Skip to content

Fix for when draft model hidden dimension is different from target model hidden dimension #183

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
wants to merge 2 commits into
base: main
Choose a base branch
from
Open

Fix for when draft model hidden dimension is different from target model hidden dimension #183

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
204e3d5
init
yilian49 Aug 26, 2025
36442ca
update draft model
yilian49 Aug 26, 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
10 changes: 8 additions & 2 deletions scripts/train_eagle3_offline.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -226,8 +226,14 @@ def main():
.cuda()
.to(torch.bfloat16)
)
draft_model.load_embedding(args.target_model_path, embedding_key=args.embedding_key)
draft_model.freeze_embedding()
if (
not hasattr(draft_model_config, "target_hidden_size")
or draft_model_config.target_hidden_size == draft_model_config.hidden_size
):
Comment on lines +229 to +232
Copy link
Contributor

Choose a reason for hiding this comment

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

medium

This condition can be made more concise and idiomatic by using getattr with a default value. This avoids explicitly checking for the attribute's existence before accessing it.

On a related note, this entire conditional block is duplicated in scripts/train_eagle3_online.py. Consider refactoring the model setup logic into a shared utility function to improve maintainability and avoid future inconsistencies.

Suggested change
if (
not hasattr(draft_model_config, "target_hidden_size")
or draft_model_config.target_hidden_size == draft_model_config.hidden_size
):
if (
getattr(draft_model_config, "target_hidden_size", draft_model_config.hidden_size)
== draft_model_config.hidden_size
):

draft_model.load_embedding(
args.target_model_path, embedding_key=args.embedding_key
)
draft_model.freeze_embedding()
print_with_rank("Initialized draft model")

# build dataloaders
Expand Down
10 changes: 8 additions & 2 deletions scripts/train_eagle3_online.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -259,8 +259,14 @@ def main():
.cuda()
.to(torch.bfloat16)
)
draft_model.load_embedding(args.target_model_path, embedding_key=args.embedding_key)
draft_model.freeze_embedding()
if (
not hasattr(draft_model_config, "target_hidden_size")
or draft_model_config.target_hidden_size == draft_model_config.hidden_size
):
Comment on lines +262 to +265
Copy link
Contributor

Choose a reason for hiding this comment

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

medium

This condition can be made more concise and idiomatic by using getattr with a default value. This avoids explicitly checking for the attribute's existence before accessing it.

On a related note, this entire conditional block is duplicated from scripts/train_eagle3_offline.py. Consider refactoring the model setup logic into a shared utility function to improve maintainability and avoid future inconsistencies.

Suggested change
if (
not hasattr(draft_model_config, "target_hidden_size")
or draft_model_config.target_hidden_size == draft_model_config.hidden_size
):
if (
getattr(draft_model_config, "target_hidden_size", draft_model_config.hidden_size)
== draft_model_config.hidden_size
):

draft_model.load_embedding(
args.target_model_path, embedding_key=args.embedding_key
)
draft_model.freeze_embedding()
print_with_rank("Initialized draft model")

# build dataloaders
Expand Down
15 changes: 8 additions & 7 deletions specforge/modeling/draft/llama3_eagle.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -793,14 +793,15 @@ def __init__(self, config, quant_config=None, attention_backend="sdpa") -> None:
)
self.midlayer = LlamaDecoderLayer(config, attention_backend=attention_backend)

self.hidden_size = config.hidden_size
if hasattr(config, "target_hidden_size"):
self.fc = torch.nn.Linear(
config.target_hidden_size * 3, config.hidden_size, bias=False
)
self.target_hidden_size = config.target_hidden_size
else:
self.fc = torch.nn.Linear(
config.hidden_size * 3, config.hidden_size, bias=False
)
self.target_hidden_size = config.hidden_size

self.fc = torch.nn.Linear(
self.target_hidden_size * 3, self.hidden_size, bias=False
)

self.norm = LlamaRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
self.lm_head = nn.Linear(
Expand Down Expand Up @@ -874,7 +875,7 @@ def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor:

def project_hidden_states(self, hidden_states: torch.Tensor) -> torch.Tensor:
# eagle 3 requires hidden states from 3 layers
assert hidden_states.size(-1) == self.config.hidden_size * 3
assert hidden_states.size(-1) == self.target_hidden_size * 3
return self.fc(hidden_states)

def compute_logits(self, hidden_states: torch.Tensor) -> torch.Tensor:
Expand Down