EngineX/xc-llm-ascend
3
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Revert "drop ascend scheduler" (#4580)

Browse Source 
Reverts vllm-project/vllm-ascend#4498
- vLLM version: v0.11.2
- vLLM main: https://github.com/vllm-project/vllm/commit/v0.11.2
This commit is contained in:
Mengqing Cao
2025-11-29 22:20:48 +08:00
committed by GitHub
parent 4dbe4fd123
commit 517fd9272d
52 changed files with 2948 additions and 85 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
tests/e2e/nightly/features/test_mtpx_deepseek_r1_0528_w8a8.py
View File

@@ -73,7 +73,11 @@ async def test_models(model: str, mode: str) -> None:
"VLLM_RPC_TIMEOUT": "3600000",
"VLLM_EXECUTE_MODEL_TIMEOUT_SECONDS": "3600000"
}
additional_config: dict[str, Any] = {}
additional_config: dict[str, Any] = {
"ascend_scheduler_config": {
"enabled": False
},
}
speculative_config = {
"num_speculative_tokens": 2,
"method": "deepseek_mtp"

3
tests/e2e/nightly/features/test_prefix_cache_deepseek_r1_0528_w8a8.py
View File

@@ -74,6 +74,9 @@ async def test_models(model: str) -> None:
"PYTORCH_NPU_ALLOC_CONF": "expandable_segments:True",
}
additional_config = {
"ascend_scheduler_config": {
"enabled": False
},
"torchair_graph_config": {
"enabled": True,
"enable_multistream_moe": False,

7
tests/e2e/nightly/features/test_prefix_cache_qwen3_32b_int8.py
View File

@@ -68,7 +68,12 @@ aisbench_cases75 = [{
async def test_models(model: str) -> None:
port = get_open_port()
env_dict = {"TASK_QUEUE_ENABLE": "1", "HCCL_OP_EXPANSION_MODE": "AIV"}
additional_config = {"enable_weight_nz_layout": True}
additional_config = {
"ascend_scheduler_config": {
"enabled": False
},
"enable_weight_nz_layout": True
}
server_args = [
"--quantization", "ascend", "--reasoning-parser", "qwen3",
"--tensor-parallel-size", "4", "--port",

3
tests/e2e/nightly/features/test_qwen3_32b_int8_a3_feature_stack3.py
View File

@@ -83,7 +83,8 @@ async def test_models(model: str, tp_size: int) -> None:
"0.9", "--block-size", "128", "--max-num-seqs", "256",
"--enforce-eager", "--max-model-len", "35840",
"--max-num-batched-tokens", "35840", "--additional-config",
'{"enable_weight_nz_layout":true}', "--compilation-config",
'{"ascend_scheduler_config":{"enabled":true},"enable_weight_nz_layout":true}',
"--compilation-config",
'{"cudagraph_mode":"FULL_DECODE_ONLY", "cudagraph_capture_sizes":[1,8,24,48,60]}'
]
with RemoteOpenAIServer(model,

8
tests/e2e/nightly/models/test_deepseek_r1_0528_w8a8.py
View File

@@ -33,6 +33,7 @@ MODES = [
"single",
"aclgraph",
"aclgraph_mlapo",
"no_chunkprefill",
]
prompts = [
@@ -81,6 +82,9 @@ async def test_models(model: str, mode: str) -> None:
"method": "deepseek_mtp"
}
additional_config = {
"ascend_scheduler_config": {
"enabled": False
},
"torchair_graph_config": {
"enabled": True,
"enable_multistream_moe": False,
@@ -108,6 +112,10 @@ async def test_models(model: str, mode: str) -> None:
if mode == "aclgraph_mlapo":
env_dict["VLLM_ASCEND_ENABLE_MLAPO"] = "1"
additional_config["torchair_graph_config"] = {"enabled": False}
if mode == "no_chunkprefill":
additional_config["ascend_scheduler_config"] = {"enabled": True}
i = server_args.index("--max-num-batched-tokens") + 1
server_args[i] = "36864"
server_args.extend(["--additional-config", json.dumps(additional_config)])
request_keyword_args: dict[str, Any] = {
**api_keyword_args,

3
tests/e2e/nightly/models/test_deepseek_r1_w8a8_eplb.py
View File

@@ -71,6 +71,9 @@ async def test_models(model: str) -> None:
"cudagraph_mode": "FULL_DECODE_ONLY"
}
additional_config: dict[str, Any] = {
"ascend_scheduler_config": {
"enabled": False
},
"torchair_graph_config": {
"enabled": True
},

