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

[1/N] Refactor nightly test structure (#5479)

Browse Source 
### What this PR does / why we need it?
This patch is a series of refactoring actions, including clarifying the
directory structure of nightly tests, refactoring the config retrieval
logic, and optimizing the workflow, etc. This is the first step:
refactoring the directory structure of nightly to make it more readable
and logical.

- vLLM version: v0.13.0
- vLLM main:
5326c89803

Signed-off-by: wangli <wangli858794774@gmail.com>
This commit is contained in:
Li Wang
2025-12-30 19:03:02 +08:00
committed by GitHub
parent c85cc045f8
commit e760aae1df
59 changed files with 475 additions and 471 deletions
Download Patch File Download Diff File Expand all files Collapse all files

57
tests/e2e/nightly/multi_node/config/models/DeepSeek-R1-W8A8-A2.yaml
View File

@@ -1,57 +0,0 @@
test_name: "test DeepSeek-R1-W8A8 on A2"
model: "vllm-ascend/DeepSeek-R1-0528-W8A8"
num_nodes: 2
npu_per_node: 8
env_common:
VLLM_USE_MODELSCOPE: true
HCCL_BUFFSIZE: 1024
SERVER_PORT: 8080
OMP_PROC_BIND: false
OMP_NUM_THREADS: 10
deployment:
-
server_cmd: >
vllm serve vllm-ascend/DeepSeek-R1-0528-W8A8
--host 0.0.0.0
--port $SERVER_PORT
--data-parallel-size 4
--data-parallel-size-local 2
--data-parallel-address $LOCAL_IP
--data-parallel-rpc-port 13399
--no-enable-prefix-caching
--max-num-seqs 16
--tensor-parallel-size 4
--max-model-len 36864
--max-num-batched-tokens 6000
--enable-expert-parallel
--trust-remote-code
--quantization ascend
--gpu-memory-utilization 0.9
--enforce-eager
--speculative-config '{"num_speculative_tokens": 1, "method":"mtp"}'
--additional-config '{"enable_weight_nz_layout":true}'
-
server_cmd: >
vllm serve vllm-ascend/DeepSeek-R1-0528-W8A8
--headless
--data-parallel-size 4
--data-parallel-rpc-port 13399
--data-parallel-size-local 2
--data-parallel-start-rank 2
--data-parallel-address $MASTER_IP
--no-enable-prefix-caching
--max-num-seqs 16
--tensor-parallel-size 4
--max-model-len 36864
--max-num-batched-tokens 6000
--enable-expert-parallel
--trust-remote-code
--quantization ascend
--gpu-memory-utilization 0.9
--enforce-eager
--speculative-config '{"num_speculative_tokens": 1, "method":"mtp"}'
--additional-config '{"enable_weight_nz_layout":true}'
benchmarks:

195
tests/e2e/nightly/multi_node/config/models/DeepSeek-R1-W8A8-EPLB.yaml
View File

