diff --git a/docker/Dockerfile.npu b/docker/Dockerfile.npu
index 0d91c0b32..f83cc512a 100644
--- a/docker/Dockerfile.npu
+++ b/docker/Dockerfile.npu
@@ -1,5 +1,5 @@
 ARG VLLM_ASCEND_IMAGE=quay.io/atlas-ci/vllm-ascend
-ARG VLLM_ASCEND_VERSION=v0.28.0
+ARG VLLM_ASCEND_VERSION=v0.29.0
 ARG VLLM_ASCEND_DEVICE_SUFFIX=
 FROM ${VLLM_ASCEND_IMAGE}:${VLLM_ASCEND_VERSION}${VLLM_ASCEND_DEVICE_SUFFIX}
 
diff --git a/docker/Dockerfile.npu.ci b/docker/Dockerfile.npu.ci
index 0a150fdd8..b65b803d6 100644
--- a/docker/Dockerfile.npu.ci
+++ b/docker/Dockerfile.npu.ci
@@ -1,5 +1,5 @@
 ARG VLLM_ASCEND_IMAGE=quay.io/atlas-ci/vllm-ascend
-ARG VLLM_ASCEND_VERSION=v0.28.0
+ARG VLLM_ASCEND_VERSION=v0.29.0
 ARG VLLM_ASCEND_DEVICE_SUFFIX=
 FROM ${VLLM_ASCEND_IMAGE}:${VLLM_ASCEND_VERSION}${VLLM_ASCEND_DEVICE_SUFFIX}
 
diff --git a/docs/getting_started/installation/npu/npu.inc.md b/docs/getting_started/installation/npu/npu.inc.md
index 9ff477522..21127caf2 100644
--- a/docs/getting_started/installation/npu/npu.inc.md
+++ b/docs/getting_started/installation/npu/npu.inc.md
@@ -10,25 +10,25 @@ The recommended way to use vLLM-Omni on NPU is through the vLLM-Ascend pre-built
 === "A2"
 
     ```bash
-    export IMAGE=quay.io/atlas-ci/vllm-ascend:v0.28.0
+    export IMAGE=quay.io/atlas-ci/vllm-ascend:v0.29.0
     ```
 
 === "A3"
 
     ```bash
-    export IMAGE=quay.io/atlas-ci/vllm-ascend:v0.28.0-a3
+    export IMAGE=quay.io/atlas-ci/vllm-ascend:v0.29.0-a3
     ```
 
 === "A5"
 
     ```bash
-    export IMAGE=quay.io/atlas-ci/vllm-ascend:v0.28.0-a5
+    export IMAGE=quay.io/atlas-ci/vllm-ascend:v0.29.0-a5
     ```
 
 === "310P (Experimental)"
 
     ```bash
-    export IMAGE=quay.io/atlas-ci/vllm-ascend:v0.28.0-310p
+    export IMAGE=quay.io/atlas-ci/vllm-ascend:v0.29.0-310p
     ```
 
 ```bash
@@ -53,7 +53,7 @@ docker run --rm \
 
 # Inside the container, install vLLM-Omni from source
 cd /vllm-workspace
-git clone -b v0.28.0 https://github.com/vllm-project/vllm-omni.git
+git clone -b v0.29.0rc1 https://github.com/vllm-project/vllm-omni.git
 cd vllm-omni
 pip install -v -e . --no-build-isolation
 # or VLLM_OMNI_TARGET_DEVICE=npu pip install -v -e .
@@ -71,10 +71,10 @@ We are keeping [issue #886](https://github.com/vllm-project/vllm-omni/issues/886
 
 # --8<-- [start:installation-main]
 
-You can also install vLLM-Omni from the latest main branch if you want to use the latest features or bug fixes. Use the aligned vLLM-Ascend v0.28.0 image described above, and check [issue #886](https://github.com/vllm-project/vllm-omni/issues/886) for the status of the latest vLLM-Omni commit on NPU.
+You can also install vLLM-Omni from the latest main branch if you want to use the latest features or bug fixes. Use the aligned vLLM-Ascend v0.29.0 image described above, and check [issue #886](https://github.com/vllm-project/vllm-omni/issues/886) for the status of the latest vLLM-Omni commit on NPU.
 
