[CI][Doc] Optimize multi-node CI (#3565)
### What this PR does / why we need it?
This pull request mainly do the following things:
1. Add a doc for multi-node CI, The main content is the mechanism
principle and how to contribute
2. Simplify the config yaml for more developer-friendly
3. Optimized the mooncake installation script to prevent accidental
failures during installation
4. Fix the workflow to ensure the kubernetes can be apply correctly
5. Add Qwen3-235B-W8A8 disaggregated_prefill test
6. Add GLM-4.5 multi dp test
7. Add 2p1d 4nodes disaggregated_prefill test
8. Refactor nightly tests
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
- vLLM version: v0.11.0rc3
- vLLM main:
17c540a993
---------
Signed-off-by: wangli <wangli858794774@gmail.com>
This commit is contained in:
Li Wang
163
tests/e2e/nightly/multi_node/config/models/DeepSeek-R1-W8A8.yaml
Normal file
163
tests/e2e/nightly/multi_node/config/models/DeepSeek-R1-W8A8.yaml
Normal file
@@ -0,0 +1,163 @@
|
||||
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]
|
||||
ranktable_gen_path: "examples/disaggregated_prefill_v1/gen_ranktable.py"
|
||||
ranktable_path: "/tmp/ranktable.json"
|
||||
|
||||
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":"deepseek_mtp"}'
|
||||
--kv-transfer-config
|
||||
'{"kv_connector": "LLMDataDistCMgrConnector",
|
||||
"kv_buffer_device": "npu",
|
||||
"kv_role": "kv_producer",
|
||||
"kv_parallel_size": 1,
|
||||
"kv_port": "20001",
|
||||
"engine_id": "0",
|
||||
"kv_connector_module_path": "vllm_ascend.distributed.llmdatadist_c_mgr_connector"
|
||||
}'
|
||||
--additional-config
|
||||
'{"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
|
||||
--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":"deepseek_mtp"}'
|
||||
--kv-transfer-config
|
||||
'{"kv_connector": "LLMDataDistCMgrConnector",
|
||||
"kv_buffer_device": "npu",
|
||||
"kv_role": "kv_producer",
|
||||
"kv_parallel_size": 1,
|
||||
"kv_port": "20001",
|
||||
"engine_id": "0",
|
||||
"kv_connector_module_path": "vllm_ascend.distributed.llmdatadist_c_mgr_connector"
|
||||
}'
|
||||
--additional-config
|
||||
'{"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
|
||||
--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":"deepseek_mtp"}'
|
||||
--kv-transfer-config
|
||||
'{"kv_connector": "LLMDataDistCMgrConnector",
|
||||
"kv_buffer_device": "npu",
|
||||
"kv_role": "kv_consumer",
|
||||
"kv_parallel_size": 1,
|
||||
"kv_port": "20001",
|
||||
"engine_id": "0",
|
||||
"kv_connector_module_path": "vllm_ascend.distributed.llmdatadist_c_mgr_connector"
|
||||
}'
|
||||
--additional-config
|
||||
'{"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
|
||||
--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":"deepseek_mtp"}'
|
||||
--kv-transfer-config
|
||||
'{"kv_connector": "LLMDataDistCMgrConnector",
|
||||
"kv_buffer_device": "npu",
|
||||
"kv_role": "kv_consumer",
|
||||
"kv_parallel_size": 1,
|
||||
"kv_port": "20001",
|
||||
"engine_id": "0",
|
||||
"kv_connector_module_path": "vllm_ascend.distributed.llmdatadist_c_mgr_connector"
|
||||
}'
|
||||
--additional-config
|
||||
'{"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
|
||||
dataset_path: vllm-ascend/GSM8K-in3500-bs400
|
||||
request_conf: vllm_api_stream_chat
|
||||
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
|
||||
num_prompts: 1
|
||||
max_out_len: 2
|
||||
batch_size: 1
|
||||
baseline: 5
|
||||
threshold: 0.97
|
||||
acc:
|
||||
case_type: accuracy
|
||||
dataset_path: vllm-ascend/AIME2024
|
||||
request_conf: vllm_api_general_chat
