drop ascend scheduler

.github/workflows/_e2e_test.yaml

@@ -91,10 +91,8 @@ jobs:
           pytest -sv tests/e2e/singlecard/test_completion_with_prompt_embeds.py
           pytest -sv tests/e2e/singlecard/test_aclgraph.py
           pytest -sv tests/e2e/singlecard/test_aclgraph_mem.py
-          pytest -sv tests/e2e/singlecard/test_ascend_scheduler.py
           pytest -sv tests/e2e/singlecard/test_bge_model.py
           pytest -sv tests/e2e/singlecard/test_camem.py
-          pytest -sv tests/e2e/singlecard/test_chunked.py
           pytest -sv tests/e2e/singlecard/test_embedding.py
           # pytest -sv tests/e2e/singlecard/test_embedding_aclgraph.py
           pytest -sv tests/e2e/singlecard/test_guided_decoding.py

docs/source/tutorials/DeepSeek-V3.2-Exp.md

@@ -108,7 +108,7 @@ vllm serve vllm-ascend/DeepSeek-V3.2-Exp-W8A8 \
 --trust-remote-code \
 --no-enable-prefix-caching \
 --gpu-memory-utilization 0.92 \
---additional-config '{"ascend_scheduler_config":{"enabled":true},"torchair_graph_config":{"enabled":true,"graph_batch_sizes":[16]}}'
+--additional-config '{"torchair_graph_config":{"enabled":true,"graph_batch_sizes":[16]}}'
 ```
 
 ### Multi-node Deployment
@@ -160,7 +160,7 @@ vllm serve /root/.cache/Modelers_Park/DeepSeek-V3.2-Exp \
 --trust-remote-code \
 --no-enable-prefix-caching \
 --gpu-memory-utilization 0.9 \
---additional-config '{"ascend_scheduler_config":{"enabled":true},"torchair_graph_config":{"enabled":true,"graph_batch_sizes":[16]}}'
+--additional-config '{"torchair_graph_config":{"enabled":true,"graph_batch_sizes":[16]}}'
 ```
 
 **Node 1**
@@ -204,7 +204,7 @@ vllm serve /root/.cache/Modelers_Park/DeepSeek-V3.2-Exp \
 --trust-remote-code \
 --no-enable-prefix-caching \
 --gpu-memory-utilization 0.92 \
---additional-config '{"ascend_scheduler_config":{"enabled":true},"torchair_graph_config":{"enabled":true,"graph_batch_sizes":[16]}}'
+--additional-config '{"torchair_graph_config":{"enabled":true,"graph_batch_sizes":[16]}}'
 ```
 
 ::::
@@ -252,7 +252,7 @@ vllm serve vllm-ascend/DeepSeek-V3.2-Exp-W8A8 \
 --quantization ascend \
 --no-enable-prefix-caching \
 --gpu-memory-utilization 0.9 \
---additional-config '{"ascend_scheduler_config":{"enabled":true},"torchair_graph_config":{"enabled":true,"graph_batch_sizes":[16]}}'
+--additional-config '{"torchair_graph_config":{"enabled":true,"graph_batch_sizes":[16]}}'
 ```
 
 **Node 1**
@@ -299,7 +299,7 @@ vllm serve vllm-ascend/DeepSeek-V3.2-Exp-W8A8 \
 --quantization ascend \
 --no-enable-prefix-caching \
 --gpu-memory-utilization 0.92 \
---additional-config '{"ascend_scheduler_config":{"enabled":true},"torchair_graph_config":{"enabled":true,"graph_batch_sizes":[16]}}'
+--additional-config '{"torchair_graph_config":{"enabled":true,"graph_batch_sizes":[16]}}'
 ```
 
 ::::

docs/source/tutorials/multi_node.md

@@ -137,7 +137,7 @@ vllm serve vllm-ascend/DeepSeek-V3.1-W8A8 \
 --trust-remote-code \
 --no-enable-prefix-caching \
 --gpu-memory-utilization 0.9 \
---additional-config '{"ascend_scheduler_config":{"enabled":true},"torchair_graph_config":{"enabled":true}}'
+--additional-config '{"torchair_graph_config":{"enabled":true}}'
 ```
 
 **Node 1**
@@ -182,7 +182,7 @@ vllm serve vllm-ascend/DeepSeek-V3.1-W8A8 \
 --trust-remote-code \
 --no-enable-prefix-caching \
 --gpu-memory-utilization 0.92 \
---additional-config '{"ascend_scheduler_config":{"enabled":true},"torchair_graph_config":{"enabled":true}}'
+--additional-config '{"torchair_graph_config":{"enabled":true}}'
 ```
 
 The deployment view looks like:

docs/source/tutorials/multi_node_kimi.md

@@ -93,7 +93,7 @@ vllm serve /home/cache/weights/Kimi-K2-Instruct-W8A8 \
 --trust-remote-code \
 --no-enable-prefix-caching \
 --gpu-memory-utilization 0.9 \
---additional-config '{"ascend_scheduler_config":{"enabled":true},"torchair_graph_config":{"enabled":true}}'
+--additional-config '{"torchair_graph_config":{"enabled":true}}'
 ```
 
