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

Jump to bottom

Backport: Ensure that parameters are leaf nodes when loading a model #364

Merged
merged 1 commit into from
Feb 11, 2024
Merged

Backport: Ensure that parameters are leaf nodes when loading a model #364

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
22 changes: 16 additions & 6 deletions curated_transformers/tests/models/util.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -96,8 +96,7 @@ def assert_causal_lm_output_equals_hf(
)
orig_model.eval()

for _, param in orig_model.state_dict().items():
assert param.device == torch_device
check_params_buffers(orig_model, torch_device)

hf_model = transformers.AutoModelForCausalLM.from_pretrained(
model_name,
Expand Down Expand Up @@ -153,8 +152,7 @@ def assert_decoder_output_equals_hf(
)
orig_model.eval()

for _, param in orig_model.state_dict().items():
assert param.device == torch_device
check_params_buffers(orig_model, torch_device)

hf_model = transformers.AutoModel.from_pretrained(
model_name, revision=model_revision, trust_remote_code=trust_remote_code
Expand Down Expand Up @@ -217,8 +215,7 @@ def assert_encoder_output_equals_hf(
orig_model = model_class.from_hf_hub(name=model_name, device=torch_device)
orig_model.eval()

for _, param in orig_model.state_dict().items():
assert param.device == torch_device
check_params_buffers(orig_model, torch_device)

hf_model = transformers.AutoModel.from_pretrained(model_name)
hf_model.to(torch_device)
Expand Down Expand Up @@ -362,3 +359,16 @@ def assert_model_config(model: TransformerModule, model_output: Tensor):

hidden_width = model_output.size(-1)
assert config.layer.feedforward.hidden_width == hidden_width


def check_params_buffers(model: Module, device: torch.device):
"""
Check that parameters/buffers are placed on the correct device and that
parameters are leaf nodes.
"""
for buffer in model.buffers():
assert buffer.device == device

for param in model.parameters():
assert param.device == device
assert param.is_leaf
2 changes: 1 addition & 1 deletion curated_transformers/util/serde.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -126,7 +126,7 @@ def default_tensor_to_parameter_converter(
old_param = module._parameters[parameter_name]
assert old_param is not None
_validate_replacement(old_param, tensor, module_prefix)
return Parameter(tensor, requires_grad=old_param.requires_grad).to(device=device) # type: ignore
return Parameter(tensor.to(device=device), requires_grad=old_param.requires_grad) # type: ignore


def _emplace_module_state_dict(
Expand Down
Loading