 ```bash
-# Inside the vLLM-Ascend v0.28.0 container, install vLLM-Omni from main
+# Inside the vLLM-Ascend v0.29.0 container, install vLLM-Omni from main
 cd /vllm-workspace
 git clone https://github.com/vllm-project/vllm-omni.git
 cd vllm-omni
@@ -103,7 +103,7 @@ Here's an example deployment command that has been verified on 4 x NPUs:
 === "A2"
 
     ```bash
-    export IMAGE=quay.io/ascend/vllm-omni:v0.28.0
+    export IMAGE=quay.io/ascend/vllm-omni:v0.29.0
     docker run --rm \
         --name vllm-omni-a2 \
         --shm-size=64g \
@@ -127,7 +127,7 @@ Here's an example deployment command that has been verified on 4 x NPUs:
 === "A3"
 
     ```bash
-    export IMAGE=quay.io/ascend/vllm-omni:v0.28.0-a3
+    export IMAGE=quay.io/ascend/vllm-omni:v0.29.0-a3
     docker run --rm \
         --name vllm-omni-a3 \
         --shm-size=64g \
@@ -151,7 +151,7 @@ Here's an example deployment command that has been verified on 4 x NPUs:
 === "A5"
 
     ```bash
-    export IMAGE=quay.io/ascend/vllm-omni:v0.28.0-a5
+    export IMAGE=quay.io/ascend/vllm-omni:v0.29.0-a5
     docker run --rm \
         --name vllm-omni-a5 \
         --shm-size=64g \
@@ -175,7 +175,7 @@ Here's an example deployment command that has been verified on 4 x NPUs:
 === "310P (Experimental)"
 
     ```bash
-    export IMAGE=quay.io/ascend/vllm-omni:v0.28.0-310p
+    export IMAGE=quay.io/ascend/vllm-omni:v0.29.0-310p
     docker run --rm \
         --name vllm-omni-310p \
         --shm-size=64g \
diff --git a/vllm_omni/platforms/npu/worker/npu_ar_model_runner.py b/vllm_omni/platforms/npu/worker/npu_ar_model_runner.py
index 734178e01..7bc6c21c6 100644
--- a/vllm_omni/platforms/npu/worker/npu_ar_model_runner.py
+++ b/vllm_omni/platforms/npu/worker/npu_ar_model_runner.py
@@ -334,6 +334,8 @@ class NPUARModelRunner(OmniNPUModelRunner, OmniConnectorModelRunnerMixin, Duplex
         if not isinstance(info, dict):
             return None
         val = info.get("omni_final_stage_id")
+        if val is None:
+            return None
         try:
             return int(val)
         except (TypeError, ValueError):
@@ -512,7 +514,7 @@ class NPUARModelRunner(OmniNPUModelRunner, OmniConnectorModelRunnerMixin, Duplex
                         # returns True. before returning early here we call
                         # dummy run to ensure coordinate_batch_across_dp
                         # is called into to avoid out of sync issues.
-                        self._dummy_run(1)
+                        self._dummy_run(1, skip_gdn_state_update=True)
 
                     kv_ids = self.kv_extracted_req_ids
                     self.kv_extracted_req_ids = None
@@ -1035,6 +1037,8 @@ class NPUARModelRunner(OmniNPUModelRunner, OmniConnectorModelRunnerMixin, Duplex
                     propose_draft_token_ids(sampler_output.sampled_token_ids)
 
         (
+            num_nans_in_logits,
+            num_nans_device,
             logprobs_lists,
             valid_sampled_token_ids,
             prompt_logprobs_dict,
@@ -1278,6 +1282,7 @@ class NPUARModelRunner(OmniNPUModelRunner, OmniConnectorModelRunnerMixin, Duplex
             kv_connector_output=kv_connector_output,
             ec_connector_output=ec_connector_output if self.supports_mm_inputs else None,
             cudagraph_stats=cudagraph_stats,
+            num_nans_in_logits=num_nans_in_logits,
         )
         model_runner_output.kv_extracted_req_ids = kv_extracted_req_ids
         model_runner_output.routed_experts = routed_experts_lists
@@ -1318,6 +1323,7 @@ class NPUARModelRunner(OmniNPUModelRunner, OmniConnectorModelRunnerMixin, Duplex
             invalid_req_indices=invalid_req_indices,
             async_output_copy_stream=self.async_output_copy_stream,
             vocab_size=self.input_batch.vocab_size,
+            num_nans=num_nans_device,
         )
         self.input_batch.set_async_sampled_token_ids(
             async_output.sampled_token_ids_cpu,
diff --git a/vllm_omni/platforms/npu/worker/npu_generation_model_runner.py b/vllm_omni/platforms/npu/worker/npu_generation_model_runner.py
index fc26882cd..59dfc88a3 100644
--- a/vllm_omni/platforms/npu/worker/npu_generation_model_runner.py
+++ b/vllm_omni/platforms/npu/worker/npu_generation_model_runner.py
@@ -166,7 +166,7 @@ class NPUGenerationModelRunner(OmniNPUModelRunner, OmniConnectorModelRunnerMixin
                         # returns True. before returning early here we call
                         # dummy run to ensure coordinate_batch_across_dp
                         # is called into to avoid out of sync issues.
-                        self._dummy_run(1)
+                        self._dummy_run(1, skip_gdn_state_update=True)
                     if not has_kv_transfer_group():
                         # Return empty ModelRunnerOutput if no work to do.
                         return self.attach_omni_connector_output(EMPTY_MODEL_RUNNER_OUTPUT)
@@ -636,6 +636,7 @@ class NPUGenerationModelRunner(OmniNPUModelRunner, OmniConnectorModelRunnerMixin
         profile_seq_lens: int | None = None,
         profile_cpp: bool = False,
         randomize_inputs: bool = False,
+        skip_gdn_state_update: bool = False,
     ) -> tuple[torch.Tensor, torch.Tensor]:
         # only support eager mode and piecewise graph now
         assert cudagraph_runtime_mode is None or cudagraph_runtime_mode.is_valid_runtime_mode()
@@ -768,7 +769,10 @@ class NPUGenerationModelRunner(OmniNPUModelRunner, OmniConnectorModelRunnerMixin
                 self.query_start_loc.np[1 : num_reqs_padded + 1] = cum_num_tokens
                 self.query_start_loc.copy_to_gpu()
                 if self._has_gdn:
-                    self.gdn_query_start_loc.np[1 : num_reqs_padded + 1] = cum_num_tokens
+                    if skip_gdn_state_update:
+                        self.gdn_query_start_loc.np.fill(0)
+                    else:
+                        self.gdn_query_start_loc.np[1 : num_reqs_padded + 1] = cum_num_tokens
                     self.gdn_query_start_loc.copy_to_gpu()
 
                 if not profile_cpp:
@@ -786,6 +790,12 @@ class NPUGenerationModelRunner(OmniNPUModelRunner, OmniConnectorModelRunnerMixin
                 # rows as well so device-side metadata does not see stale block ids.
                 self.input_batch.block_table.commit_block_table(num_reqs_padded)
 
+                # Invalidate slots before backends copy metadata for dummy execution.
+                if not is_graph_capturing:
+                    for kv_cache_gid in range(len(self.kv_cache_config.kv_cache_groups)):
+                        blk_table = self.input_batch.block_table[kv_cache_gid]
+                        blk_table.slot_mapping.gpu.fill_(-1)
+
                 pad_attn = cudagraph_runtime_mode == CUDAGraphMode.FULL
                 attn_metadata, _ = self._build_attention_metadata(
                     num_tokens=num_tokens_unpadded,
@@ -796,11 +806,10 @@ class NPUGenerationModelRunner(OmniNPUModelRunner, OmniConnectorModelRunnerMixin
                     ubatch_slices=ubatch_slices_padded if pad_attn else ubatch_slices,
                     for_cudagraph_capture=is_graph_capturing,
                     num_scheduled_tokens_np=num_scheduled_tokens,
+                    cudagraph_runtime_mode=cudagraph_runtime_mode,
+                    batch_descriptor=batch_desc,
+                    skip_gdn_state_update=skip_gdn_state_update,
                 )
-                if not is_graph_capturing:
-                    for kv_cache_gid in range(len(self.kv_cache_config.kv_cache_groups)):
-                        blk_table = self.input_batch.block_table[kv_cache_gid]
-                        blk_table.slot_mapping.gpu.fill_(-1)
 
