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

[CI][Doc] Optimize multi-node CI (#3565)

Browse Source 
### 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
2025-10-25 09:23:47 +08:00
committed by GitHub
parent 292cf339c3
commit 7f73c28a24
21 changed files with 1165 additions and 378 deletions
Download Patch File Download Diff File Expand all files Collapse all files

163
tests/e2e/nightly/multi_node/config/models/DeepSeek-R1-W8A8.yaml Normal file
View 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

8
tests/e2e/nightly/multi_node/config/models/DeepSeek-V3.yaml
View File

@@ -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
View 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

8
tests/e2e/nightly/multi_node/config/models/Qwen3-235B-A3B.yaml
View File

@@ -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
View 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

139
tests/e2e/nightly/multi_node/config/multi_node_config.py
View File

@@ -1,6 +1,7 @@
import logging
import os
import subprocess
from dataclasses import dataclass
from typing import Optional
import regex as re
@@ -15,6 +16,16 @@ from tests.e2e.nightly.multi_node.config.utils import (get_avaliable_port,
setup_logger()
logger = logging.getLogger(__name__)
DISAGGREGATED_PREFILL_PROXY_SCRIPT = "examples/disaggregated_prefill_v1/load_balance_proxy_layerwise_server_example.py"
DISAGGEGATED_PREFILL_PORT = 5333
@dataclass
class NodeInfo:
index: int
ip: str
server_cmd: str
headless: bool
server_port: int
class MultiNodeConfig:
@@ -22,38 +33,50 @@ class MultiNodeConfig:
def __init__(self,
model: str,
test_name: str,
num_nodes: int = 2,
npu_per_node: int = 16,
server_port: int = 8080,
headless: bool = False,
disaggregated_prefill: Optional[dict] = None,
envs: Optional[dict] = None,
server_cmd: str = "",
nodes_info: Optional[list[NodeInfo]] = None,
perf_cmd: Optional[str] = None,
acc_cmd: Optional[str] = None):
self.test_name = test_name
self.model = model
self.num_nodes = num_nodes
self.nodes_info = nodes_info or []
self.num_nodes = len(self.nodes_info)
self.npu_per_node = npu_per_node
self.envs = envs if envs is not None else {}
self.server_port = server_port
if disaggregated_prefill:
self.proxy_port = get_avaliable_port()
self.headless = headless
self.server_cmd = server_cmd
self.envs = envs if envs is not None else {}
self.proxy_port = get_avaliable_port()
self.perf_cmd = perf_cmd
self.acc_cmd = acc_cmd
assert perf_cmd is not None, "perf_cmd must be provided"
assert acc_cmd is not None, "acc_cmd must be provided"
assert server_cmd is not None, "server_cmd must be provided"
self.cur_index = os.getenv("LWS_WORKER_INDEX", 0)
self.cur_index = int(os.getenv("LWS_WORKER_INDEX", 0))
self.cur_ip = get_cur_ip()
self.nic_name = get_net_interface(self.cur_ip)
self.cluster_ips = get_cluster_ips(num_nodes)
self.cluster_ips = get_cluster_ips(self.num_nodes)
self.cur_node_info: NodeInfo = self.nodes_info[self.cur_index]
self.disaggregated_prefill = disaggregated_prefill
self._init_disaggregated_prefill()
self._init_dist_env()
self.server_cmd = self._expand_env_vars(self.server_cmd, self.envs)
self.server_cmd = self._expand_env_vars(self.cur_node_info.server_cmd,
self.envs)
def _init_disaggregated_prefill(self):
if self.disaggregated_prefill:
decode_host_index = self.disaggregated_prefill.get(
"decoder_host_index")
if not decode_host_index:
raise RuntimeError("got empty decode_host_index")
self.decode_start_index: int = decode_host_index[0]
self.num_prefillers = self.decode_start_index
