36a0470de1
### What this PR does / why we need it? The env `VLLM_ASCEND_ENABLE_FUSED_MC2` should only enabled in the decoder node during Prefill-Decode Disaggregation scenario --------- Signed-off-by: wangli <wangli858794774@gmail.com>
67 lines
1.9 KiB
YAML
67 lines
1.9 KiB
YAML
# ==========================================
|
|
# ACTUAL TEST CASES
|
|
# ==========================================
|
|
|
|
test_cases:
|
|
- name: "Qwen3.5-397B-A17B-w8a8-mtp"
|
|
model: "Eco-Tech/Qwen3.5-397B-A17B-w8a8-mtp"
|
|
envs:
|
|
VLLM_USE_MODELSCOPE: "true"
|
|
PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
|
|
HCCL_BUFFSIZE: "512"
|
|
OMP_PROC_BIND: "false"
|
|
OMP_NUM_THREADS: "1"
|
|
TASK_QUEUE_ENABLE: "1"
|
|
SERVER_PORT: "DEFAULT_PORT"
|
|
VLLM_ASCEND_ENABLE_FLASHCOMM1: "1"
|
|
server_cmd:
|
|
- "--tensor-parallel-size"
|
|
- "16"
|
|
- "--data-parallel-size"
|
|
- "1"
|
|
- "--enable-expert-parallel"
|
|
- "--port"
|
|
- "$SERVER_PORT"
|
|
- "--seed"
|
|
- "1024"
|
|
- "--max-num-seqs"
|
|
- "32"
|
|
- "--quantization"
|
|
- "ascend"
|
|
- "--max-model-len"
|
|
- "133000"
|
|
- "--max-num-batched-tokens"
|
|
- "8096"
|
|
- "--trust-remote-code"
|
|
- "--gpu-memory-utilization"
|
|
- "0.9"
|
|
- "--no-enable-prefix-caching"
|
|
- "--additional-config"
|
|
- '{"enable_cpu_binding":true,"multistream_overlap_shared_expert":true}'
|
|
- "--speculative_config"
|
|
- '{"method": "qwen3_5_mtp","num_speculative_tokens": 3,"enforce_eager": true}'
|
|
- "--compilation-config"
|
|
- '{"cudagraph_mode":"FULL_DECODE_ONLY"}'
|
|
- "--async-scheduling"
|
|
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: 32768
|
|
batch_size: 32
|
|
baseline: 95
|
|
threshold: 5
|
|
perf:
|
|
case_type: performance
|
|
dataset_path: vllm-ascend/GSM8K-in3500-bs8000-qwen3
|
|
request_conf: vllm_api_stream_chat
|
|
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
|
|
num_prompts: 80
|
|
max_out_len: 1500
|
|
batch_size: 20
|
|
request_rate: 0
|
|
baseline: 768.6894
|
|
threshold: 0.97
|