Merge branch 'vllm-project:main' into main

Author: florenceCH

.github/workflows/format_pr_body.yaml

@@ -36,7 +36,7 @@ jobs:
 
       - name: Get vLLM version
         run: |
-          VLLM_COMMIT=9607d5eb449711b349d4c2bee0a9c94afcc7ed14
+          VLLM_COMMIT=f225ea7dd98e9f29752e5c032cd4a8ee1d712f16
           echo "VLLM_COMMIT=https://github.com/vllm-project/vllm/commit/$VLLM_COMMIT" >> $GITHUB_ENV
 
       - name: Checkout repository

.github/workflows/vllm_ascend_test.yaml

@@ -42,7 +42,7 @@ jobs:
   lint:
     uses: ./.github/workflows/pre-commit.yml
     with:
-      vllm: 9607d5eb449711b349d4c2bee0a9c94afcc7ed14
+      vllm: f225ea7dd98e9f29752e5c032cd4a8ee1d712f16
 
   changes:
     runs-on: ubuntu-latest
@@ -83,7 +83,7 @@ jobs:
         VLLM_USE_MODELSCOPE: True
     strategy:
       matrix:
-        vllm_version: [9607d5eb449711b349d4c2bee0a9c94afcc7ed14, v0.10.2]
+        vllm_version: [f225ea7dd98e9f29752e5c032cd4a8ee1d712f16, v0.10.2]
     steps:
       - name: Install packages
         run: |
@@ -138,7 +138,7 @@ jobs:
     name: e2e-light
     strategy:
       matrix:
-        vllm_version: [9607d5eb449711b349d4c2bee0a9c94afcc7ed14, v0.10.2]
+        vllm_version: [f225ea7dd98e9f29752e5c032cd4a8ee1d712f16, v0.10.2]
     # Note (yikun): If CI resource are limited we can split job into two chain jobs
     needs: [lint, changes]
     # only trigger e2e test after lint passed and the change is e2e related with pull request.

.github/workflows/vllm_ascend_test_full.yaml

@@ -68,7 +68,7 @@ jobs:
     name: e2e-full
     strategy:
       matrix:
-        vllm_version: [9607d5eb449711b349d4c2bee0a9c94afcc7ed14, v0.10.2]
+        vllm_version: [f225ea7dd98e9f29752e5c032cd4a8ee1d712f16, v0.10.2]
     needs: [changes]
     if: ${{ needs.changes.outputs.e2e_tracker == 'true' }}
     uses: ./.github/workflows/_e2e_test.yaml

docs/source/developer_guide/modeling/adding_a_new_model.md

@@ -61,7 +61,6 @@ from torch import nn
 from vllm.attention import Attention
 from vllm.config import VllmConfig
 from vllm.sequence import IntermediateTensors
-from vllm.model_executor.sampling_metadata import SamplingMetadata
 
 class CustomAttention(nn.Module):
     def __init__(self, vllm_config: VllmConfig, prefix: str):

docs/source/tutorials/multi_node_ray.md

@@ -91,7 +91,7 @@ After setting up the containers and installing vllm-ascend on each node, follow
 
 Choose one machine as the head node and the others as worker nodes. Before proceeding, use `ip addr` to check your `nic_name` (network interface name).
 
-Set the `ASCEND_RT_VISIBLE_DEVICES` environment variable to specify the NPU devices to use. For Ray versions above 2.1, also set the `RAY_EXPERIMENTAL_NOSET_ASCEND_RT_VISIBLE_DEVICES` variable to avoid device recognition issues. The `--num-gpus` parameter defines the number of NPUs to be used on each node.
+Set the `ASCEND_RT_VISIBLE_DEVICES` environment variable to specify the NPU devices to use. For Ray versions above 2.1, also set the `RAY_EXPERIMENTAL_NOSET_ASCEND_RT_VISIBLE_DEVICES` variable to avoid device recognition issues.
 
 Below are the commands for the head and worker nodes:
 
@@ -109,7 +109,7 @@ export GLOO_SOCKET_IFNAME={nic_name}
 export TP_SOCKET_IFNAME={nic_name}
 export RAY_EXPERIMENTAL_NOSET_ASCEND_RT_VISIBLE_DEVICES=1
 export ASCEND_RT_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
-ray start --head --num-gpus=8
+ray start --head
 ```
 
 **Worker node**:
@@ -125,20 +125,22 @@ export GLOO_SOCKET_IFNAME={nic_name}
 export TP_SOCKET_IFNAME={nic_name}
 export RAY_EXPERIMENTAL_NOSET_ASCEND_RT_VISIBLE_DEVICES=1
 export ASCEND_RT_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
-ray start --address='{head_node_ip}:6379' --num-gpus=8 --node-ip-address={local_ip}
+ray start --address='{head_node_ip}:6379' --node-ip-address={local_ip}
 ```
 
 Once the cluster is started on multiple nodes, execute `ray status` and `ray list nodes` to verify the Ray cluster's status. You should see the correct number of nodes and NPUs listed.
 
