Skip to content

fixed TP layers in profiler #30

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.

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
wants to merge 1 commit into
base: main
Choose a base branch
from
Open
Show file tree
Hide file tree
Changes from all commits
Commits
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
45 changes: 34 additions & 11 deletions examples/megatron_example.py
Original file line number Diff line number Diff line change
Expand Up @@ -79,16 +79,32 @@ def tokenize_fn(prompts):
tensor_parallel_size=2,
)

# Create a hook function
module_name = "layers.28.feed_forward.w3"
hook_function = HookFunction(
module_name=module_name,
# Define some editing function
def do_something(current_module, inputs, save_ctx, modules):
save_ctx.my_tensor = inputs
print(f"Saving tensor: {hash(inputs)}")
return inputs * 20

def do_another_thing(current_module, inputs, save_ctx, modules):
past_tensor = save_ctx.my_tensor
print(f"Retrieved tensor: {hash(past_tensor)}")
return inputs + past_tensor

# Define hook functions
my_hook_function_1 = HookFunction(
module_name="layers.20.feed_forward.w3",
expected_shape=(None, None, 13824),
editing_function=None,
editing_function=do_something,
)
my_hook_function_2 = HookFunction(
module_name="layers.28.feed_forward.w3",
expected_shape=(None, None, 13824),
editing_function=do_another_thing,
)

# Register hook function with the model
model.register_forward_hook(hook_function)
model.register_forward_hook(my_hook_function_1)
model.register_forward_hook(my_hook_function_2)

# Tokenize a prompt
inputs = tokenize_fn(prompts)
Expand All @@ -99,12 +115,19 @@ def tokenize_fn(prompts):
# Activations are only dumped to main process. Activations per-module key
# are accumulated in a list.
if torch.distributed.get_rank() == 0:
activation = activation_dict[module_name][0]
print(f"Activation shape: {activation.shape}")
print(activation)
activation_1 = activation_dict["layers.20.feed_forward.w3"][0]
print(f"Activation shape: {activation_1.shape}")
print(activation_1)

assert activation_1.shape[0] == 4
assert activation_1.shape[-1] == 13824

activation_2 = activation_dict["layers.28.feed_forward.w3"][0]
print(f"Activation shape: {activation_2.shape}")
print(activation_2)

assert activation.shape[0] == 4
assert activation.shape[-1] == 13824
assert activation_2.shape[0] == 4
assert activation_2.shape[-1] == 13824


if __name__ == "__main__":
Expand Down
4 changes: 3 additions & 1 deletion profiling/utils.py
Original file line number Diff line number Diff line change
Expand Up @@ -55,9 +55,11 @@ def __init__(self, model_dim, n_layers, is_distributed, config):
RowParallelLinear(
model_dim,
model_dim,
input_is_parallel=True,
config=config,
init_method=init_method,
bias=False,
input_is_parallel=True,
skip_bias_add=False,
),
)
for _ in range(self.n_layers // 2)
Expand Down