1
tests/e2e/nightly/models/test_deepseek_v3_2_exp_w8a8.py
View File

@@ -92,6 +92,7 @@ async def test_models(model: str, tp_size: int, dp_size: int,
"--gpu-memory-utilization",
"0.9",
"--additional-config",
'{"ascend_scheduler_config":{"enabled":true},'
'"torchair_graph_config":{"enabled":true,"graph_batch_sizes":[16]}}',
]
if full_graph:

5
tests/e2e/nightly/models/test_qwen2_5_vl_32b.py
View File

@@ -85,8 +85,9 @@ async def test_models(model: str, tp_size: int) -> None:
str(tp_size), "--port",
str(port), "--max-model-len", "30000", "--max-num-batched-tokens",
"40000", "--max-num-seqs", "400", "--trust-remote-code",
"--gpu-memory-utilization", "0.8", "--compilation_config",
'{"cudagraph_mode": "FULL_DECODE_ONLY"}'
"--gpu-memory-utilization", "0.8", "--additional-config",
'{"ascend_scheduler_config":{"enabled":false}}',
"--compilation_config", '{"cudagraph_mode": "FULL_DECODE_ONLY"}'
]
request_keyword_args: dict[str, Any] = {
**api_keyword_args,

6
tests/e2e/nightly/models/test_qwen3_235b_a22b_w8a8_eplb.py
View File

@@ -60,7 +60,11 @@ async def test_models(model: str) -> None:
"PYTORCH_NPU_ALLOC_CONF": "expandable_segments:True",
"VLLM_ASCEND_ENABLE_FLASHCOMM1": "1"
}
additional_config: dict[str, Any] = {}
additional_config: dict[str, Any] = {
"ascend_scheduler_config": {
"enabled": False
},
}
compilation_config = {"cudagraph_mode": "FULL_DECODE_ONLY"}
server_args = [
"--quantization", "ascend", "--async-scheduling",

6
tests/e2e/nightly/models/test_qwen3_235b_w8a8.py
View File

@@ -63,6 +63,11 @@ async def test_models(model: str, mode: str) -> None:
"PYTORCH_NPU_ALLOC_CONF": "expandable_segments:True",
"VLLM_ASCEND_ENABLE_FLASHCOMM1": "1"
}
additional_config: dict[str, Any] = {
"ascend_scheduler_config": {
"enabled": False
},
}
compilation_config = {"cudagraph_mode": "FULL_DECODE_ONLY"}
server_args = [
"--quantization", "ascend", "--async-scheduling",
@@ -77,6 +82,7 @@ async def test_models(model: str, mode: str) -> None:
server_args.extend(
["--compilation-config",
json.dumps(compilation_config)])
server_args.extend(["--additional-config", json.dumps(additional_config)])
request_keyword_args: dict[str, Any] = {
**api_keyword_args,
}

2
tests/e2e/nightly/models/test_qwq_32b.py
View File

@@ -93,6 +93,8 @@ async def test_models(model: str, mode: str, tp_size: int) -> None:
server_args.remove(
'{"cudagraph_mode":"FULL_DECODE_ONLY", "cudagraph_capture_sizes": [1, 8, 24, 48, 60]}'
)
server_args.append("--additional-config")
server_args.append('{"ascend_scheduler_config":{"enabled":true}}')
server_args.append("--enforce-eager")
request_keyword_args: dict[str, Any] = {
**api_keyword_args,

4
tests/e2e/nightly/multi_node/config/models/DeepSeek-R1-W8A8-A2-torchair.yaml
View File

@@ -30,7 +30,7 @@ deployment:
--quantization ascend
--gpu-memory-utilization 0.9
--speculative-config '{"num_speculative_tokens": 1, "method":"deepseek_mtp"}'
--additional-config '{"torchair_graph_config":{"enabled":true,"enable_multistream_moe":true},"chunked_prefill_for_mla":true,"enable_weight_nz_layout":true}'
--additional-config '{"ascend_scheduler_config":{"enabled":false},"torchair_graph_config":{"enabled":true,"enable_multistream_moe":true},"chunked_prefill_for_mla":true,"enable_weight_nz_layout":true}'
-
server_cmd: >
@@ -51,7 +51,7 @@ deployment:
--quantization ascend
--gpu-memory-utilization 0.9
--speculative-config '{"num_speculative_tokens": 1, "method":"deepseek_mtp"}'
--additional-config '{"torchair_graph_config":{"enabled":true,"enable_multistream_moe":true},"chunked_prefill_for_mla":true,"enable_weight_nz_layout":true}'
--additional-config '{"ascend_scheduler_config":{"enabled":false},"torchair_graph_config":{"enabled":true,"enable_multistream_moe":true},"chunked_prefill_for_mla":true,"enable_weight_nz_layout":true}'
benchmarks:
acc:
case_type: accuracy