-## Start the Online Inference Service on multinode
-In the container, you can use vLLM as if all NPUs were on a single node. vLLM will utilize NPU resources across all nodes in the Ray cluster. You only need to run the vllm command on one node.
+## Start the Online Inference Service on multinode scenario
+In the container, you can use vLLM as if all NPUs were on a single node. vLLM will utilize NPU resources across all nodes in the Ray cluster.
+
+**You only need to run the vllm command on one node.**
 
 To set up parallelism, the common practice is to set the `tensor-parallel-size` to the number of NPUs per node, and the `pipeline-parallel-size` to the number of nodes.
 
 For example, with 16 NPUs across 2 nodes (8 NPUs per node), set the tensor parallel size to 8 and the pipeline parallel size to 2:
 
 ```shell
-vllm Qwen/Qwen3-235B-A22B \
+vllm serve Qwen/Qwen3-235B-A22B \
   --distributed-executor-backend ray \
   --pipeline-parallel-size 2 \
   --tensor-parallel-size 8 \
@@ -154,7 +156,7 @@ vllm Qwen/Qwen3-235B-A22B \
 Alternatively, if you want to use only tensor parallelism, set the tensor parallel size to the total number of NPUs in the cluster. For example, with 16 NPUs across 2 nodes, set the tensor parallel size to 16:
 
 ```shell
-vllm Qwen/Qwen3-235B-A22B \
+vllm serve Qwen/Qwen3-235B-A22B \
   --distributed-executor-backend ray \
   --tensor-parallel-size 16 \
   --enable-expert-parallel \

examples/offline_disaggregated_prefill_npu.py

@@ -79,7 +79,7 @@ def run_prefill(prefill_done, process_close):
 
 
 def run_decode(prefill_done):
-    os.environ['VLLM_LLMDD_RPC_PORT'] = '6634'
+    os.environ['VLLM_ASCEND_LLMDD_RPC_PORT'] = '6634'
     # ranktable.json needs be generated using gen_ranktable.sh
     # from the examples/disaggregated_prefill_v1 module in the main branch.
     os.environ['DISAGGREGATED_PREFILL_RANK_TABLE_PATH'] = "./ranktable.json"

tests/e2e/model_utils.py

@@ -19,7 +19,12 @@
 
 from typing import Dict, List, Optional, Sequence, Tuple, Union
 
-from vllm.sequence import PromptLogprobs, SampleLogprobs
+from vllm_ascend.utils import vllm_version_is
+
+if vllm_version_is("0.10.2"):
+    from vllm.sequence import PromptLogprobs, SampleLogprobs
+else:
+    from vllm.logprobs import PromptLogprobs, SampleLogprobs
 
 TokensText = Tuple[List[int], str]
 

tests/e2e/pd_disaggreate/run_edge_case_test.sh

@@ -70,7 +70,7 @@ run_tests_for_model() {
   # Start prefill instance
   PREFILL_PORT=8001
 
-  BASE_CMD="ASCEND_RT_VISIBLE_DEVICES=0 VLLM_LLMDD_RPC_PORT=5559 vllm serve $model_name \
+  BASE_CMD="ASCEND_RT_VISIBLE_DEVICES=0 VLLM_ASCEND_LLMDD_RPC_PORT=5559 vllm serve $model_name \
   --port $PREFILL_PORT \
   --seed 1024 \
   --enforce-eager \
@@ -90,7 +90,7 @@ run_tests_for_model() {
   DECODE_PORT=8002
 
   # Build the command with or without model-specific args
-  BASE_CMD="ASCEND_RT_VISIBLE_DEVICES=1 VLLM_LLMDD_RPC_PORT=6000 vllm serve $model_name \
+  BASE_CMD="ASCEND_RT_VISIBLE_DEVICES=1 VLLM_ASCEND_LLMDD_RPC_PORT=6000 vllm serve $model_name \
   --port $DECODE_PORT \
   --seed 1024 \
   --enforce-eager \

tests/ut/attention/test_mla_v1.py

@@ -554,7 +554,11 @@ def test_mla_preprocess(self, magic_npu_fetch):
         self.impl.num_kv_heads = self.impl.num_heads
 
         decode_res, prefill_res = self.impl._mla_preprocess(
-            hidden_states, kv_cache, attn_metadata, need_gather_q_kv=False)
+            "mock_layer",
+            hidden_states,
+            kv_cache,
+            attn_metadata,
+            need_gather_q_kv=False)
 
         self.assertIsNotNone(decode_res)
         self.assertIsNotNone(prefill_res)

tests/ut/core/test_schedule_config.py

@@ -27,7 +27,6 @@ def setUp(self):
             max_model_len=8192,
             is_multimodal_model=False,
             send_delta_data=False,
-            scheduler_delay_factor=0,
         )
 
     def test_initialize_from_config_with_default(self):
@@ -90,21 +89,6 @@ def test_not_implemented_send_delta_data(self):
             str(context.exception),
         )
 
-    def test_not_implemented_delay_factor(self):
-        with self.assertRaises(NotImplementedError) as context:
-            AscendSchedulerConfig.initialize_from_config(
-                self.basic_scheduler_config,
-                AscendSchedulerConfig(
-                    delay_factor=1,
-                    max_num_batched_tokens=2048,
-                    max_model_len=2048,
-                ),
-            )
-        self.assertIn(
-            "currently AscendScheduler doesn't support scheduler_delay_factor",
-            str(context.exception),
-        )
-
     def test_no_override(self):
         ascend_config = AscendSchedulerConfig.initialize_from_config(
             self.basic_scheduler_config, {})