@@ -1,195 +0,0 @@
test_name: "test DeepSeek-R1-W8A8 disaggregated_prefill"
model: "vllm-ascend/DeepSeek-R1-0528-W8A8"
num_nodes: 4
npu_per_node: 16
env_common:
VLLM_USE_MODELSCOPE: true
HCCL_BUFFSIZE: 1024
SERVER_PORT: 8080
OMP_PROC_BIND: false
OMP_NUM_THREADS: 10
PYTORCH_NPU_ALLOC_CONF: expandable_segments:True
HCCL_DETERMINISTIC: True
TASK_QUEUE_ENABLE: 1
HCCL_OP_RETRY_ENABLE: "L0:0, L1:0, L2:0"
DYNAMIC_EPLB: true
disaggregated_prefill:
enabled: true
prefiller_host_index: [0, 1]
decoder_host_index: [2]
deployment:
-
server_cmd: >
vllm serve vllm-ascend/DeepSeek-R1-0528-W8A8
--host 0.0.0.0
--port $SERVER_PORT
--data-parallel-size 2
--data-parallel-size-local 2
--tensor-parallel-size 8
--enforce-eager
--enable-expert-parallel
--seed 1024
--quantization ascend
--max-num-seqs 4
--max-model-len 36864
--max-num-batched-tokens 16384
--trust-remote-code
--gpu-memory-utilization 0.9
--speculative-config '{"num_speculative_tokens": 1, "method":"mtp"}'
--kv-transfer-config
'{"kv_connector": "MooncakeConnectorV1",
"kv_role": "kv_producer",
"kv_port": "30000",
"engine_id": "0",
"kv_connector_module_path": "vllm_ascend.distributed.mooncake_connector",
"kv_connector_extra_config": {
"prefill": {
"dp_size": 2,
"tp_size": 8
},
"decode": {
"dp_size": 32,
"tp_size": 1
}
}
}'
--additional-config
'{"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
--host 0.0.0.0
--port $SERVER_PORT
--data-parallel-size 2
--data-parallel-size-local 2
--tensor-parallel-size 8
--enforce-eager
--enable-expert-parallel
--seed 1024
--quantization ascend
--max-num-seqs 4
--max-model-len 36864
--max-num-batched-tokens 16384
--trust-remote-code
--gpu-memory-utilization 0.9
--speculative-config '{"num_speculative_tokens": 1, "method":"mtp"}'
--kv-transfer-config
'{"kv_connector": "MooncakeConnectorV1",
"kv_role": "kv_producer",
"kv_port": "30100",
"engine_id": "1",
"kv_connector_module_path": "vllm_ascend.distributed.mooncake_connector",
"kv_connector_extra_config": {
"prefill": {
"dp_size": 2,
"tp_size": 8
},
"decode": {
"dp_size": 32,
"tp_size": 1
}
}
}'
--additional-config
'{"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
--host 0.0.0.0
--port $SERVER_PORT
--data-parallel-size 32
--data-parallel-size-local 16
--data-parallel-start-rank 0
--data-parallel-address $LOCAL_IP
--data-parallel-rpc-port 13389
--tensor-parallel-size 1
--enable-expert-parallel
--seed 1024
--quantization ascend
--max-num-seqs 28
--max-model-len 36864
--max-num-batched-tokens 256
--trust-remote-code
--gpu-memory-utilization 0.9
--speculative-config '{"num_speculative_tokens": 1, "method":"mtp"}'
--kv-transfer-config
'{"kv_connector": "MooncakeConnectorV1",
"kv_role": "kv_consumer",
"kv_port": "30200",
"engine_id": "2",
"kv_connector_module_path": "vllm_ascend.distributed.mooncake_connector",
"kv_connector_extra_config": {
"prefill": {
"dp_size": 2,
"tp_size": 8
},
"decode": {
"dp_size": 32,
"tp_size": 1
}
}
}'
--additional-config
'{"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
--headless
--data-parallel-size 32
--data-parallel-size-local 16
--data-parallel-start-rank 16
--data-parallel-address $MASTER_IP
--data-parallel-rpc-port 13389
--tensor-parallel-size 1
--enable-expert-parallel
--seed 1024
--quantization ascend
--max-num-seqs 28
--max-model-len 36864
--max-num-batched-tokens 256
--trust-remote-code
--gpu-memory-utilization 0.9
--speculative-config '{"num_speculative_tokens": 1, "method":"mtp"}'
--kv-transfer-config
'{"kv_connector": "MooncakeConnectorV1",
"kv_role": "kv_consumer",
"kv_port": "30200",
"engine_id": "2",
"kv_connector_module_path": "vllm_ascend.distributed.mooncake_connector",
"kv_connector_extra_config": {
"prefill": {
"dp_size": 2,
"tp_size": 8
},
"decode": {
"dp_size": 32,
"tp_size": 1
}
}
}'
--additional-config
'{"multistream_overlap_shared_expert":true,"dynamic_eplb":true,"num_iterations_eplb_update":2048,"num_wait_worker_iterations":200}'
benchmarks:
perf:
case_type: performance
dataset_path: vllm-ascend/GSM8K-in3500-bs2800
request_conf: vllm_api_stream_chat
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
num_prompts: 2800
max_out_len: 1500
batch_size: 700
request_rate: 11.2
baseline: 1
threshold: 0.97
acc:
case_type: accuracy
dataset_path: vllm-ascend/gsm8k
request_conf: vllm_api_general_chat
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_chat_prompt
max_out_len: 32768
batch_size: 512
baseline: 95
threshold: 5