 **Node 1**
@@ -137,7 +137,7 @@ vllm serve /home/cache/weights/Kimi-K2-Instruct-W8A8 \
 --trust-remote-code \
 --no-enable-prefix-caching \
 --gpu-memory-utilization 0.92 \
---additional-config '{"ascend_scheduler_config":{"enabled":true},"torchair_graph_config":{"enabled":true}}'
+--additional-config '{"torchair_graph_config":{"enabled":true}}'
 ```
 
 The deployment view looks like:

docs/source/tutorials/multi_npu_moge.md

@@ -158,11 +158,6 @@ if __name__ == "__main__":
             'torchair_graph_config': {
             'enabled': True,
             },
-            'ascend_scheduler_config':{
-            'enabled': True,
-            'enable_chunked_prefill' : False,
-            'chunked_prefill_enabled': False
-            },
             })
 
     outputs = llm.generate(prompts, sampling_params)

docs/source/user_guide/configuration/additional_config.md

@@ -27,7 +27,6 @@ The following table lists additional configuration options available in vLLM Asc
 | Name                                | Type | Default | Description                                                                                                                                   |
 |-------------------------------------|------|---------|-----------------------------------------------------------------------------------------------------------------------------------------------|
 | `torchair_graph_config`             | dict | `{}`    | Configuration options for torchair graph mode                                                                                                    |
-| `ascend_scheduler_config`           | dict | `{}`    | Configuration options for ascend scheduler                                                                                                       |
 | `weight_prefetch_config`            | dict | `{}`    | Configuration options for weight prefetch                                                                                                        |
 | `refresh`                           | bool | `false` | Whether to refresh global Ascend configuration content. This is usually used by rlhf or ut/e2e test case.                                      |
 | `expert_map_path`                   | str  | `None`  | When using expert load balancing for an MoE model, an expert map path needs to be passed in.                                                 |
@@ -61,18 +60,6 @@ The details of each configuration option are as follows:
 | `enable_kv_nz`| bool | `False` | Whether to enable KV Cache NZ layout. This option only takes effect on models using MLA (for example, DeepSeek). |
 | `enable_super_kernel` | bool | `False` | Whether to enable super kernel to fuse operators in deepseek moe layers. This option only takes effects on moe models using dynamic w8a8 quantization.|
 
-**ascend_scheduler_config**
-
-| Name | Type | Default | Description |
-| ---- | ---- | ------- | ----------- |
-| `enabled` | bool | `False` | Whether to enable ascend scheduler for V1 engine.|
-| `enable_pd_transfer` | bool | `False` | Whether to enable P-D transfer. When it is enabled, decode is started only when prefill of all requests is done. This option only takes effect on offline inference. |
-| `decode_max_num_seqs` | int | `0` | Whether to change max_num_seqs of decode phase when P-D transfer is enabled. This option only takes effect when enable_pd_transfer is True. |
-| `max_long_partial_prefills` | Union[int, float] | `float('inf')` | The maximum number of prompts longer than long_prefill_token_threshold that will be prefilled concurrently. |
-| `long_prefill_token_threshold` | Union[int, float] | `float('inf')` | a request is considered long if the prompt is longer than this number of tokens. |
-
-ascend_scheduler_config also supports the options from [vllm scheduler config](https://docs.vllm.ai/en/stable/api/vllm/config.html#vllm.config.SchedulerConfig). For example, you can add `enable_chunked_prefill: True` to ascend_scheduler_config as well.
-
 **weight_prefetch_config**
 
 | Name             | Type | Default                                                     | Description                        |
@@ -93,12 +80,6 @@ An example of additional configuration is as follows:
         "graph_batch_sizes_init": False,
         "enable_kv_nz": False
     },
-    "ascend_scheduler_config": {
-        "enabled": True,
-        "enable_chunked_prefill": True,
-        "max_long_partial_prefills": 1,
-        "long_prefill_token_threshold": 4096,
-    },
     "weight_prefetch_config": {
         "enabled": True,
         "prefetch_ratio": {

docs/source/user_guide/feature_guide/graph_mode.md

@@ -45,14 +45,14 @@ import os
 from vllm import LLM
 
 # TorchAirGraph only works without chunked-prefill now
-model = LLM(model="path/to/DeepSeek-R1-0528", additional_config={"torchair_graph_config": {"enabled": True},"ascend_scheduler_config": {"enabled": True}})
+model = LLM(model="path/to/DeepSeek-R1-0528", additional_config={"torchair_graph_config": {"enabled": True}})
 outputs = model.generate("Hello, how are you?")
 ```
 