self.num_decoders = self.num_nodes - self.num_prefillers
if self.disaggregated_prefill.get(
"ranktable_gen_path") is not None:
self._gen_ranktable()
def _init_dist_env(self):
self.envs["HCCL_IF_IP"] = self.cur_ip
@@ -62,7 +85,17 @@ class MultiNodeConfig:
self.envs["HCCL_SOCKET_IFNAME"] = self.nic_name
self.envs["LOCAL_IP"] = self.cur_ip
self.envs["NIC_NAME"] = self.nic_name
self.envs["MASTER_IP"] = self.cluster_ips[0]
if self.disaggregated_prefill:
self.envs[
"DISAGGREGATED_PREFILL_RANK_TABLE_PATH"] = self.disaggregated_prefill.get(
"ranktable_path")
if self.cur_index < self.decode_start_index:
self.envs["MASTER_IP"] = self.cluster_ips[0]
else:
self.envs["MASTER_IP"] = self.cluster_ips[
self.decode_start_index]
ascend_path = "/usr/local/Ascend/ascend-toolkit/latest/python/site-packages"
self.envs[
"LD_LIBRARY_PATH"] = f"{ascend_path}:{self.envs.get('LD_LIBRARY_PATH', os.environ.get('LD_LIBRARY_PATH', ''))}"
@@ -172,15 +205,21 @@ class MultiNodeConfig:
deployments = config_data.get("deployment", [])
assert len(deployments) == num_nodes, \
f"Number of deployments ({len(deployments)}) must match num_nodes ({num_nodes})"
for deployment in deployments:
if deployment.get("local_index") == int(
os.getenv("LWS_WORKER_INDEX", 0)):
envs_extend = deployment.get("env_extend", {})
if envs_extend:
envs.update(envs_extend)
server_cmd = deployment.get("server_cmd")
headless = deployment.get("headless", False)
break
cluster_ips = get_cluster_ips(num_nodes)
nodes_info = []
for index, deployment in enumerate(deployments):
# after assert len(deployments) == num_nodes, we can assume that this will must have a match
server_cmd = deployment.get("server_cmd", "")
headless = "--headless" in server_cmd
nodes_info.append(
NodeInfo(ip=cluster_ips[index],
index=index,
headless=headless,
server_port=server_port,
server_cmd=server_cmd))
benchmarks = config_data.get("benchmarks", {})
assert benchmarks is not None, "benchmarks must be provided"
perf_cmd = benchmarks["perf"]
@@ -188,13 +227,11 @@ class MultiNodeConfig:
return cls(model=model,
test_name=test_name,
num_nodes=num_nodes,
npu_per_node=npu_per_node,
envs=envs,
server_port=server_port,
headless=headless,
disaggregated_prefill=disaggregated_prefill,
server_cmd=server_cmd,
nodes_info=nodes_info,
perf_cmd=perf_cmd,
acc_cmd=acc_cmd)
@@ -204,4 +241,52 @@ class MultiNodeConfig:
@property
def is_master(self):
return int(self.cur_index) == 0
return self.cur_index == 0
def _gen_ranktable(self):
cluster_ip = self.cluster_ips
assert len(cluster_ip) > 0
nnodes = self.num_nodes
node_rank = self.cur_index
master_addr = cluster_ip[0]
master_port = DISAGGEGATED_PREFILL_PORT
assert self.disaggregated_prefill is not None
ranktable_gen_path = self.disaggregated_prefill.get(
"ranktable_gen_path")
ranktable_path = self.disaggregated_prefill.get("ranktable_path")
assert ranktable_gen_path is not None and ranktable_path is not None
if os.path.exists(str(ranktable_path)):
return
local_host = self.cur_ip
cmd = [
"torchrun",
"--nproc_per_node",
"1",
"--nnodes",
str(nnodes),
"--node_rank",
str(node_rank),
"--master_addr",
master_addr,
"--master_port",
str(master_port),
ranktable_gen_path,
"--ranktable-path",
str(ranktable_path),
"--local-host",
local_host,
"--prefill-device-cnt",
str(self.npu_per_node * self.num_prefillers),
"--decode-device-cnt",
str(self.npu_per_node * self.num_decoders),
]
env = os.environ.copy()
assert self.nic_name is not None
env["GLOO_SOCKET_IFNAME"] = self.nic_name
subprocess.run(cmd, env=env, check=True)
assert os.path.exists(
str(ranktable_path)), "failed generate ranktable.json"