194
tests/e2e/nightly/multi_node/config/models/DeepSeek-R1-W8A8.yaml
View File

@@ -1,194 +0,0 @@
test_name: "test DeepSeek-R1-W8A8 disaggregated_prefill"
model: "vllm-ascend/DeepSeek-R1-0528-W8A8"
num_nodes: 4
npu_per_node: 16
env_common:
VLLM_USE_MODELSCOPE: true
HCCL_BUFFSIZE: 1024
SERVER_PORT: 8080
OMP_PROC_BIND: false
OMP_NUM_THREADS: 10
PYTORCH_NPU_ALLOC_CONF: expandable_segments:True
HCCL_DETERMINISTIC: True
TASK_QUEUE_ENABLE: 1
HCCL_OP_RETRY_ENABLE: "L0:0, L1:0, L2:0"
disaggregated_prefill:
enabled: true
prefiller_host_index: [0, 1]
decoder_host_index: [2]
deployment:
-
server_cmd: >
vllm serve vllm-ascend/DeepSeek-R1-0528-W8A8
--host 0.0.0.0
--port $SERVER_PORT
--data-parallel-size 2
--data-parallel-size-local 2
--tensor-parallel-size 8
--enforce-eager
--enable-expert-parallel
--seed 1024
--quantization ascend
--max-num-seqs 4
--max-model-len 36864
--max-num-batched-tokens 16384
--trust-remote-code
--gpu-memory-utilization 0.9
--speculative-config '{"num_speculative_tokens": 1, "method":"mtp"}'
--kv-transfer-config
'{"kv_connector": "MooncakeConnectorV1",
"kv_role": "kv_producer",
"kv_port": "30000",
"engine_id": "0",
"kv_connector_module_path": "vllm_ascend.distributed.mooncake_connector",
"kv_connector_extra_config": {
"prefill": {
"dp_size": 2,
"tp_size": 8
},
"decode": {
"dp_size": 32,
"tp_size": 1
}
}
}'
--additional-config
'{"enable_prefill_optimizations":true,"enable_weight_nz_layout":true}'
-
server_cmd: >
vllm serve vllm-ascend/DeepSeek-R1-0528-W8A8
--host 0.0.0.0
--port $SERVER_PORT
--data-parallel-size 2
--data-parallel-size-local 2
--tensor-parallel-size 8
--enforce-eager
--enable-expert-parallel
--seed 1024
--quantization ascend
--max-num-seqs 4
--max-model-len 36864
--max-num-batched-tokens 16384
--trust-remote-code
--gpu-memory-utilization 0.9
--speculative-config '{"num_speculative_tokens": 1, "method":"mtp"}'
--kv-transfer-config
'{"kv_connector": "MooncakeConnectorV1",
"kv_role": "kv_producer",
"kv_port": "30100",
"engine_id": "1",
"kv_connector_module_path": "vllm_ascend.distributed.mooncake_connector",
"kv_connector_extra_config": {
"prefill": {
"dp_size": 2,
"tp_size": 8
},
"decode": {
"dp_size": 32,
"tp_size": 1
}
}
}'
--additional-config
'{"enable_prefill_optimizations":true,"enable_weight_nz_layout":true}'
-
server_cmd: >
vllm serve vllm-ascend/DeepSeek-R1-0528-W8A8
--host 0.0.0.0
--port $SERVER_PORT
--data-parallel-size 32
--data-parallel-size-local 16
--data-parallel-start-rank 0
--data-parallel-address $LOCAL_IP
--data-parallel-rpc-port 13389
--tensor-parallel-size 1
--enable-expert-parallel
--seed 1024
--quantization ascend
--max-num-seqs 28
--max-model-len 36864
--max-num-batched-tokens 256
--trust-remote-code
--gpu-memory-utilization 0.9
--speculative-config '{"num_speculative_tokens": 1, "method":"mtp"}'
--kv-transfer-config
'{"kv_connector": "MooncakeConnectorV1",
"kv_role": "kv_consumer",
"kv_port": "30200",
"engine_id": "2",
"kv_connector_module_path": "vllm_ascend.distributed.mooncake_connector",
"kv_connector_extra_config": {
"prefill": {
"dp_size": 2,
"tp_size": 8
},
"decode": {
"dp_size": 32,
"tp_size": 1
}
}
}'
--additional-config
'{"multistream_overlap_shared_expert":true}'
-
server_cmd: >
vllm serve vllm-ascend/DeepSeek-R1-0528-W8A8
--headless
--data-parallel-size 32
--data-parallel-size-local 16
--data-parallel-start-rank 16
--data-parallel-address $MASTER_IP
--data-parallel-rpc-port 13389
--tensor-parallel-size 1
--enable-expert-parallel
--seed 1024
--quantization ascend
--max-num-seqs 28
--max-model-len 36864
--max-num-batched-tokens 256
--trust-remote-code
--gpu-memory-utilization 0.9
--speculative-config '{"num_speculative_tokens": 1, "method":"mtp"}'
--kv-transfer-config
'{"kv_connector": "MooncakeConnectorV1",
"kv_role": "kv_consumer",
"kv_port": "30200",
"engine_id": "2",
"kv_connector_module_path": "vllm_ascend.distributed.mooncake_connector",
"kv_connector_extra_config": {
"prefill": {
"dp_size": 2,
"tp_size": 8
},
"decode": {
"dp_size": 32,
"tp_size": 1
}
}
}'
--additional-config
'{"multistream_overlap_shared_expert":true}'
benchmarks:
perf:
case_type: performance
dataset_path: vllm-ascend/GSM8K-in3500-bs2800
request_conf: vllm_api_stream_chat
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
num_prompts: 2800
max_out_len: 1500
batch_size: 700
request_rate: 11.2
baseline: 1
threshold: 0.97
acc:
case_type: accuracy
dataset_path: vllm-ascend/gsm8k
request_conf: vllm_api_general_chat
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_chat_prompt
max_out_len: 32768
batch_size: 512
baseline: 95
threshold: 5