 Online example:
 
 ```shell
-vllm serve path/to/DeepSeek-R1-0528 --additional-config='{"torchair_graph_config": {"enabled": true},"ascend_scheduler_config": {"enabled": true}}'
+vllm serve path/to/DeepSeek-R1-0528 --additional-config='{"torchair_graph_config": {"enabled": true}}'
 ```
 
 You can find more details about additional configuration [here](../configuration/additional_config.md).

examples/offline_inference_npu_long_seq.py

@@ -42,7 +42,6 @@
         enable_chunked_prefill=False,
         max_num_batched_tokens=2048,
         max_model_len=1024,
-        additional_config={"ascend_scheduler_config": {"enabled": False}},
         max_num_seqs=1,
         block_size=128,
         gpu_memory_utilization=0.9

examples/run_dp_server.sh

@@ -28,4 +28,4 @@ vllm serve Qwen/Qwen1.5-MoE-A2.7B  \
   --gpu-memory-utilization 0.9 \
   --trust-remote-code \
   --enforce-eager \
-  --additional-config '{"ascend_scheduler_config":{"enabled":true},"torchair_graph_config":{"enabled":false, "use_cached_graph":false}}'
+  --additional-config '{"torchair_graph_config":{"enabled":false, "use_cached_graph":false}}'

tests/e2e/310p/test_offline_inference_parallel_310p.py

@@ -24,15 +24,12 @@
 MODELS = [
     "IntervitensInc/pangu-pro-moe-model",
 ]
-# set additional config for ascend scheduler and torchair graph
+# set additional config for torchair graph
 ADDITIONAL_CONFIG = [{
     "additional_config": {
         "torchair_graph_config": {
             "enabled": True
         },
-        "ascend_scheduler_config": {
-            "enabled": True,
-        }
     }
 }]
 

tests/e2e/multicard/test_expert_parallel.py

@@ -15,23 +15,14 @@ def test_e2e_ep_correctness(model_name):
     max_tokens = 5
 
     # FIXME: Really strange that chunked prefill might lead to different results, investigate further
-    with VllmRunner(
-            model_name,
-            tensor_parallel_size=2,
-            additional_config={"ascend_scheduler_config": {
-                "enabled": True
-            }},
-            enforce_eager=False) as vllm_model:
+    with VllmRunner(model_name, tensor_parallel_size=2,
+                    enforce_eager=False) as vllm_model:
         tp_output = vllm_model.generate_greedy(example_prompts, max_tokens)
 
-    with VllmRunner(
-            model_name,
-            tensor_parallel_size=2,
-            enable_expert_parallel=True,
-            additional_config={"ascend_scheduler_config": {
-                "enabled": True
-            }},
-            enforce_eager=False) as vllm_model:
+    with VllmRunner(model_name,
+                    tensor_parallel_size=2,
+                    enable_expert_parallel=True,
+                    enforce_eager=False) as vllm_model:
         ep_output = vllm_model.generate_greedy(example_prompts, max_tokens)
 
     check_outputs_equal(

tests/e2e/multicard/test_fused_moe_allgather_ep.py

@@ -49,13 +49,7 @@ def test_generate_with_allgather():
                     tensor_parallel_size=2,
                     max_model_len=1024,
                     dtype="auto",
-                    enable_expert_parallel=True,
-                    additional_config={
-                        "ascend_scheduler_config": {
-                            "enabled": True,
-                            "chunked_prefill_enabled": False,
-                        },
-                    }) as vllm_model:
+                    enable_expert_parallel=True) as vllm_model:
         vllm_model.generate(example_prompts, sampling_params)
 
 
@@ -76,11 +70,5 @@ def test_generate_with_alltoall():
                     tensor_parallel_size=2,
                     max_model_len=1024,
                     dtype="auto",
-                    enable_expert_parallel=True,
-                    additional_config={
-                        "ascend_scheduler_config": {
-                            "enabled": True,
-                            "chunked_prefill_enabled": False,
-                        },
-                    }) as vllm_model:
+                    enable_expert_parallel=True) as vllm_model:
         vllm_model.generate(example_prompts, sampling_params)