         with self.maybe_dummy_run_with_lora(
             self.lora_config,
diff --git a/vllm_omni/platforms/npu/worker/npu_model_runner.py b/vllm_omni/platforms/npu/worker/npu_model_runner.py
index 8f0369808..bd7c7dce4 100644
--- a/vllm_omni/platforms/npu/worker/npu_model_runner.py
+++ b/vllm_omni/platforms/npu/worker/npu_model_runner.py
@@ -1,5 +1,5 @@
 # SPDX-License-Identifier: Apache-2.0
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM-Omni project
 
 from functools import partial
 from typing import Any
@@ -126,6 +126,7 @@ class OmniNPUModelRunner(OmniGPUModelRunner, NPUModelRunner):
         profile_seq_lens: int | None = None,
         profile_cpp: bool = False,
         randomize_inputs: bool = False,
+        skip_gdn_state_update: bool = False,
     ) -> tuple[torch.Tensor, torch.Tensor]:
         # only support eager mode and piecewise graph now
         assert cudagraph_runtime_mode is None or cudagraph_runtime_mode.is_valid_runtime_mode()
@@ -257,7 +258,10 @@ class OmniNPUModelRunner(OmniGPUModelRunner, NPUModelRunner):
                 self.query_start_loc.np[1 : num_reqs_padded + 1] = cum_num_tokens
                 self.query_start_loc.copy_to_gpu()
                 if self._has_gdn:
-                    self.gdn_query_start_loc.np[1 : num_reqs_padded + 1] = cum_num_tokens
+                    if skip_gdn_state_update:
+                        self.gdn_query_start_loc.np.fill(0)
+                    else:
+                        self.gdn_query_start_loc.np[1 : num_reqs_padded + 1] = cum_num_tokens
                     self.gdn_query_start_loc.copy_to_gpu()
 
                 if not profile_cpp:
@@ -275,6 +279,12 @@ class OmniNPUModelRunner(OmniGPUModelRunner, NPUModelRunner):
                 # rows as well so device-side metadata does not see stale block ids.
                 self.input_batch.block_table.commit_block_table(num_reqs_padded)
 
+                # Invalidate slots before backends copy metadata for dummy execution.
+                if not is_graph_capturing:
+                    for kv_cache_gid in range(len(self.kv_cache_config.kv_cache_groups)):
+                        blk_table = self.input_batch.block_table[kv_cache_gid]
+                        blk_table.slot_mapping.gpu.fill_(-1)
+
                 pad_attn = cudagraph_runtime_mode == CUDAGraphMode.FULL
                 attn_metadata, _ = self._build_attention_metadata(
                     num_tokens=num_tokens_unpadded,
@@ -285,11 +295,10 @@ class OmniNPUModelRunner(OmniGPUModelRunner, NPUModelRunner):
                     ubatch_slices=ubatch_slices_padded if pad_attn else ubatch_slices,
                     for_cudagraph_capture=is_graph_capturing,
                     num_scheduled_tokens_np=num_scheduled_tokens,
+                    cudagraph_runtime_mode=cudagraph_runtime_mode,
+                    batch_descriptor=batch_desc,
+                    skip_gdn_state_update=skip_gdn_state_update,
                 )
-                if not is_graph_capturing:
-                    for kv_cache_gid in range(len(self.kv_cache_config.kv_cache_groups)):
-                        blk_table = self.input_batch.block_table[kv_cache_gid]
-                        blk_table.slot_mapping.gpu.fill_(-1)
 
         with self.maybe_dummy_run_with_lora(
             self.lora_config,