111
tests/e2e/nightly/multi_node/config/models/DeepSeek-V3.yaml
View File

@@ -1,111 +0,0 @@
# For disaggregated mode, set is_disaggregated: true, and set the following parameters:
# Prefiller_index: the hosts index of the node running prefiller
# Decoder_index: the hosts index of the node running decoder
# Suppose we have **4 nodes** running a 2P1D setup (2 Prefillers + 1 Decoder):
# ┌───────────────┬───────────────┬───────────────┬───────────────┐
# │ node0 │ node1 │ node2 │ node3 │
# │ Prefiller #1 │ Prefiller #2 │ Decoder │ Decoder │
# └───────────────┴───────────────┴───────────────┴───────────────┘
# For the prefiller nodes. the hosts should be node0 and node1
# For the decoder nodes. we only have 1 decoder node(dp+tp+ep across node2 and node3. Where node3 is running with headless mode)
# So the prefiller_host_index is [0, 1], and the decoder_host_index is [2]
test_name: "test DeepSeek-V3 disaggregated_prefill"
model: "vllm-ascend/DeepSeek-V3-W8A8"
num_nodes: 2
npu_per_node: 16
env_common:
VLLM_USE_MODELSCOPE: true
OMP_PROC_BIND: false
OMP_NUM_THREADS: 100
HCCL_BUFFSIZE: 1024
SERVER_PORT: 8080
NUMEXPR_MAX_THREADS: 128
DISAGGREGATED_PREFILL_PROXY_SCRIPT: "examples/disaggregated_prefill_v1/load_balance_proxy_server_example.py"
# For None kubernetes deployment, list the IPs of all nodes used in order as follow
# cluster_hosts: []
disaggregated_prefill:
enabled: true
prefiller_host_index: [0]
decoder_host_index: [1]
deployment:
-
server_cmd: >
vllm serve "vllm-ascend/DeepSeek-V3-W8A8"
--host 0.0.0.0
--port $SERVER_PORT
--data-parallel-size 2
--data-parallel-size-local 2
--tensor-parallel-size 8
--seed 1024
--enforce-eager
--enable-expert-parallel
--max-num-seqs 16
--max-model-len 8192
--max-num-batched-tokens 8192
--quantization ascend
--trust-remote-code
--no-enable-prefix-caching
--gpu-memory-utilization 0.9
--kv-transfer-config
'{"kv_connector": "MooncakeConnectorV1",
"kv_role": "kv_producer",
"kv_port": "30000",
"engine_id": "0",
"kv_connector_module_path": "vllm_ascend.distributed.mooncake_connector",
"kv_connector_extra_config": {
"prefill": {
"dp_size": 2,
"tp_size": 8
},
"decode": {
"dp_size": 2,
"tp_size": 8
}
}
}'
-
server_cmd: >
vllm serve "vllm-ascend/DeepSeek-V3-W8A8"
--host 0.0.0.0
--port $SERVER_PORT
--data-parallel-size 2
--data-parallel-size-local 2
--tensor-parallel-size 8
--seed 1024
--quantization ascend
--max-num-seqs 16
--max-model-len 8192
--max-num-batched-tokens 8192
--enable-expert-parallel
--trust-remote-code
--no-enable-prefix-caching
--gpu-memory-utilization 0.9
--kv-transfer-config
'{"kv_connector": "MooncakeConnectorV1",
"kv_role": "kv_consumer",
"kv_port": "30200",
"engine_id": "1",
"kv_connector_module_path": "vllm_ascend.distributed.mooncake_connector",
"kv_connector_extra_config": {
"prefill": {
"dp_size": 2,
"tp_size": 8
},
"decode": {
"dp_size": 2,
"tp_size": 8
}
}
}'
benchmarks:
acc:
case_type: accuracy
dataset_path: vllm-ascend/gsm8k-lite
request_conf: vllm_api_general_chat
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_chat_prompt
max_out_len: 4096
batch_size: 512
baseline: 95
threshold: 5

