Remove VLLM_USE_V1 (#4086)

Drop VLLM_USE_V1 usage.  This env has been removed from vLLM already.

- vLLM version: v0.11.0
- vLLM main:
83f478bb19

Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>

.github/workflows/nightly_benchmarks.yaml

@@ -46,14 +46,13 @@ jobs:
   test:
     if: ${{ contains(github.event.pull_request.labels.*.name, 'performance-test') && contains(github.event.pull_request.labels.*.name, 'ready-for-test') || github.event_name == 'schedule' || github.event_name == 'workflow_dispatch' }}
 
-    name: Benchmarks/vLLM=${{ matrix.vllm_branch }}, vLLM-Ascend=${{ matrix.vllm_ascend_branch }}, use_v1=${{ matrix.vllm_use_v1 }}
+    name: Benchmarks/vLLM=${{ matrix.vllm_branch }}, vLLM-Ascend=${{ matrix.vllm_ascend_branch }}
     runs-on: 'linux-arm64-npu-static-8'
     strategy:
       matrix:
         include:
           - vllm_branch: v0.11.0
             vllm_ascend_branch: main
-            vllm_use_v1: 1
       max-parallel: 1
     container:
       image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.3.rc1-910b-ubuntu22.04-py3.11
@@ -73,7 +72,6 @@ jobs:
         VLLM_USE_MODELSCOPE: True
         ES_OM_DOMAIN: ${{ secrets.ES_OM_DOMAIN }}
         ES_OM_AUTHORIZATION: ${{ secrets.ES_OM_AUTHORIZATION }}
-        VLLM_USE_V1: ${{ matrix.vllm_use_v1 }}
     steps:
       - name: Check npu and CANN info
         run: |
@@ -200,7 +198,6 @@ jobs:
             --created_at "$commit_time_no_tz" \
             --res_dir ./benchmarks/results \
             --error "$ERROR_MSG" \
-            --extra_feat '{"VLLM_USE_V1": "${{ matrix.vllm_use_v1 }}"}'
             rm -rf ./benchmarks/results
             cd -
           done < commit_log.txt

docs/source/tutorials/multi_node.md

@@ -160,7 +160,6 @@ export TP_SOCKET_IFNAME=$nic_name
 export HCCL_SOCKET_IFNAME=$nic_name
 export OMP_PROC_BIND=false
 export OMP_NUM_THREADS=100
-export VLLM_USE_V1=1
 export HCCL_BUFFSIZE=1024
 
 vllm serve vllm-ascend/DeepSeek-V3.1-W8A8 \

docs/source/tutorials/multi_node_kimi.md

@@ -70,7 +70,6 @@ export TP_SOCKET_IFNAME=$nic_name
 export HCCL_SOCKET_IFNAME=$nic_name
 export OMP_PROC_BIND=false
 export OMP_NUM_THREADS=100
-export VLLM_USE_V1=1
 export HCCL_BUFFSIZE=1024
 
 # The w8a8 weight can be obtained from https://www.modelscope.cn/models/vllm-ascend/Kimi-K2-Instruct-W8A8
@@ -116,7 +115,6 @@ export TP_SOCKET_IFNAME=$nic_name
 export HCCL_SOCKET_IFNAME=$nic_name
 export OMP_PROC_BIND=false
 export OMP_NUM_THREADS=100
-export VLLM_USE_V1=1
 export HCCL_BUFFSIZE=1024
 
 vllm serve /home/cache/weights/Kimi-K2-Instruct-W8A8 \

docs/source/tutorials/multi_node_pd_disaggregation_llmdatadist.md

@@ -104,7 +104,6 @@ export HCCL_SOCKET_IFNAME="eth0"
 export DISAGGREGATED_PREFILL_RANK_TABLE_PATH="/path/to/your/generated/ranktable.json"
 export OMP_PROC_BIND=false
 export OMP_NUM_THREADS=10
-export VLLM_USE_V1=1
 
 vllm serve /model/Qwen3-30B-A3B  \
   --host 0.0.0.0 \
@@ -144,7 +143,6 @@ export HCCL_SOCKET_IFNAME="eth0"
 export DISAGGREGATED_PREFILL_RANK_TABLE_PATH="/path/to/your/generated/ranktable.json"
 export OMP_PROC_BIND=false
 export OMP_NUM_THREADS=10
-export VLLM_USE_V1=1
 
 vllm serve /model/Qwen3-30B-A3B  \
   --host 0.0.0.0 \
@@ -183,7 +181,6 @@ export HCCL_SOCKET_IFNAME="eth0"
 export DISAGGREGATED_PREFILL_RANK_TABLE_PATH="/path/to/your/generated/ranktable.json"
 export OMP_PROC_BIND=false
 export OMP_NUM_THREADS=10
-export VLLM_USE_V1=1
 
 vllm serve /model/Qwen3-30B-A3B  \
   --host 0.0.0.0 \

docs/source/tutorials/multi_node_pd_disaggregation_mooncake.md

@@ -107,7 +107,6 @@ export HCCL_IF_IP=192.0.0.1
 export GLOO_SOCKET_IFNAME="eth0"  # network card name
 export TP_SOCKET_IFNAME="eth0"
 export HCCL_SOCKET_IFNAME="eth0"
-export VLLM_USE_V1=1
 export HCCL_BUFFSIZE=1024
 export OMP_PROC_BIND=false
 export OMP_NUM_THREADS=10
@@ -162,7 +161,6 @@ export HCCL_IF_IP=192.0.0.2
 export GLOO_SOCKET_IFNAME="eth0"  # network card name
 export TP_SOCKET_IFNAME="eth0"
 export HCCL_SOCKET_IFNAME="eth0"
-export VLLM_USE_V1=1
 export HCCL_BUFFSIZE=1024
 export OMP_PROC_BIND=false
 export OMP_NUM_THREADS=10
@@ -217,7 +215,6 @@ export HCCL_IF_IP=192.0.0.3
 export GLOO_SOCKET_IFNAME="eth0"  # network card name
 export TP_SOCKET_IFNAME="eth0"
 export HCCL_SOCKET_IFNAME="eth0"
-export VLLM_USE_V1=1
 export HCCL_BUFFSIZE=2048
 export OMP_PROC_BIND=false
 export OMP_NUM_THREADS=10
@@ -274,7 +271,6 @@ export HCCL_IF_IP=192.0.0.4
 export GLOO_SOCKET_IFNAME="eth0"  # network card name
 export TP_SOCKET_IFNAME="eth0"
 export HCCL_SOCKET_IFNAME="eth0"
-export VLLM_USE_V1=1
 export HCCL_BUFFSIZE=2048
 export OMP_PROC_BIND=false
 export OMP_NUM_THREADS=10
@@ -338,7 +334,6 @@ export HCCL_IF_IP=192.0.0.1
 export GLOO_SOCKET_IFNAME="eth0"  # network card name
 export TP_SOCKET_IFNAME="eth0"
 export HCCL_SOCKET_IFNAME="eth0"
-export VLLM_USE_V1=1
 export HCCL_BUFFSIZE=1024
 export OMP_PROC_BIND=false
 export OMP_NUM_THREADS=10
@@ -393,7 +388,6 @@ export HCCL_IF_IP=192.0.0.2
 export GLOO_SOCKET_IFNAME="eth0"  # network card name
 export TP_SOCKET_IFNAME="eth0"
 export HCCL_SOCKET_IFNAME="eth0"
-export VLLM_USE_V1=1
 export HCCL_BUFFSIZE=1024
 export OMP_PROC_BIND=false
 export OMP_NUM_THREADS=10
@@ -448,7 +442,6 @@ export HCCL_IF_IP=192.0.0.3
 export GLOO_SOCKET_IFNAME="eth0"  # network card name
 export TP_SOCKET_IFNAME="eth0"
 export HCCL_SOCKET_IFNAME="eth0"
-export VLLM_USE_V1=1
 export HCCL_BUFFSIZE=2048
 export OMP_PROC_BIND=false
 export OMP_NUM_THREADS=10
@@ -505,7 +498,6 @@ export HCCL_IF_IP=192.0.0.4
 export GLOO_SOCKET_IFNAME="eth0"  # network card name
 export TP_SOCKET_IFNAME="eth0"
 export HCCL_SOCKET_IFNAME="eth0"
-export VLLM_USE_V1=1
 export HCCL_BUFFSIZE=2048
 export OMP_PROC_BIND=false
 export OMP_NUM_THREADS=10

