Skip to content

fix(pt): remove optimizer state from multi-task change-bias to reduce file size #4954

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

Draft
wants to merge 3 commits into
base: devel
Choose a base branch
from
Draft

fix(pt): remove optimizer state from multi-task change-bias to reduce file size #4954

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
10806cc
Initial plan
Copilot Sep 3, 2025
efb1cf2
fix(pt): optimize change-bias for multi-task models to save only sele…
Copilot Sep 3, 2025
5028e76
fix(pt): remove optimizer state from multi-task change-bias to reduce…
Copilot Sep 3, 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
53 changes: 44 additions & 9 deletions deepmd/pt/entrypoints/main.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,6 +1,5 @@
# SPDX-License-Identifier: LGPL-3.0-or-later
import argparse
import copy
import json
import logging
import os
Expand Down Expand Up @@ -400,7 +399,8 @@ def change_bias(
old_state_dict = torch.load(
input_file, map_location=env.DEVICE, weights_only=True
)
model_state_dict = copy.deepcopy(old_state_dict.get("model", old_state_dict))
# Only copy model_params, not the entire state dict, to avoid memory bloat
model_state_dict = old_state_dict.get("model", old_state_dict)
model_params = model_state_dict["_extra_state"]["model_params"]
elif input_file.endswith(".pth"):
old_model = torch.jit.load(input_file, map_location=env.DEVICE)
Expand Down Expand Up @@ -495,14 +495,49 @@ def change_bias(
output_path = (
output if output is not None else input_file.replace(".pt", "_updated.pt")
)
wrapper = ModelWrapper(model)
if "model" in old_state_dict:
old_state_dict["model"] = wrapper.state_dict()
old_state_dict["model"]["_extra_state"] = model_state_dict["_extra_state"]
if multi_task:
# For multi-task models, save only the selected branch as a single-head model
single_head_model = updated_model
wrapper = ModelWrapper(single_head_model)

# Create single-head model parameters
single_head_params = model_params["model_dict"][model_branch].copy()

# Save only the selected branch with single-head structure
if "model" in old_state_dict:
# For multi-task models, don't include optimizer state to reduce file size
state_to_save = {
"model": wrapper.state_dict(),
}
# Update the model's extra state to reflect single-head parameters
state_to_save["model"]["_extra_state"] = {
"model_params": single_head_params,
"train_infos": model_state_dict["_extra_state"].get(
"train_infos", {"lr": 0.001, "step": 0}
),
}
torch.save(state_to_save, output_path)
else:
state_to_save = wrapper.state_dict()
state_to_save["_extra_state"] = {
"model_params": single_head_params,
"train_infos": model_state_dict["_extra_state"].get(
"train_infos", {"lr": 0.001, "step": 0}
),
}
torch.save(state_to_save, output_path)
else:
old_state_dict = wrapper.state_dict()
old_state_dict["_extra_state"] = model_state_dict["_extra_state"]
torch.save(old_state_dict, output_path)
# For single-task models, keep existing behavior
wrapper = ModelWrapper(model)
if "model" in old_state_dict:
old_state_dict["model"] = wrapper.state_dict()
old_state_dict["model"]["_extra_state"] = model_state_dict[
"_extra_state"
]
else:
old_state_dict = wrapper.state_dict()
old_state_dict["_extra_state"] = model_state_dict["_extra_state"]
torch.save(old_state_dict, output_path)
else:
# for .pth
output_path = (
Expand Down