51
tests/e2e/nightly/multi_node/config/models/DeepSeek-V3_2-Exp-bf16.yaml
View File

@@ -1,51 +0,0 @@
test_name: "test DeepSeek-V3.2-Exp-bf16 multi-dp"
model: "Yanguan/DeepSeek-V3.2-Exp-bf16"
num_nodes: 2
npu_per_node: 16
env_common:
VLLM_USE_MODELSCOPE: true
OMP_PROC_BIND: false
OMP_NUM_THREADS: 100
HCCL_BUFFSIZE: 1024
SERVER_PORT: 8080
VLLM_ASCEND_ENABLE_MLAPO: 0
deployment:
-
server_cmd: >
vllm serve "Yanguan/DeepSeek-V3.2-Exp-bf16"
--host 0.0.0.0
--port $SERVER_PORT
--data-parallel-address $LOCAL_IP
--data-parallel-size 2
--data-parallel-size-local 1
--data-parallel-rpc-port 13389
--tensor-parallel-size 16
--seed 1024
--enable-expert-parallel
--max-num-seqs 16
--max-model-len 17450
--max-num-batched-tokens 17450
--trust-remote-code
--no-enable-prefix-caching
--gpu-memory-utilization 0.9
-
server_cmd: >
vllm serve "Yanguan/DeepSeek-V3.2-Exp-bf16"
--headless
--data-parallel-size 2
--data-parallel-size-local 1
--data-parallel-start-rank 1
--data-parallel-address $MASTER_IP
--data-parallel-rpc-port 13389
--tensor-parallel-size 16
--seed 1024
--max-num-seqs 16
--max-model-len 17450
--max-num-batched-tokens 17450
--enable-expert-parallel
--trust-remote-code
--no-enable-prefix-caching
--gpu-memory-utilization 0.92
benchmarks:

72
tests/e2e/nightly/multi_node/config/models/Qwen3-235B-A22B-A2.yaml
View File

@@ -1,72 +0,0 @@
test_name: "test Qwen3-235B-A22B multi-dp on A2"
model: "Qwen/Qwen3-235B-A22B"
num_nodes: 2
npu_per_node: 8
env_common:
VLLM_USE_MODELSCOPE: true
OMP_PROC_BIND: false
OMP_NUM_THREADS: 1
HCCL_BUFFSIZE: 1024
SERVER_PORT: 8080
NUMEXPR_MAX_THREADS: 128
TASK_QUEUE_ENABLE: 1
PYTORCH_NPU_ALLOC_CONF: expandable_segments:True
deployment:
-
server_cmd: >
vllm serve "Qwen/Qwen3-235B-A22B"
--host 0.0.0.0
--port $SERVER_PORT
--data-parallel-size 2
--data-parallel-size-local 1
--data-parallel-address $LOCAL_IP
--data-parallel-rpc-port 13389
--tensor-parallel-size 8
--seed 1024
--enable-expert-parallel
--max-num-seqs 128
--max-model-len 40960
--max-num-batched-tokens 256
--trust-remote-code
--gpu-memory-utilization 0.9
--async-scheduling
-
server_cmd: >
vllm serve "Qwen/Qwen3-235B-A22B"
--headless
--data-parallel-size 2
--data-parallel-size-local 1
--data-parallel-start-rank 1
--data-parallel-address $MASTER_IP
--data-parallel-rpc-port 13389
--tensor-parallel-size 8
--seed 1024
--max-num-seqs 128
--max-model-len 40960
--max-num-batched-tokens 256
--enable-expert-parallel
--trust-remote-code
--gpu-memory-utilization 0.9
--async-scheduling
benchmarks:
perf:
case_type: performance
dataset_path: vllm-ascend/GSM8K-in3500-bs2800
request_conf: vllm_api_stream_chat
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
num_prompts: 2800
max_out_len: 1500
batch_size: 256
request_rate: 4.8
baseline: 1
threshold: 0.97
acc:
case_type: accuracy
dataset_path: vllm-ascend/gsm8k-lite
request_conf: vllm_api_general_chat
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_chat_prompt
max_out_len: 7680
batch_size: 256
baseline: 96
threshold: 5

70
tests/e2e/nightly/multi_node/config/models/Qwen3-235B-A22B.yaml
View File

@@ -1,70 +0,0 @@
test_name: "test Qwen3-235B-A22B multi-dp"
model: "Qwen/Qwen3-235B-A22B"
num_nodes: 2
npu_per_node: 16
env_common:
VLLM_USE_MODELSCOPE: true
OMP_PROC_BIND: false
OMP_NUM_THREADS: 100
HCCL_BUFFSIZE: 1024
SERVER_PORT: 8080
NUMEXPR_MAX_THREADS: 128
deployment:
-
server_cmd: >
vllm serve "Qwen/Qwen3-235B-A22B"
--host 0.0.0.0
--port $SERVER_PORT
--data-parallel-size 4
--data-parallel-size-local 2
--data-parallel-address $LOCAL_IP
--data-parallel-rpc-port 13389
--tensor-parallel-size 8
--seed 1024
--enable-expert-parallel
--max-num-seqs 32
--max-model-len 8192
--max-num-batched-tokens 8192
--trust-remote-code
--no-enable-prefix-caching
--gpu-memory-utilization 0.9
-
server_cmd: >
vllm serve "Qwen/Qwen3-235B-A22B"
--headless
--data-parallel-size 4
--data-parallel-size-local 2
--data-parallel-start-rank 2
--data-parallel-address $MASTER_IP
--data-parallel-rpc-port 13389
--tensor-parallel-size 8
--seed 1024
--max-num-seqs 32
--max-model-len 8192
--max-num-batched-tokens 8192
--enable-expert-parallel
--trust-remote-code
--no-enable-prefix-caching
--gpu-memory-utilization 0.9
benchmarks:
perf:
case_type: performance
dataset_path: vllm-ascend/GSM8K-in3500-bs2800
request_conf: vllm_api_stream_chat
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
num_prompts: 2800
max_out_len: 1500
batch_size: 700
request_rate: 11.2
baseline: 1
threshold: 0.97
acc:
case_type: accuracy
dataset_path: vllm-ascend/gsm8k
request_conf: vllm_api_general_chat
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_chat_prompt
max_out_len: 7680
batch_size: 512
baseline: 95
threshold: 3

91
tests/e2e/nightly/multi_node/config/models/Qwen3-235B-W8A8-EPLB.yaml
View File

