Skip to content

support dlrm eager train#382

Draft
ShawnXuan wants to merge 1 commit intomainfrom
dev_dlrm_eager
Draft

support dlrm eager train#382
ShawnXuan wants to merge 1 commit intomainfrom
dev_dlrm_eager

Conversation

@ShawnXuan
Copy link
Contributor

No description provided.

save_model("initial_checkpoint")

opt = flow.optim.SGD(dlrm_module.parameters(), lr=args.learning_rate)
lr_scheduler = make_lr_scheduler(args, opt)
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
lr_scheduler = make_lr_scheduler(args, opt)
lr_scheduler = make_lr_scheduler(args, opt)
opt = flow.one_embedding.Optimizer(
opt, embeddings=[dlrm_module.embedding.one_embedding]
)

test_case=g${num_gpus}_lr${lr}_t${train_batches}_b${train_batch_size}_d${decay_batches}
echo $test_case

export ONEFLOW_ONE_EMBEDDING_EAGER=1
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
export ONEFLOW_ONE_EMBEDDING_EAGER=1


opt = flow.optim.SGD(dlrm_module.parameters(), lr=args.learning_rate)
lr_scheduler = make_lr_scheduler(args, opt)
loss_fn = flow.nn.BCEWithLogitsLoss(reduction="none").to("cuda")
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

这里应该reduction直接设置为mean,才有可能用那个bce fuse kernel

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants