Skip to content

NAS export refactor + skip conversion on minitron restore #424

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
merged 2 commits into from
Oct 11, 2025
Merged

NAS export refactor + skip conversion on minitron restore #424

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
b6f831f
[Minor] Pruning doc update + bring minitron import to mtp.* instead o…
kevalmorabia97 Oct 10, 2025
7a4394e
NAS export refactor + skip conversion on minitron restore
kevalmorabia97 Oct 11, 2025
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
10 changes: 9 additions & 1 deletion examples/megatron-lm/README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -110,6 +110,8 @@ Coming soon ...

### ⭐ Pruning

Checkout pruning [getting started section](../pruning/README.md#getting-started) and [guidelines](../pruning/README.md#pruning-guidelines) for configuring pruning parameters in the pruning README.

Pruning is supported for GPT and Mamba models in Pipeline Parallel mode. Available pruning options are:

- `TARGET_FFN_HIDDEN_SIZE`
Expand All @@ -121,14 +123,20 @@ Pruning is supported for GPT and Mamba models in Pipeline Parallel mode. Availab
- `TARGET_NUM_LAYERS`
- `LAYERS_TO_DROP` (comma separated, 1-indexed list of layer numbers to directly drop)

Example for depth pruning Qwen3-8B from 36 to 24 layers:

```sh
PP=1 \
TARGET_NUM_LAYERS=24 \
HF_MODEL_CKPT=<pretrained_model_name_or_path> \
MLM_MODEL_SAVE=/tmp/Qwen3-8B-DPruned \
MLM_MODEL_SAVE=Qwen3-8B-Pruned \
bash megatron-lm/examples/post_training/modelopt/prune.sh qwen/Qwen3-8B
```

> [!TIP]
> If number of layers in the model is not divisible by pipeline parallel size (PP), you can configure uneven
> PP by setting `MLM_EXTRA_ARGS="--decoder-first-pipeline-num-layers <X> --decoder-last-pipeline-num-layers <Y>"`

## Learn More About Configuration

For simplicity, we use `shell` scripts and variables as arguments. Each script has at least 1 positional
Expand Down
4 changes: 4 additions & 0 deletions modelopt/torch/prune/__init__.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -21,6 +21,10 @@

# nas is a required - so let's check if it's available
import modelopt.torch.nas
from modelopt.torch.utils import import_plugin

from . import fastnas, gradnas, plugins
from .pruning import *

with import_plugin("mcore_minitron", verbose=False):
from .plugins import mcore_minitron
9 changes: 8 additions & 1 deletion modelopt/torch/prune/plugins/mcore_minitron.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -37,6 +37,7 @@
HAS_MAMBA,
_DynamicMCoreLanguageModel,
SUPPORTED_MODELS,
drop_mcore_language_model_layers,
)
# isort: on

Expand Down Expand Up @@ -70,7 +71,13 @@
"num_layers",
}

__all__ = ["MCoreMinitronConfig", "MCoreMinitronModeDescriptor", "MCoreMinitronSearcher"]
__all__ = [
"SUPPORTED_HPARAMS",
"MCoreMinitronConfig",
"MCoreMinitronModeDescriptor",
"MCoreMinitronSearcher",
"drop_mcore_language_model_layers",
]


class MCoreMinitronSearcher(BaseSearcher):
Expand Down