -382,7 +382,6 @@ def load_megatron_model(
model_cfg.context_parallel_size = 1
model_cfg.expert_model_parallel_size = 1
model_cfg.expert_tensor_parallel_size = 1
model_cfg.moe_extended_tp = False
model_cfg.sequence_parallel = False
model_cfg.perform_initialization = False
model_cfg.virtual_pipeline_model_parallel_size = None
Expand Down
1 change: 0 additions & 1 deletion tests/unit_tests/peft/test_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -62,7 +62,6 @@ def __init__(self):
self.perform_initialization = True
self.use_cpu_initialization = False
self.gradient_accumulation_fusion = False
self.async_tensor_model_parallel_allreduce = False


class MockColumnParallelLinear(ColumnParallelLinear):
Expand Down
3 changes: 0 additions & 3 deletions tests/unit_tests/training/test_model_load_save.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -525,7 +525,6 @@ def test_load_megatron_model_resets_defaults(self, mock_load_model_config, mock_
cfg.context_parallel_size = 2
cfg.expert_model_parallel_size = 2
cfg.expert_tensor_parallel_size = 2
cfg.moe_extended_tp = True
cfg.sequence_parallel = True
cfg.virtual_pipeline_model_parallel_size = 2
cfg.hierarchical_context_parallel_sizes = [2, 2]
Expand All @@ -545,7 +544,6 @@ def test_load_megatron_model_resets_defaults(self, mock_load_model_config, mock_
assert cfg.context_parallel_size == 1
assert cfg.expert_model_parallel_size == 1
assert cfg.expert_tensor_parallel_size == 1
assert cfg.moe_extended_tp is False
assert cfg.sequence_parallel is False
assert cfg.virtual_pipeline_model_parallel_size is None
assert cfg.hierarchical_context_parallel_sizes is None
Expand All @@ -561,7 +559,6 @@ def test_load_megatron_model_applies_overrides(self, mock_load_model_config, moc
cfg.context_parallel_size = 1
cfg.expert_model_parallel_size = 1
cfg.expert_tensor_parallel_size = 1
cfg.moe_extended_tp = False
cfg.sequence_parallel = False
cfg.virtual_pipeline_model_parallel_size = None
cfg.hierarchical_context_parallel_sizes = None
Expand Down
Loading
Loading