|
||||
dataset_conf: aime2024/aime2024_gen_0_shot_chat_prompt
|
||||
max_out_len: 10
|
||||
batch_size: 32
|
||||
baseline: 1
|
||||
threshold: 1
|
||||
@@ -26,10 +26,6 @@ disaggregated_prefill:
|
||||
|
||||
deployment:
|
||||
-
|
||||
local_index: 0
|
||||
master_index: 0
|
||||
headless: false
|
||||
env_extend:
|
||||
server_cmd: >
|
||||
vllm serve "vllm-ascend/DeepSeek-V3-W8A8"
|
||||
--host 0.0.0.0
|
||||
@@ -66,10 +62,6 @@ deployment:
|
||||
}'
|
||||
|
||||
-
|
||||
local_index: 1
|
||||
master_index: 0
|
||||
headless: true
|
||||
env_extend:
|
||||
server_cmd: >
|
||||
vllm serve "vllm-ascend/DeepSeek-V3-W8A8"
|
||||
--host 0.0.0.0
|
||||
|
||||
68
tests/e2e/nightly/multi_node/config/models/GLM-4_5.yaml
Normal file
68
tests/e2e/nightly/multi_node/config/models/GLM-4_5.yaml
Normal file
@@ -0,0 +1,68 @@
|
||||
test_name: "test GLM-4.5 multi-dp"
|
||||
model: "ZhipuAI/GLM-4.5"
|
||||
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
|
||||
|
||||
deployment:
|
||||
-
|
||||
server_cmd: >
|
||||
vllm serve "ZhipuAI/GLM-4.5"
|
||||
--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 16
|
||||
--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 "ZhipuAI/GLM-4.5"
|
||||
--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 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
|
||||
benchmarks:
|
||||
perf:
|
||||
case_type: performance
|
||||
dataset_path: vllm-ascend/GSM8K-in3500-bs400
|
||||
request_conf: vllm_api_stream_chat
|
||||
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
|
||||
num_prompts: 1
|
||||
max_out_len: 2
|
||||
batch_size: 1
|
||||
baseline: 5
|
||||
threshold: 0.97
|
||||
acc:
|
||||
case_type: accuracy
|
||||
dataset_path: vllm-ascend/AIME2024
|
||||
request_conf: vllm_api_general_chat
|
||||
dataset_conf: aime2024/aime2024_gen_0_shot_chat_prompt
|
||||
max_out_len: 10
|
||||
batch_size: 32
|
||||
baseline: 1
|
||||
threshold: 1
|
||||
@@ -11,10 +11,6 @@ env_common:
|
||||
|
||||
deployment:
|
||||
-
|
||||
local_index: 0
|
||||
master_index: 0
|
||||
headless: false
|
||||
env_extend:
|
||||
server_cmd: >
|
||||
vllm serve "Qwen/Qwen3-235B-A22B"
|
||||
--host 0.0.0.0
|
||||
@@ -33,10 +29,6 @@ deployment:
|
||||
--no-enable-prefix-caching
|
||||
--gpu-memory-utilization 0.9
|
||||
-
|
||||
local_index: 1
|
||||
master_index: 0
|
||||
headless: true
|
||||
env_extend:
|
||||
server_cmd: >
|
||||
vllm serve "Qwen/Qwen3-235B-A22B"
|
||||
--headless
|
||||
|
||||
105
tests/e2e/nightly/multi_node/config/models/Qwen3-235B-W8A8.yaml
Normal file
105
tests/e2e/nightly/multi_node/config/models/Qwen3-235B-W8A8.yaml
Normal file
@@ -0,0 +1,105 @@
|
||||
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
|
||||
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": "MooncakeConnector",
|
||||
"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": "MooncakeConnector",
|
||||
"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:
|
||||
perf:
|
||||
case_type: performance
|
||||
dataset_path: vllm-ascend/GSM8K-in3500-bs400
|
||||
request_conf: vllm_api_stream_chat
|
||||
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
|
||||
num_prompts: 1
|
||||
max_out_len: 2
|
||||
batch_size: 1
|
||||
baseline: 5
|
||||
threshold: 0.97
|
||||
acc:
|
||||
case_type: accuracy
|
||||
dataset_path: vllm-ascend/AIME2024
|
||||
request_conf: vllm_api_general_chat
|
||||
dataset_conf: aime2024/aime2024_gen_0_shot_chat_prompt
|
||||
max_out_len: 10
|
||||
batch_size: 32
|
||||
baseline: 1
|
||||
threshold: 1
|
||||
Reference in New Issue
Block a user