@@ -1,91 +0,0 @@
test_name: "test Qwen3-235B-A22B-W8A8 disaggregated_prefill"
model: "vllm-ascend/Qwen3-235B-A22B-W8A8"
num_nodes: 2
npu_per_node: 16
env_common:
VLLM_USE_MODELSCOPE: true
OMP_PROC_BIND: false
OMP_NUM_THREADS: 100
HCCL_BUFFSIZE: 1024
SERVER_PORT: 8080
DYNAMIC_EPLB: true
disaggregated_prefill:
enabled: true
prefiller_host_index: [0]
decoder_host_index: [1]
deployment:
-
server_cmd: >
vllm serve "vllm-ascend/Qwen3-235B-A22B-W8A8"
--host 0.0.0.0
--port $SERVER_PORT
--data-parallel-size 2
--data-parallel-size-local 2
--tensor-parallel-size 8
--seed 1024
--enable-expert-parallel
--max-num-seqs 16
--max-model-len 8192
--max-num-batched-tokens 8192
--quantization ascend
--trust-remote-code
--no-enable-prefix-caching
--gpu-memory-utilization 0.9
--kv-transfer-config
'{"kv_connector": "MooncakeConnectorV1",
"kv_role": "kv_producer",
"kv_port": "30000",
"engine_id": "0",
"kv_connector_module_path": "vllm_ascend.distributed.mooncake_connector",
"kv_connector_extra_config": {
"prefill": {
"dp_size": 2,
"tp_size": 8
},
"decode": {
"dp_size": 2,
"tp_size": 8
}
}
}'
--additional-config
'{"dynamic_eplb":true,"num_iterations_eplb_update":2048,"num_wait_worker_iterations":200}'
-
server_cmd: >
vllm serve "vllm-ascend/Qwen3-235B-A22B-W8A8"
--host 0.0.0.0
--port $SERVER_PORT
--data-parallel-size 2
--data-parallel-size-local 2
--tensor-parallel-size 8
--seed 1024
--quantization ascend
--max-num-seqs 16
--max-model-len 8192
--max-num-batched-tokens 8192
--enable-expert-parallel
--trust-remote-code
--no-enable-prefix-caching
--gpu-memory-utilization 0.9
--kv-transfer-config
'{"kv_connector": "MooncakeConnectorV1",
"kv_role": "kv_consumer",
"kv_port": "30200",
"engine_id": "1",
"kv_connector_module_path": "vllm_ascend.distributed.mooncake_connector",
"kv_connector_extra_config": {
"prefill": {
"dp_size": 2,
"tp_size": 8
},
"decode": {
"dp_size": 2,
"tp_size": 8
}
}
}'
--additional-config
'{"dynamic_eplb":true,"num_iterations_eplb_update":2048,"num_wait_worker_iterations":200}'
benchmarks:

87
tests/e2e/nightly/multi_node/config/models/Qwen3-235B-W8A8.yaml
View File

@@ -1,87 +0,0 @@
test_name: "test Qwen3-235B-A22B-W8A8 disaggregated_prefill"
model: "vllm-ascend/Qwen3-235B-A22B-W8A8"
num_nodes: 2
npu_per_node: 16
env_common:
VLLM_USE_MODELSCOPE: true
OMP_PROC_BIND: false
OMP_NUM_THREADS: 100
HCCL_BUFFSIZE: 1024
SERVER_PORT: 8080
NUMEXPR_MAX_THREADS: 128
disaggregated_prefill:
enabled: true
prefiller_host_index: [0]
decoder_host_index: [1]
deployment:
-
server_cmd: >
vllm serve "vllm-ascend/Qwen3-235B-A22B-W8A8"
--host 0.0.0.0
--port $SERVER_PORT
--data-parallel-size 2
--data-parallel-size-local 2
--tensor-parallel-size 8
--seed 1024
--enable-expert-parallel
--max-num-seqs 16
--max-model-len 8192
--max-num-batched-tokens 8192
--quantization ascend
--trust-remote-code
--no-enable-prefix-caching
--gpu-memory-utilization 0.9
--kv-transfer-config
'{"kv_connector": "MooncakeConnectorV1",
"kv_role": "kv_producer",
"kv_port": "30000",
"engine_id": "0",
"kv_connector_module_path": "vllm_ascend.distributed.mooncake_connector",
"kv_connector_extra_config": {
"prefill": {
"dp_size": 2,
"tp_size": 8
},
"decode": {
"dp_size": 2,
"tp_size": 8
}
}
}'
-
server_cmd: >
vllm serve "vllm-ascend/Qwen3-235B-A22B-W8A8"
--host 0.0.0.0
--port $SERVER_PORT
--data-parallel-size 2
--data-parallel-size-local 2
--tensor-parallel-size 8
--seed 1024
--quantization ascend
--max-num-seqs 16
--max-model-len 8192
--max-num-batched-tokens 8192
--enable-expert-parallel
--trust-remote-code
--no-enable-prefix-caching
--gpu-memory-utilization 0.9
--kv-transfer-config
'{"kv_connector": "MooncakeConnectorV1",
"kv_role": "kv_consumer",
"kv_port": "30200",
"engine_id": "1",
"kv_connector_module_path": "vllm_ascend.distributed.mooncake_connector",
"kv_connector_extra_config": {
"prefill": {
"dp_size": 2,
"tp_size": 8
},
"decode": {
"dp_size": 2,
"tp_size": 8
}
}
}'
benchmarks: