diff --git a/vllm_ascend/ops/register_custom_ops.py b/vllm_ascend/ops/register_custom_ops.py
index 06a52ae1..b7100991 100644
--- a/vllm_ascend/ops/register_custom_ops.py
+++ b/vllm_ascend/ops/register_custom_ops.py
@@ -116,7 +116,7 @@ def _maybe_prefetch_mlp_gate_up_proj_impl(x_dependency: torch.Tensor,
     except AssertionError:
         return
 
-    if not forward_context.prefetch_mlp_enabled:
+    if not getattr(forward_context, 'prefetch_mlp_enabled', False):
         return
     model_instance = forward_context.model_instance
     prefetch_stream = forward_context.prefetch_stream
@@ -173,7 +173,7 @@ def _maybe_prefetch_mlp_down_proj_impl(x_dependency: torch.Tensor) -> None:
     except AssertionError:
         return
 
-    if not forward_context.prefetch_mlp_enabled:
+    if not getattr(forward_context, 'prefetch_mlp_enabled', False):
         return
     forward_context.prefetch_mlp_down_proj = True
     model_instance = forward_context.model_instance
@@ -202,7 +202,7 @@ def _maybe_wait_prefetch_done_impl(x: torch.Tensor) -> None:
     except AssertionError:
         return
 
-    if not forward_context.prefetch_mlp_enabled:
+    if not getattr(forward_context, 'prefetch_mlp_enabled', False):
         return
     if forward_context.prefetch_mlp_gate_up_proj or \
         forward_context.prefetch_mlp_down_proj:
diff --git a/vllm_ascend/patch/worker/patch_qwen2_5_omni.py b/vllm_ascend/patch/worker/patch_qwen2_5_omni.py
index f52d1a1d..c272edb3 100644
--- a/vllm_ascend/patch/worker/patch_qwen2_5_omni.py
+++ b/vllm_ascend/patch/worker/patch_qwen2_5_omni.py
@@ -18,8 +18,7 @@
 import torch
 import torch.nn as nn
 from vllm.model_executor.models.qwen2_5_omni_thinker import (
-    Qwen2_5_VLImageInputs, Qwen2_5_VLVideoInputs,
-    Qwen2_5OmniThinkerForConditionalGeneration)
+    Qwen2_5_VLImageInputs, Qwen2_5_VLVideoInputs)
 
 from vllm_ascend.ascend_forward_context import set_ascend_forward_context
 
@@ -65,8 +64,3 @@ class AscendQwen2_5OmniThinkerForConditionalGeneration(nn.Module):
         sizes = grid_thw.prod(-1) // merge_size // merge_size
 
         return video_embeds.split(sizes.tolist())
-
-
-# NOTE: These will be removed after ascend_forward_context is refactored.
-Qwen2_5OmniThinkerForConditionalGeneration._process_image_input = AscendQwen2_5OmniThinkerForConditionalGeneration._process_image_input
-Qwen2_5OmniThinkerForConditionalGeneration._process_video_input = AscendQwen2_5OmniThinkerForConditionalGeneration._process_video_input
diff --git a/vllm_ascend/patch/worker/patch_qwen2_5_vl.py b/vllm_ascend/patch/worker/patch_qwen2_5_vl.py
index 7db4323d..62a1e67e 100644
--- a/vllm_ascend/patch/worker/patch_qwen2_5_vl.py
+++ b/vllm_ascend/patch/worker/patch_qwen2_5_vl.py
@@ -20,9 +20,9 @@ import torch
 import torch.nn as nn
 import torch.nn.functional as F
 import torch_npu
-from vllm.model_executor.models.qwen2_5_vl import (
-    Qwen2_5_VisionAttention, Qwen2_5_VLForConditionalGeneration,
-    Qwen2_5_VLImageInputs, Qwen2_5_VLVideoInputs)
+from vllm.model_executor.models.qwen2_5_vl import (Qwen2_5_VisionAttention,
+                                                   Qwen2_5_VLImageInputs,
+                                                   Qwen2_5_VLVideoInputs)
 from vllm.model_executor.models.qwen2_vl import Qwen2VisionAttention
 from vllm.model_executor.models.vision import run_dp_sharded_mrope_vision_model
 
@@ -169,7 +169,3 @@ class AscendQwen2_5_VLForConditionalGeneration(nn.Module):
 # NOTE: This will be removed after MMEncoderAttention has been extract as a CustomOp in vllm.
 Qwen2VisionAttention.forward = AscendQwen2_5_VisionAttention.forward
 Qwen2_5_VisionAttention.forward = AscendQwen2_5_VisionAttention.forward
-
-# NOTE: These will be removed after ascend_forward_context is refactored.
-Qwen2_5_VLForConditionalGeneration._process_image_input = AscendQwen2_5_VLForConditionalGeneration._process_image_input
-Qwen2_5_VLForConditionalGeneration._process_video_input = AscendQwen2_5_VLForConditionalGeneration._process_video_input