Skip to content

[bugfix] fix deepseek v4 fp8 #135

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
Jintao-Huang merged 1 commit into from
Jun 24, 2026
Merged

[bugfix] fix deepseek v4 fp8 #135

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
4 changes: 2 additions & 2 deletions src/mcore_bridge/model/gpts/deepseek_v4.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -70,7 +70,7 @@ def __init__(self, config, *args, **kwargs):
super().__init__(config, *args, **kwargs)
self.layer_type = self.config.hf_config.layer_types[self.layer_number - 1]
self.rope_layer_type = 'main' if self.layer_type == 'sliding_attention' else 'compress'
if getattr(config, 'fp8_param', False):
if config.fp8_param:
Comment thread
Jintao-Huang marked this conversation as resolved.
group_proj_in_size = self.query_projection_size // config.o_groups
del self.linear_o_group_proj
self.linear_o_group_proj = te.GroupedLinear(
Expand Down Expand Up @@ -442,7 +442,7 @@ def _set_o_group_proj_grouped(self, mg_attn, hf_state_dict, to_mcore):
scale_invs.append(param._rowwise_scale_inv)
else:
weights.append(param.data)
hf_state_dict['wo_a.weight'] = torch.cat(weights, dim=0)
hf_state_dict['wo_a.weight'] = torch.cat(weights, dim=0).view(torch.float8_e4m3fn)
Comment thread
Jintao-Huang marked this conversation as resolved.
if scale_invs:
hf_state_dict['wo_a.weight_scale_inv'] = torch.cat(scale_invs, dim=0)

Expand Down
Loading