tests/e2e/multicard/test_offline_inference_distributed.py

@@ -82,9 +82,6 @@ def test_models_distributed_DeepSeek_multistream_moe():
                     "enabled": True,
                 },
                 "enable_multistream_moe": True,
-                "ascend_scheduler_config": {
-                    "enabled": True,
-                },
                 "refresh": True,
             },
     ) as vllm_model:
@@ -154,14 +151,9 @@ def test_models_distributed_DeepSeek_W4A8DYNAMIC(model):
             quantization="ascend",
             enforce_eager=True,
             enable_expert_parallel=True,
-            additional_config={
-                "torchair_graph_config": {
-                    "enabled": False,
-                },
-                "ascend_scheduler_config": {
-                    "enabled": True,
-                }
-            },
+            additional_config={"torchair_graph_config": {
+                "enabled": False,
+            }},
     ) as vllm_model:
         vllm_model.generate_greedy(prompts, max_tokens)
 

tests/e2e/multicard/test_prefix_caching.py

@@ -1,6 +1,6 @@
 # SPDX-License-Identifier: Apache-2.0
 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
-"""Compare the with and without prefix caching on V1 scheduler or AscendScheduler."""
+"""Compare the with and without prefix caching on V1 scheduler."""
 
 import pytest
 
@@ -84,67 +84,3 @@ def test_prefix_cache_with_v1_scheduler(model: str, max_tokens: int) -> None:
         name_0="vllm_output",
         name_1="prefix_cache_output",
     )
-
-
-@pytest.mark.skip(reason="Fix me, the accuracy is not correct")
-@pytest.mark.parametrize("model", MODELS)
-@pytest.mark.parametrize("max_tokens", [50])
-def test_prefix_cache_with_ascend_scheduler(model: str,
-                                            max_tokens: int) -> None:
-
-    with VllmRunner(model,
-                    additional_config={
-                        'ascend_scheduler_config': {
-                            'enabled': True,
-                        },
-                    },
-                    enforce_eager=False,
-                    max_model_len=2048,
-                    tensor_parallel_size=2,
-                    gpu_memory_utilization=0.7) as vllm_model:
-        vllm_output = vllm_model.generate_greedy(INPUT_PROMPTS, max_tokens)
-
-    with VllmRunner(model,
-                    additional_config={
-                        'ascend_scheduler_config': {
-                            'enabled': True,
-                            'enable_prefix_caching': True,
-                        },
-                    },
-                    enforce_eager=False,
-                    max_model_len=2048,
-                    tensor_parallel_size=2,
-                    gpu_memory_utilization=0.7) as vllm_model:
-        prefix_cache_output = vllm_model.generate_greedy(
-            INPUT_PROMPTS, max_tokens)
-
-    # TODO: enable apc and chunked prefill with ascend scheduler will lead accuracy problem.
-    # Disable it now. Fix it or drop the ascend scheduler in the future.
-    # with VllmRunner(model,
-    #                 additional_config={
-    #                     'ascend_scheduler_config': {
-    #                         'enabled': True,
-    #                         'enable_prefix_caching': True,
-    #                         "enable_chunked_prefill": True,
-    #                     },
-    #                 },
-    #                 enforce_eager=True,
-    #                 max_model_len=2048,
-    #                 tensor_parallel_size=2,
-    #                 gpu_memory_utilization=0.7) as vllm_model:
-    #     chunk_prefill_prefix_cache_output = vllm_model.generate_greedy(
-    #         INPUT_PROMPTS, max_tokens)
-
-    check_outputs_equal(
-        outputs_0_lst=vllm_output,
-        outputs_1_lst=prefix_cache_output,
-        name_0="vllm_output",
-        name_1="prefix_cache_output",
-    )
-
-    # check_outputs_equal(
-    #     outputs_0_lst=chunk_prefill_prefix_cache_output,
-    #     outputs_1_lst=prefix_cache_output,
-    #     name_0="chunk_prefill_prefix_cache_output",
-    #     name_1="prefix_cache_output",
-    # )

tests/e2e/multicard/test_qwen3_next.py

@@ -24,6 +24,7 @@
 import os
 from unittest.mock import patch
 
+import pytest
 from modelscope import snapshot_download  # type: ignore
 
 from tests.e2e.conftest import VllmRunner
@@ -63,6 +64,8 @@ def test_models_distributed_Qwen3_NEXT_TP4_FULL_DECODE_ONLY():
         del vllm_model
 
 
+@pytest.mark.skip(
+    reason="Qwen3-Next + MTP doesn't work with chunked prefill. Fix Me")
 def test_models_distributed_Qwen3_NEXT_MTP_TP4_SIMILARITY():
     example_prompts = [
         "Hello, my name is",
@@ -89,12 +92,6 @@ def test_models_distributed_Qwen3_NEXT_MTP_TP4_SIMILARITY():
                     gpu_memory_utilization=0.8,
                     distributed_executor_backend="mp",
                     enforce_eager=True,
-                    additional_config={
-                        "ascend_scheduler_config": {
-                            "enabled": True,
-                            "enable_chunked_prefill": False
-                        }
-                    },
                     speculative_config={
                         "method": "qwen3_next_mtp",
                         "num_speculative_tokens": 1