docs/source/tutorials/multi_node_qwen3vl.md

@@ -70,7 +70,6 @@ export TP_SOCKET_IFNAME=$nic_name
 export HCCL_SOCKET_IFNAME=$nic_name
 export OMP_PROC_BIND=false
 export OMP_NUM_THREADS=100
-export VLLM_USE_V1=1
 export HCCL_BUFFSIZE=1024
 
 vllm serve Qwen/Qwen3-VL-235B-A22B-Instruct \
@@ -112,7 +111,6 @@ export TP_SOCKET_IFNAME=$nic_name
 export HCCL_SOCKET_IFNAME=$nic_name
 export OMP_PROC_BIND=false
 export OMP_NUM_THREADS=100
-export VLLM_USE_V1=1
 export HCCL_BUFFSIZE=1024
 
 vllm serve Qwen/Qwen3-VL-235B-A22B-Instruct \

docs/source/user_guide/feature_guide/kv_pool_mooncake.md

@@ -82,7 +82,6 @@ The content of the multi_producer.sh script:
 export LD_LIBRARY_PATH=/usr/local/Ascend/ascend-toolkit/latest/python/site-packages:$LD_LIBRARY_PATH
 export PYTHONPATH=$PYTHONPATH:/xxxxx/vllm
 export MOONCAKE_CONFIG_PATH="/xxxxxx/mooncake.json"
-export VLLM_USE_V1=1
 export ASCEND_RT_VISIBLE_DEVICES=0,1,2,3
 export ACL_OP_INIT_MODE=1
 export ASCEND_BUFFER_POOL=4:8
@@ -145,7 +144,6 @@ The content of multi_consumer.sh:
 export LD_LIBRARY_PATH=/usr/local/Ascend/ascend-toolkit/latest/python/site-packages:$LD_LIBRARY_PATH
 export PYTHONPATH=$PYTHONPATH:/xxxxx/vllm
 export MOONCAKE_CONFIG_PATH="/xxxxx/mooncake.json"
-export VLLM_USE_V1=1
 export ASCEND_RT_VISIBLE_DEVICES=4,5,6,7
 export ACL_OP_INIT_MODE=1
 export ASCEND_BUFFER_POOL=4:8
@@ -246,7 +244,6 @@ Content of mixed_department.sh:
 export LD_LIBRARY_PATH=/usr/local/Ascend/ascend-toolkit/latest/python/site-packages:$LD_LIBRARY_PATH
 export PYTHONPATH=$PYTHONPATH:/xxxxx/vllm
 export MOONCAKE_CONFIG_PATH="/xxxxxx/mooncake.json"
-export VLLM_USE_V1=1
 export ASCEND_RT_VISIBLE_DEVICES=0,1,2,3
 export ACL_OP_INIT_MODE=1
 export ASCEND_BUFFER_POOL=4:8

examples/disaggregated_prefill_v1/README.md

@@ -41,7 +41,6 @@ export HCCL_SOCKET_IFNAME="eth0"
 export DISAGGREGATED_PREFILL_RANK_TABLE_PATH=/vllm-workspace/vllm-ascend/examples/disaggregated_prefill_v1/ranktable.json
 export OMP_PROC_BIND=false
 export OMP_NUM_THREADS=100
-export VLLM_USE_V1=1
 export VLLM_ASCEND_LLMDD_RPC_PORT=5559
 
 vllm serve /models/deepseek_r1_w8a8 \
@@ -82,7 +81,6 @@ export HCCL_SOCKET_IFNAME="eth0"
 export DISAGGREGATED_PREFILL_RANK_TABLE_PATH=/vllm-workspace/vllm-ascend/examples/disaggregated_prefill_v1/ranktable.json
 export OMP_PROC_BIND=false
 export OMP_NUM_THREADS=100
-export VLLM_USE_V1=1
 export VLLM_ASCEND_LLMDD_RPC_PORT=5659
 
 vllm serve /models/deepseek_r1_w8a8 \
@@ -126,7 +124,6 @@ export HCCL_SOCKET_IFNAME="eth0"
 export DISAGGREGATED_PREFILL_RANK_TABLE_PATH=/vllm-workspace/vllm-ascend/examples/disaggregated_prefill_v1/ranktable.json
 export OMP_PROC_BIND=false
 export OMP_NUM_THREADS=100
-export VLLM_USE_V1=1
 export VLLM_ASCEND_LLMDD_RPC_PORT=5759
 
 vllm serve /models/deepseek_r1_w8a8 \
@@ -168,7 +165,6 @@ export HCCL_SOCKET_IFNAME="eth0"
 export DISAGGREGATED_PREFILL_RANK_TABLE_PATH=/vllm-workspace/vllm-ascend/examples/disaggregated_prefill_v1/ranktable.json
 export OMP_PROC_BIND=false
 export OMP_NUM_THREADS=100
-export VLLM_USE_V1=1
 export VLLM_ASCEND_LLMDD_RPC_PORT=5859
 
 vllm serve /models/deepseek_r1_w8a8 \

examples/disaggregated_prefill_v1/run_server.sh

@@ -7,8 +7,6 @@ export DISAGGREGATED_PREFILL_RANK_TABLE_PATH=path-to-rank-table
 export OMP_PROC_BIND=false
 export OMP_NUM_THREADS=100
 
-export VLLM_USE_V1=1
-
 vllm serve model_path \
   --host 0.0.0.0 \
   --port 20002 \

examples/external_online_dp/run_dp_template.sh

@@ -11,8 +11,6 @@ export HCCL_DETERMINISTIC=True
 export HCCL_BUFFSIZE=1024
 export TASK_QUEUE_ENABLE=1
 
-export VLLM_USE_V1=1
-
 export ASCEND_RT_VISIBLE_DEVICES=$1
 
 vllm serve model_path \

examples/run_dp_server.sh

@@ -7,7 +7,6 @@ export HCCL_SOCKET_IFNAME="eth0"
 export OMP_PROC_BIND=false
 export OMP_NUM_THREADS=100
 
-export VLLM_USE_V1=1
 export VLLM_USE_MODELSCOPE=true
 
 export ASCEND_LAUNCH_BLOCKING=0

tests/e2e/nightly/features/test_prefix_cache_deepseek_r1_0528_w8a8.py

@@ -72,7 +72,6 @@ async def test_models(model: str) -> None:
         "OMP_PROC_BIND": "false",
         "HCCL_BUFFSIZE": "1024",
         "PYTORCH_NPU_ALLOC_CONF": "expandable_segments:True",
-        "VLLM_USE_V1": "1"
     }
     additional_config = {
         "ascend_scheduler_config": {

tests/e2e/nightly/models/test_qwq_32b.py

@@ -73,7 +73,6 @@ async def test_models(model: str, mode: str, tp_size: int) -> None:
     env_dict = {
         "TASK_QUEUE_ENABLE": "1",
         "OMP_PROC_BIND": "false",
-        "VLLM_USE_V1": "1",
         "HCCL_OP_EXPANSION_MODE": "AIV",
         "VLLM_ASCEND_ENABLE_FLASHCOMM": "1",
         "VLLM_ASCEND_ENABLE_TOPK_OPTIMIZE": "1",

tests/e2e/pd_disaggreate/run_edge_case_test.sh

@@ -2,7 +2,6 @@
 export LCCL_DETERMINISTIC=1
 export HCCL_DETERMINISTIC=true
 export CLOSE_MATMUL_K_SHIFT=1
-export VLLM_USE_V1=1
 
 set -xe
 

tests/e2e/singlecard/test_aclgraph_mem.py

@@ -30,8 +30,6 @@ from vllm_ascend.worker.model_runner_v1 import NPUModelRunner
 MODELS = ["Qwen/Qwen3-0.6B", "vllm-ascend/DeepSeek-V2-Lite-W8A8"]
 
