From 81c979091e13053db8aae25f116ba6faa8410731 Mon Sep 17 00:00:00 2001
From: wucong25 <412916467@qq.com>
Date: Wed, 24 Jun 2026 17:49:00 +0800
Subject: [PATCH 1/4] test ci

---
 verl/checkpoint_engine/hccl_checkpoint_engine.py | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/verl/checkpoint_engine/hccl_checkpoint_engine.py b/verl/checkpoint_engine/hccl_checkpoint_engine.py
index df777cfae5e..73176ff82f0 100644
--- a/verl/checkpoint_engine/hccl_checkpoint_engine.py
+++ b/verl/checkpoint_engine/hccl_checkpoint_engine.py
@@ -149,7 +149,8 @@ def finalize(self):
 
         self.send_buf = None
         self.recv_buf = None
-        torch.npu.empty_cache()
+        from verl.utils.device import get_torch_device
+        get_torch_device().empty_cache()
 
     @classmethod
     def build_topology(cls, actor_wg_world_size: int, rollout_world_size: int, metadata: list[dict]):

From b66e87d9bffe626a6ae0f11f7e426cc530ab3fbf Mon Sep 17 00:00:00 2001
From: wucong25 <412916467@qq.com>
Date: Mon, 29 Jun 2026 14:42:41 +0800
Subject: [PATCH 2/4] Update test_multi_modal.py

---
 tests/experimental/agent_loop/test_multi_modal.py | 1 +
 1 file changed, 1 insertion(+)

diff --git a/tests/experimental/agent_loop/test_multi_modal.py b/tests/experimental/agent_loop/test_multi_modal.py
index 35f4600b524..556f3d30ce5 100644
--- a/tests/experimental/agent_loop/test_multi_modal.py
+++ b/tests/experimental/agent_loop/test_multi_modal.py
@@ -66,6 +66,7 @@ def init_config() -> DictConfig:
     config.actor_rollout_ref.rollout.prompt_length = 10240
     config.actor_rollout_ref.rollout.response_length = 4096
     config.actor_rollout_ref.rollout.n = 4
+    config.actor_rollout_ref.rollout.seed = 0
     config.actor_rollout_ref.rollout.agent.num_workers = 2
     config.actor_rollout_ref.rollout.skip_tokenizer_init = True
 

From da2c0780cb6554381bcd3a0b63f5a273c369e213 Mon Sep 17 00:00:00 2001
From: wucong25 <412916467@qq.com>
Date: Mon, 29 Jun 2026 14:44:12 +0800
Subject: [PATCH 3/4] Update hccl_checkpoint_engine.py

---
 verl/checkpoint_engine/hccl_checkpoint_engine.py | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/verl/checkpoint_engine/hccl_checkpoint_engine.py b/verl/checkpoint_engine/hccl_checkpoint_engine.py
index 73176ff82f0..df777cfae5e 100644
--- a/verl/checkpoint_engine/hccl_checkpoint_engine.py
+++ b/verl/checkpoint_engine/hccl_checkpoint_engine.py
@@ -149,8 +149,7 @@ def finalize(self):
 
         self.send_buf = None
         self.recv_buf = None
-        from verl.utils.device import get_torch_device
-        get_torch_device().empty_cache()
+        torch.npu.empty_cache()
 
     @classmethod
     def build_topology(cls, actor_wg_world_size: int, rollout_world_size: int, metadata: list[dict]):

From 327e0b39657b777347a0ba11b24b1e0e7d3c822f Mon Sep 17 00:00:00 2001
From: wucong25 <412916467@qq.com>
Date: Mon, 29 Jun 2026 14:47:30 +0800
Subject: [PATCH 4/4] Update vllm_ascend.yml

---
 .github/workflows/vllm_ascend.yml | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/.github/workflows/vllm_ascend.yml b/.github/workflows/vllm_ascend.yml
index 7c485862e09..eaf2a210baf 100644
--- a/.github/workflows/vllm_ascend.yml
+++ b/.github/workflows/vllm_ascend.yml
@@ -110,7 +110,7 @@ jobs:
         run: |
           ray stop --force
           python3 examples/data_preprocess/gsm8k.py --local_dataset_path ${HOME}/models/hf_data/gsm8k
-      - name: Test the latest vLLM Rollout async with agent loop
+      - name: Test the latest vLLM Rollout async with agent loop with seed 0(non deter)
         run: |
           export HCCL_HOST_SOCKET_PORT_RANGE=auto
           export HCCL_NPU_SOCKET_PORT_RANGE=auto