Skip to content

[Loader V1] modify layername for DeepSeekV3 #3336

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

Merged
merged 4 commits into from
Aug 13, 2025
Merged

[Loader V1] modify layername for DeepSeekV3 #3336

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
4749156
[Loader V1] modify layername for DeepSeekV3
zeroRains Aug 11, 2025
0fddee8
Merge branch 'develop' into description
zeroRains Aug 12, 2025
b7f294f
Merge branch 'develop' into description
yuanlehome Aug 12, 2025
bca7eba
Merge branch 'develop' into description
XieYunshen Aug 12, 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
9 changes: 4 additions & 5 deletions fastdeploy/model_executor/models/deepseek_v3.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -539,7 +539,7 @@ def __init__(
prefix="deepseek_v3.embed_tokens",
)

self.decoder_layers = nn.LayerList(
self.layers = nn.LayerList(
[
DeepSeekV3DecoderLayer(
fd_config,
Expand All @@ -564,7 +564,7 @@ def load_state_dict(self, state_dict):
self.norm.load_state_dict(state_dict)
for i in range(self.num_layers):
logger.info(f"Start load layer {i}")
self.decoder_layers[i].load_state_dict(state_dict)
self.layers[i].load_state_dict(state_dict)

def forward(
self,
Expand All @@ -578,7 +578,7 @@ def forward(

residual = None
for i in range(self.num_layers):
hidden_states, residual = self.decoder_layers[i](
hidden_states, residual = self.layers[i](
forward_meta,
hidden_states,
residual,
Expand Down Expand Up @@ -658,12 +658,11 @@ def load_weights(self, weights_iterator) -> None:

for loaded_weight_name, loaded_weight in weights_iterator:
loaded_weight_name = loaded_weight_name.replace("deepseek_v3", "model")
loaded_weight_name = loaded_weight_name.replace("layers", "decoder_layers")

for param_name, weight_name, shard_id in stacked_params_mapping:
if weight_name not in loaded_weight_name:
continue
if "mlp.experts." in loaded_weight_name and loaded_weight_name not in params_dict:
if "mlp.experts." in loaded_weight_name:
continue
model_param_name = loaded_weight_name.replace(weight_name, param_name)

Expand Down
Loading