[CI] Align multi-node nightly test paramter with corresponding tutorials document (#5756)

### What this PR does / why we need it?
Align multi-node nightly test paramter with tutorials documents.

### Does this PR introduce _any_ user-facing change?
NA

### How was this patch tested?
Test locally and nighly e2e multi-node test cases.

- vLLM version: v0.13.0
- vLLM main:
2f4e6548ef

---------

Signed-off-by: leo-pony <nengjunma@outlook.com>
This commit is contained in:
Nengjun Ma
2026-01-12 09:00:31 +08:00
committed by GitHub
parent 6880c1b383
commit 297f6deb09
10 changed files with 66 additions and 35 deletions

View File

@@ -3,11 +3,12 @@ model: "vllm-ascend/DeepSeek-R1-0528-W8A8"
num_nodes: 2
npu_per_node: 16
env_common:
HCCL_OP_EXPANSION_MODE: AIV
VLLM_USE_MODELSCOPE: true
HCCL_BUFFSIZE: 1024
HCCL_BUFFSIZE: 768
SERVER_PORT: 8080
OMP_PROC_BIND: false
OMP_NUM_THREADS: 10
OMP_NUM_THREADS: 1
PYTORCH_NPU_ALLOC_CONF: expandable_segments:True
HCCL_DETERMINISTIC: True
TASK_QUEUE_ENABLE: 1
@@ -34,10 +35,10 @@ deployment:
--seed 1024
--quantization ascend
--max-num-seqs 4
--max-model-len 36864
--max-model-len 32768
--max-num-batched-tokens 16384
--trust-remote-code
--gpu-memory-utilization 0.9
--gpu-memory-utilization 0.85
--enable-chunked-prefill
--speculative-config '{"num_speculative_tokens": 3, "method":"mtp"}'
--kv-transfer-config
@@ -72,10 +73,10 @@ deployment:
--seed 1024
--quantization ascend
--max-num-seqs 4
--max-model-len 36864
--max-model-len 32768
--max-num-batched-tokens 256
--trust-remote-code
--gpu-memory-utilization 0.9
--gpu-memory-utilization 0.85
--compilation_config '{"cudagraph_capture_sizes":[4,8,12,16],"cudagraph_mode": "FULL_DECODE_ONLY"}'
--enable-chunked-prefill
--speculative-config '{"num_speculative_tokens": 3, "method":"mtp"}'
@@ -103,7 +104,7 @@ benchmarks:
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
max_out_len: 24576
batch_size: 512
baseline: 95
threshold: 5