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

deepseek overflow fix #349

Merged
merged 1 commit into from
Jan 6, 2025
Merged

deepseek overflow fix #349

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
14 changes: 10 additions & 4 deletions vllm/model_executor/models/deepseek_v2.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -147,11 +147,11 @@ def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
shared_output = self.shared_experts(hidden_states)
# router_logits: (num_tokens, n_experts)
router_logits, _ = self.gate(hidden_states)
final_hidden_states = self.experts(
hidden_states=hidden_states,
router_logits=router_logits) * self.routed_scaling_factor
final_hidden_states = self.experts(hidden_states=hidden_states,
router_logits=router_logits)
if shared_output is not None:
final_hidden_states = final_hidden_states + shared_output
final_hidden_states = final_hidden_states + shared_output \
* (1. / self.routed_scaling_factor)
if self.tp_size > 1:
final_hidden_states = tensor_model_parallel_all_reduce(
final_hidden_states)
Expand Down Expand Up @@ -375,6 +375,7 @@ def __init__(
eps=config.rms_norm_eps)
self.post_attention_layernorm = RMSNorm(config.hidden_size,
eps=config.rms_norm_eps)
self.routed_scaling_factor = config.routed_scaling_factor

def forward(
self,
Expand All @@ -399,9 +400,14 @@ def forward(
)

# Fully Connected
if isinstance(self.mlp, DeepseekV2MoE):
hidden_states *= 1. / self.mlp.routed_scaling_factor
Copy link
Collaborator

Choose a reason for hiding this comment

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

Just curious, why not /= ?

Copy link
Author

Choose a reason for hiding this comment

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

Division is slower? Not sure whether this can be optimized

hidden_states, residual = self.post_attention_layernorm(
hidden_states, residual)
hidden_states = self.mlp(hidden_states)
if isinstance(self.mlp, DeepseekV2MLP):
hidden_states *= 1. / self.routed_scaling_factor
residual *= 1. / self.routed_scaling_factor
return hidden_states, residual


Expand Down
Loading