Skip to content

Respect cuda off for DPO training #324

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

Open
zpitroda wants to merge 3 commits into
base: develop
Choose a base branch
from
Open

Respect cuda off for DPO training #324

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
ddfcd15
Merge pull request #306 from mindverse/release_0428
kevin-mindverse Apr 29, 2025
245bb1e
fix:hotfix use_previous_params problem (#320)
kevin-mindverse Apr 30, 2025
632b98f
Respect cuda toggle
zpitroda Apr 30, 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
15 changes: 10 additions & 5 deletions lpm_kernel/L2/dpo/dpo_train.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -42,13 +42,18 @@ def training_data_processor(args, SYS = "You are a helpful assistant.\n\n"):
return training_data

def train(args):
# Read CUDA toggle from training params
training_params = TrainingParamsManager.get_latest_training_params()
use_cuda = training_params.get('use_cuda', False)
device = torch.device('cuda' if use_cuda and torch.cuda.is_available() else 'cpu')

tokenizer = AutoTokenizer.from_pretrained(args.base_model_path, padding_side="left")
model = AutoModelForCausalLM.from_pretrained(
args.base_model_path,
trust_remote_code=True,
ignore_mismatched_sizes=True,
torch_dtype=torch.float32, # CPU doesn't support bfloat16
)
args.base_model_path,
trust_remote_code=True,
ignore_mismatched_sizes=True,
torch_dtype="auto",
).to(device)
time_str = get_east_eight_time_formatted()

# merged_model = model.merge_and_unload()
Expand Down
3 changes: 1 addition & 2 deletions lpm_kernel/api/domains/trainprocess/routes.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -288,7 +288,6 @@ def retrain():
data_synthesis_mode: Mode for data synthesis (optional)
use_cuda: Whether to use CUDA for training (optional)
is_cot: Whether to use Chain of Thought (optional)
use_previous_params: Whether to use previous training parameters (optional, default True)

Returns:
Response: JSON response
Expand Down Expand Up @@ -318,7 +317,7 @@ def retrain():
is_cot = data.get("is_cot", None)

# Log the received parameters
logger.info(f"Retrain parameters: model_name={model_name}, learning_rate={learning_rate}, number_of_epochs={number_of_epochs}, concurrency_threads={concurrency_threads}, data_synthesis_mode={data_synthesis_mode}, use_cuda={use_cuda}, is_cot={is_cot}, use_previous_params={use_previous_params}")
logger.info(f"Retrain parameters: model_name={model_name}, learning_rate={learning_rate}, number_of_epochs={number_of_epochs}, concurrency_threads={concurrency_threads}, data_synthesis_mode={data_synthesis_mode}, use_cuda={use_cuda}, is_cot={is_cot}")

# Create training service instance
train_service = TrainProcessService(current_model_name=model_name)
Expand Down