4
tests/e2e/nightly/multi_node/config/models/DeepSeek-R1-W8A8-A2.yaml
View File

@@ -31,7 +31,7 @@ deployment:
--gpu-memory-utilization 0.9
--enforce-eager
--speculative-config '{"num_speculative_tokens": 1, "method":"deepseek_mtp"}'
--additional-config '{"chunked_prefill_for_mla":true,"enable_weight_nz_layout":true}'
--additional-config '{"ascend_scheduler_config":{"enabled":false},"chunked_prefill_for_mla":true,"enable_weight_nz_layout":true}'
-
server_cmd: >
@@ -53,5 +53,5 @@ deployment:
--gpu-memory-utilization 0.9
--enforce-eager
--speculative-config '{"num_speculative_tokens": 1, "method":"deepseek_mtp"}'
--additional-config '{"chunked_prefill_for_mla":true,"enable_weight_nz_layout":true}'
--additional-config '{"ascend_scheduler_config":{"enabled":false},"chunked_prefill_for_mla":true,"enable_weight_nz_layout":true}'
benchmarks:

8
tests/e2e/nightly/multi_node/config/models/DeepSeek-R1-W8A8-EPLB.yaml
View File

@@ -50,7 +50,7 @@ deployment:
"kv_connector_module_path": "vllm_ascend.distributed.llmdatadist_c_mgr_connector"
}'
--additional-config
'{"torchair_graph_config":{"enabled":false,"enable_multistream_shared_expert":false},"enable_prefill_optimizations":true,"enable_weight_nz_layout":true,"dynamic_eplb":true,"num_iterations_eplb_update":2048,"num_wait_worker_iterations":200}'
'{"ascend_scheduler_config":{"enabled":false},"torchair_graph_config":{"enabled":false,"enable_multistream_shared_expert":false},"enable_prefill_optimizations":true,"enable_weight_nz_layout":true,"dynamic_eplb":true,"num_iterations_eplb_update":2048,"num_wait_worker_iterations":200}'
-
server_cmd: >
@@ -80,7 +80,7 @@ deployment:
"kv_connector_module_path": "vllm_ascend.distributed.llmdatadist_c_mgr_connector"
}'
--additional-config
'{"torchair_graph_config":{"enabled":false,"enable_multistream_shared_expert":false},"enable_prefill_optimizations":true,"enable_weight_nz_layout":true,"dynamic_eplb":true,"num_iterations_eplb_update":2048,"num_wait_worker_iterations":200}'
'{"ascend_scheduler_config":{"enabled":false},"torchair_graph_config":{"enabled":false,"enable_multistream_shared_expert":false},"enable_prefill_optimizations":true,"enable_weight_nz_layout":true,"dynamic_eplb":true,"num_iterations_eplb_update":2048,"num_wait_worker_iterations":200}'
-
server_cmd: >
vllm serve vllm-ascend/DeepSeek-R1-0528-W8A8
@@ -111,7 +111,7 @@ deployment:
"kv_connector_module_path": "vllm_ascend.distributed.llmdatadist_c_mgr_connector"
}'
--additional-config
'{"torchair_graph_config":{"enabled":true,"enable_multistream_mla":true,"graph_batch_sizes":[28],"use_cached_graph":true,"enable_super_kernel":false},"multistream_overlap_shared_expert":true,"dynamic_eplb":true,"num_iterations_eplb_update":2048,"num_wait_worker_iterations":200}'
'{"ascend_scheduler_config":{"enabled":false},"torchair_graph_config":{"enabled":true,"enable_multistream_mla":true,"graph_batch_sizes":[28],"use_cached_graph":true,"enable_super_kernel":false},"multistream_overlap_shared_expert":true,"dynamic_eplb":true,"num_iterations_eplb_update":2048,"num_wait_worker_iterations":200}'
-
server_cmd: >
vllm serve vllm-ascend/DeepSeek-R1-0528-W8A8
@@ -141,7 +141,7 @@ deployment:
"kv_connector_module_path": "vllm_ascend.distributed.llmdatadist_c_mgr_connector"
}'
--additional-config
'{"torchair_graph_config":{"enabled":true,"enable_multistream_mla":true,"graph_batch_sizes":[28],"use_cached_graph":true,"enable_super_kernel":false},"multistream_overlap_shared_expert":true,"dynamic_eplb":true,"num_iterations_eplb_update":2048,"num_wait_worker_iterations":200}'
'{"ascend_scheduler_config":{"enabled":false},"torchair_graph_config":{"enabled":true,"enable_multistream_mla":true,"graph_batch_sizes":[28],"use_cached_graph":true,"enable_super_kernel":false},"multistream_overlap_shared_expert":true,"dynamic_eplb":true,"num_iterations_eplb_update":2048,"num_wait_worker_iterations":200}'
benchmarks:
perf:
case_type: performance

