Skip to content

[TRTLLM-7008][fix] fix wideEP weights loading and args #6789

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 2 commits into from
Aug 12, 2025
+5 −5
Merged

[TRTLLM-7008][fix] fix wideEP weights loading and args #6789

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
2cf31b5
relax tensor device type check to fix wideEP loading and fix argument
dongxuy04 Aug 11, 2025
0e8540c
move comment forward
dongxuy04 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
2 changes: 1 addition & 1 deletion tensorrt_llm/_torch/modules/fused_moe/fused_moe_wide_ep.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -533,7 +533,7 @@ def forward_chunk(
self.fc31_input_scale,
self.scaling_vector_size,
sfUseUE8M0=False,
swizzedLayout=False)
isSfSwizzledLayout=False)
x_sf = x_sf.view((x_row, -1))

elif self.has_deepseek_fp8_block_scales:
Expand Down
8 changes: 4 additions & 4 deletions tensorrt_llm/_torch/modules/fused_moe/quantization.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -347,8 +347,8 @@ def load_expert_w3_w1_weight(self, module: torch.nn.Module,
Load w1 and w3 weights for each expert.
Override this method if you need to preprocess the weights differently.
"""
# device don't have to be 'cuda', e.g. 'cpu' for online EPLB
device = dst_w3_w1_weight.device
assert device.type == "cuda"
w1_weight_shard = load_weight_shard(w1_weight,
module.tp_size,
module.tp_rank,
Expand All @@ -372,8 +372,8 @@ def load_expert_w2_weight(self, module: torch.nn.Module,
Load w2 weight for each expert.
Override this method if you need to preprocess the weights differently.
"""
# device don't have to be 'cuda', e.g. 'cpu' for online EPLB
device = dst_w2_weight.device
assert device.type == "cuda"
w2_weight_shard = load_weight_shard(w2_weight,
module.tp_size,
module.tp_rank,
Expand Down Expand Up @@ -1537,8 +1537,8 @@ def load_expert_w3_w1_weight_scale_nvfp4(
self, module: torch.nn.Module, w1_weight_scale: torch.Tensor,
w3_weight_scale: torch.Tensor,
dst_w3_w1_weight_scale: torch.Tensor):
# device don't have to be 'cuda', e.g. 'cpu' for online EPLB
device = dst_w3_w1_weight_scale.device
assert device.type == "cuda"
w1_weight_scale = load_weight_shard(w1_weight_scale,
module.tp_size,
module.tp_rank,
Expand Down Expand Up @@ -1577,8 +1577,8 @@ def load_expert_w3_w1_weight_scale_nvfp4(
def load_expert_w2_weight_scale_nvfp4(self, module: torch.nn.Module,
w2_weight_scale: torch.Tensor,
dst_w2_weight_scale: torch.Tensor):
# device don't have to be 'cuda', e.g. 'cpu' for online EPLB
device = dst_w2_weight_scale.device
assert device.type == "cuda"
w2_weight_scale = load_weight_shard(w2_weight_scale,
module.tp_size,
module.tp_rank,
Expand Down