 
-@pytest.mark.skipif(os.getenv("VLLM_USE_V1") == "0",
-                    reason="aclgraph only support on v1")
 @pytest.mark.parametrize("model", MODELS)
 @pytest.mark.parametrize("max_tokens", [4])
 @patch.dict(os.environ, {"VLLM_ASCEND_FLASHCOMM2_PARALLEL_SIZE": "0"})

tests/ut/kv_connector/utils.py

@@ -27,7 +27,6 @@ else:
     from vllm.utils.hashing import sha256
 
 EOS_TOKEN_ID = 50256
-os.environ["VLLM_USE_V1"] = "1"
 
 
 def assert_scheduler_empty(scheduler: Scheduler):

vllm_ascend/models/qwen3_next.py

@@ -9,7 +9,6 @@ import torch
 from einops import rearrange
 from torch import nn
 from transformers.activations import ACT2FN
-from vllm import envs
 from vllm.attention import AttentionBackend, AttentionMetadata
 from vllm.compilation.decorators import support_torch_compile
 from vllm.config import (CacheConfig, ModelConfig, SpeculativeConfig,
@@ -668,7 +667,6 @@ class CustomQwen3NextForCausalLM(Qwen3NextForCausalLM):
         scheduler_config = vllm_config.scheduler_config
         assert not cache_config.enable_prefix_caching, \
             "Qwen3Next currently does not support prefix caching"
-        assert envs.VLLM_USE_V1, "Qwen3Next requires VLLM_USE_V1"
         self.quant_config = vllm_config.quant_config
         self.config = config
         self.scheduler_config = scheduler_config

vllm_ascend/patch/platform/patch_config.py

@@ -1,6 +1,5 @@
 import ast
 
-import vllm.envs as envs
 from vllm.config.speculative import SpeculativeConfig
 from vllm.logger import logger
 
@@ -163,11 +162,6 @@ def __post_init__(self):
 
             # Replace hf_config for EAGLE draft_model
             if self.method in ("eagle", "eagle3"):
-                if self.enable_chunked_prefill and not envs.VLLM_USE_V1:
-                    raise ValueError(
-                        "Chunked prefill and EAGLE are not compatible "
-                        "when using V0.")
-
                 from vllm.transformers_utils.configs import SpeculatorsConfig
                 from vllm.transformers_utils.configs.eagle import EAGLEConfig
 

vllm_ascend/platform.py

@@ -20,7 +20,6 @@ import os
 from typing import TYPE_CHECKING, Optional, Tuple
 
 import torch
-import vllm.envs as envs_vllm
 from vllm.logger import logger
 from vllm.platforms import Platform, PlatformEnum
 
@@ -117,8 +116,6 @@ class NPUPlatform(Platform):
 
     @classmethod
     def check_and_update_config(cls, vllm_config: VllmConfig) -> None:
-        if not envs_vllm.VLLM_USE_V1:
-            raise ValueError("vLLM Ascend does not support V0 engine.")
         # initialize ascend config from vllm additional_config
         ascend_config = init_ascend_config(vllm_config)
 

vllm_ascend/torchair/models/qwen2.py

@@ -21,7 +21,6 @@ from typing import Any, List, Optional, Union
 import torch
 import torch.nn.functional as F
 import vllm
-import vllm.envs as envs
 from torch import nn
 from transformers import Qwen2Config
 from vllm.attention import AttentionMetadata, AttentionType
@@ -112,11 +111,8 @@ class CustomQwen2Attention(Qwen2Attention):
                                    is_prefill=False,
                                    is_qwen_torchair=True)
             forward_kwargs = {}
-            if envs.VLLM_USE_V1:
             output_shape = q.shape
-                output = torch.empty(output_shape,
+            output = torch.empty(output_shape, dtype=q.dtype, device=q.device)
-                                     dtype=q.dtype,
-                                     device=q.device)
             forward_kwargs['output'] = output
 
             attn_output = self.attn.impl.forward(self.attn,

vllm_ascend/torchair/models/qwen3_moe.py

@@ -19,7 +19,6 @@
 from typing import Any, List, Optional, Union
 
 import torch
-import vllm.envs as envs
 from torch import nn
 from transformers import PretrainedConfig
 from vllm.attention import Attention, AttentionMetadata
@@ -244,11 +243,8 @@ class CustomQwen3MoeAttention(Qwen3MoeAttention):
                                    is_prefill=False,
                                    is_qwen_torchair=True)
             forward_kwargs = {}
-            if envs.VLLM_USE_V1:
             output_shape = q.shape
-                output = torch.empty(output_shape,
+            output = torch.empty(output_shape, dtype=q.dtype, device=q.device)
-                                     dtype=q.dtype,
-                                     device=q.device)
             forward_kwargs['output'] = output
 
             attn_output = self.attn.impl.forward(self.attn,