8
tests/e2e/nightly/multi_node/config/models/DeepSeek-R1-W8A8.yaml
View File

@@ -49,7 +49,7 @@ deployment:
"kv_connector_module_path": "vllm_ascend.distributed.llmdatadist_c_mgr_connector"
}'
--additional-config
'{"torchair_graph_config":{"enabled":false,"enable_multistream_shared_expert":false},"enable_prefill_optimizations":true,"enable_weight_nz_layout":true}'
'{"ascend_scheduler_config":{"enabled":false},"torchair_graph_config":{"enabled":false,"enable_multistream_shared_expert":false},"enable_prefill_optimizations":true,"enable_weight_nz_layout":true}'
-
server_cmd: >
@@ -79,7 +79,7 @@ deployment:
"kv_connector_module_path": "vllm_ascend.distributed.llmdatadist_c_mgr_connector"
}'
--additional-config
'{"torchair_graph_config":{"enabled":false,"enable_multistream_shared_expert":false},"enable_prefill_optimizations":true,"enable_weight_nz_layout":true}'
'{"ascend_scheduler_config":{"enabled":false},"torchair_graph_config":{"enabled":false,"enable_multistream_shared_expert":false},"enable_prefill_optimizations":true,"enable_weight_nz_layout":true}'
-
server_cmd: >
vllm serve vllm-ascend/DeepSeek-R1-0528-W8A8
@@ -110,7 +110,7 @@ deployment:
"kv_connector_module_path": "vllm_ascend.distributed.llmdatadist_c_mgr_connector"
}'
--additional-config
'{"torchair_graph_config":{"enabled":true,"enable_multistream_mla":true,"graph_batch_sizes":[28],"use_cached_graph":true,"enable_super_kernel":false},"multistream_overlap_shared_expert":true}'
'{"ascend_scheduler_config":{"enabled":false},"torchair_graph_config":{"enabled":true,"enable_multistream_mla":true,"graph_batch_sizes":[28],"use_cached_graph":true,"enable_super_kernel":false},"multistream_overlap_shared_expert":true}'
-
server_cmd: >
vllm serve vllm-ascend/DeepSeek-R1-0528-W8A8
@@ -140,7 +140,7 @@ deployment:
"kv_connector_module_path": "vllm_ascend.distributed.llmdatadist_c_mgr_connector"
}'
--additional-config
'{"torchair_graph_config":{"enabled":true,"enable_multistream_mla":true,"graph_batch_sizes":[28],"use_cached_graph":true,"enable_super_kernel":false},"multistream_overlap_shared_expert":true}'
'{"ascend_scheduler_config":{"enabled":false},"torchair_graph_config":{"enabled":true,"enable_multistream_mla":true,"graph_batch_sizes":[28],"use_cached_graph":true,"enable_super_kernel":false},"multistream_overlap_shared_expert":true}'
benchmarks:
perf:
case_type: performance

4
tests/e2e/nightly/multi_node/config/models/DeepSeek-V3_2-Exp-bf16.yaml
View File

@@ -29,7 +29,7 @@ deployment:
--trust-remote-code
--no-enable-prefix-caching
--gpu-memory-utilization 0.9
--additional-config '{"torchair_graph_config":{"enabled":true,"graph_batch_sizes":[16]}}'
--additional-config '{"ascend_scheduler_config":{"enabled":true},"torchair_graph_config":{"enabled":true,"graph_batch_sizes":[16]}}'
-
server_cmd: >
@@ -49,5 +49,5 @@ deployment:
--trust-remote-code
--no-enable-prefix-caching
--gpu-memory-utilization 0.92
--additional-config '{"torchair_graph_config":{"enabled":true,"graph_batch_sizes":[16]}}'
--additional-config '{"ascend_scheduler_config":{"enabled":true},"torchair_graph_config":{"enabled":true,"graph_batch_sizes":[16]}}'
benchmarks: