9615bc33fd
### What this PR does / why we need it?
This pr modifies qwen3Next nightly CI config.
(1) Add a nightly CI .
(2) Set a more precise accuracy standard
- vLLM version: v0.18.0
- vLLM main:
6a9cceb219
Signed-off-by: Your Name <you@example.com>
Co-authored-by: Your Name <you@example.com>
46 lines
1.2 KiB
YAML
46 lines
1.2 KiB
YAML
# ==========================================
|
|
# ACTUAL TEST CASES
|
|
# ==========================================
|
|
|
|
test_cases:
|
|
- name: "Qwen3-Next-80B-A3B-Instruct-W8A8"
|
|
model: "vllm-ascend/Qwen3-Next-80B-A3B-Instruct-W8A8"
|
|
envs:
|
|
OMP_NUM_THREADS: "10"
|
|
OMP_PROC_BIND: "false"
|
|
HCCL_BUFFSIZE: "1024"
|
|
SERVER_PORT: "DEFAULT_PORT"
|
|
server_cmd:
|
|
- "--quantization"
|
|
- "ascend"
|
|
- "--async-scheduling"
|
|
- "--no-enable-prefix-caching"
|
|
- "--data-parallel-size"
|
|
- "1"
|
|
- "--tensor-parallel-size"
|
|
- "4"
|
|
- "--enable-expert-parallel"
|
|
- "--port"
|
|
- "$SERVER_PORT"
|
|
- "--max-model-len"
|
|
- "40960"
|
|
- "--max-num-batched-tokens"
|
|
- "8192"
|
|
- "--max-num-seqs"
|
|
- "32"
|
|
- "--trust-remote-code"
|
|
- "--gpu-memory-utilization"
|
|
- "0.65"
|
|
- "--compilation-config"
|
|
- '{"cudagraph_capture_sizes": [32]}'
|
|
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: 96
|
|
threshold: 3
|