34386c8896
### What this PR does / why we need it? This PR fixes and simplifies the CI configuration for Qwen3 32B. The main changes are: - Remove the redundant `Qwen3-32B-Int8-A3-Feature-Stack3.yaml` config and consolidate the CI setup into `Qwen3-32B-Int8.yaml`. - Improve runtime stability by adding `PYTORCH_NPU_ALLOC_CONF=expandable_segments:True` and setting `--max-num-seqs 80`. - Update the accuracy benchmark from `aime2024` to `gsm8k-lite`, and adjust the related dataset config, output length, baseline, and threshold accordingly. These changes make the Qwen3 32B CI easier to maintain and more stable in nightly validation. --------- Signed-off-by: ZYang6263 <zy626375@gmail.com>
80 lines
1.9 KiB
YAML
80 lines
1.9 KiB
YAML
# ==========================================
|
|
# Shared Configurations
|
|
# ==========================================
|
|
|
|
_envs: &envs
|
|
TASK_QUEUE_ENABLE: "1"
|
|
PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
|
|
HCCL_OP_EXPANSION_MODE: "AIV"
|
|
VLLM_ASCEND_ENABLE_FLASHCOMM: "1"
|
|
SERVER_PORT: "DEFAULT_PORT"
|
|
|
|
_server_cmd: &server_cmd
|
|
- "--quantization"
|
|
- "ascend"
|
|
- "--no-enable-prefix-caching"
|
|
- "--tensor-parallel-size"
|
|
- "4"
|
|
- "--max-num-seqs"
|
|
- "80"
|
|
- "--port"
|
|
- "$SERVER_PORT"
|
|
- "--max-model-len"
|
|
- "40960"
|
|
- "--max-num-batched-tokens"
|
|
- "40960"
|
|
- "--block-size"
|
|
- "128"
|
|
- "--trust-remote-code"
|
|
- "--gpu-memory-utilization"
|
|
- "0.9"
|
|
- "--async-scheduling"
|
|
- "--additional-config"
|
|
- '{"weight_prefetch_config":{"enabled":true}}'
|
|
|
|
_benchmarks: &benchmarks
|
|
acc:
|
|
case_type: accuracy
|
|
dataset_path: vllm-ascend/gsm8k-lite
|
|
request_conf: vllm_api_general_chat
|
|
dataset_conf: gsm8k/gsm8k_gen_0_shot_noncot_chat_prompt
|
|
max_out_len: 10240
|
|
batch_size: 32
|
|
baseline: 96
|
|
threshold: 4
|
|
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: 304
|
|
max_out_len: 1500
|
|
batch_size: 76
|
|
baseline: 1
|
|
threshold: 0.97
|
|
|
|
# ==========================================
|
|
# ACTUAL TEST CASES
|
|
# ==========================================
|
|
|
|
test_cases:
|
|
- name: "Qwen3-32B-W8A8-aclgraph-a3"
|
|
model: "vllm-ascend/Qwen3-32B-W8A8"
|
|
envs:
|
|
<<: *envs
|
|
server_cmd: *server_cmd
|
|
server_cmd_extra:
|
|
- "--compilation-config"
|
|
- '{"cudagraph_mode":"FULL_DECODE_ONLY","cudagraph_capture_sizes":[1,12,16,20,24,32,48,60,64,68,72,76,80]}'
|
|
benchmarks:
|
|
<<: *benchmarks
|
|
|
|
- name: "Qwen3-32B-W8A8-single-a3"
|
|
model: "vllm-ascend/Qwen3-32B-W8A8"
|
|
envs:
|
|
<<: *envs
|
|
server_cmd: *server_cmd
|
|
server_cmd_extra:
|
|
- "--enforce-eager"
|
|
benchmarks:
|