[CI] Fix various failures on main (#47197)

Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
This commit is contained in:
Harry Mellor
2026-07-01 10:35:34 +01:00
committed by GitHub
parent aa8bb5562e
commit e7d0fcbc09
5 changed files with 18 additions and 14 deletions
@@ -907,7 +907,7 @@ class RoutedExperts(PluggableLayer):
# Unified loading logic for fused and non-fused experts
loaded_experts = experts_shard.unbind()
for expert_id, loaded_expert in enumerate(loaded_experts, start=start):
success = self.weight_loader(
success = param.weight_loader(
param=param,
loaded_weight=loaded_expert,
weight_name=weight_name,
@@ -1445,6 +1445,8 @@ class DeepseekV2Model(nn.Module):
hidden_states, residual = combined_states.split(
[self.hidden_size, self.hidden_size], dim=-1
)
# fused_add_rms_norm requires a contiguous residual
residual = residual.contiguous()
if idx in self.aux_hidden_state_layers:
aux_hidden_state = hidden_states + residual
if aux_hidden_state.shape[0] != positions.shape[0]:
@@ -1469,6 +1471,8 @@ class DeepseekV2Model(nn.Module):
hidden_states, residual = combined_states.split(
[self.hidden_size, self.hidden_size], dim=-1
)
# fused_add_rms_norm requires a contiguous residual
residual = residual.contiguous()
if self.end_layer in self.aux_hidden_state_layers:
aux_hidden_states.append(hidden_states + residual)
+3
View File
@@ -761,6 +761,7 @@ class Gemma3ForConditionalGeneration(
max_frames_per_batch: int,
device: torch.device,
dtype: torch.dtype,
path: str = "default",
):
from vllm.v1.worker.encoder_cudagraph_defs import (
EncoderCudaGraphCaptureInputs,
@@ -792,6 +793,7 @@ class Gemma3ForConditionalGeneration(
mm_kwargs: dict[str, Any],
max_batch_size: int,
max_frames_per_batch: int,
path: str = "default",
):
from vllm.v1.worker.encoder_cudagraph_defs import (
EncoderCudaGraphReplayBuffers,
@@ -804,6 +806,7 @@ class Gemma3ForConditionalGeneration(
def encoder_cudagraph_forward(
self,
values: dict[str, torch.Tensor],
path: str = "default",
) -> torch.Tensor:
pixel_values = values["pixel_values"]
image_features = self.vision_tower(pixel_values)