Skip to content
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

add is_trainable in kwargs #363

Merged
merged 1 commit into from
May 15, 2023
Merged

add is_trainable in kwargs #363

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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: 15 additions & 0 deletions tests/test_peft_models.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -191,3 +191,18 @@ def test_load_pretrained_peft(self):
for p1, p2 in zip(model.named_parameters(), model_from_pretrained.named_parameters()):
if p1[0] not in ["v_head.summary.weight", "v_head.summary.bias"]:
self.assertTrue(torch.allclose(p1[1], p2[1]), msg=f"{p1[0]} != {p2[0]}")

def test_continue_training_peft_model(self):
r"""
Load peft and checks that it can continue training.
"""
causal_lm_model = AutoModelForCausalLM.from_pretrained(self.causal_lm_model_id)
pretrained_model = get_peft_model(causal_lm_model, self.lora_config)

with tempfile.TemporaryDirectory() as tmp_dir:
pretrained_model.save_pretrained(tmp_dir)
# set is_trainable to True
model = AutoModelForCausalLMWithValueHead.from_pretrained(tmp_dir, is_trainable=True)
# Check that the number of trainable parameters is correct
nb_trainable_params = sum(p.numel() for p in model.parameters() if p.requires_grad)
self.assertEqual(nb_trainable_params, 10273)
6 changes: 5 additions & 1 deletion trl/models/modeling_base.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -107,9 +107,11 @@ class and the arguments that are specific to trl models. The kwargs
"""
if kwargs is not None:
peft_config = kwargs.pop("peft_config", None)
is_trainable = kwargs.pop("is_trainable", False)
trl_model_args, pretrained_kwargs, peft_int8_kwargs = cls._split_kwargs(kwargs)
else:
peft_config = None
is_trainable = False
trl_model_args = {}
pretrained_kwargs = {}
peft_int8_kwargs = {}
Expand Down Expand Up @@ -163,7 +165,9 @@ class and the arguments that are specific to trl models. The kwargs
peft_config.base_model_name_or_path, *model_args, **pretrained_kwargs
)

pretrained_model = PeftModel.from_pretrained(pretrained_model, pretrained_model_name_or_path)
pretrained_model = PeftModel.from_pretrained(
pretrained_model, pretrained_model_name_or_path, is_trainable=is_trainable
)
else:
pretrained_model = cls.transformers_parent_class.from_pretrained(
pretrained_model_name_or_path, *model_args, **pretrained_kwargs
Expand Down