67 lines
2.0 KiB
YAML
67 lines
2.0 KiB
YAML
# ==========================================
|
|
# ACTUAL TEST CASES
|
|
# ==========================================
|
|
test_cases:
|
|
- name: "Qwen3.6-35B-A3B"
|
|
model: "Eco-Tech/Qwen3.6-35B-A3B-w8a8"
|
|
envs:
|
|
PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
|
|
HCCL_OP_EXPANSION_MODE: "AIV"
|
|
OMP_PROC_BIND: "false"
|
|
OMP_NUM_THREADS: "1"
|
|
TASK_QUEUE_ENABLE: "1"
|
|
HCCL_BUFFSIZE: "1024"
|
|
VLLM_ASCEND_ENABLE_FLASHCOMM1: "1"
|
|
VLLM_ASCEND_BALANCE_SCHEDULING: "1"
|
|
SERVER_PORT: "DEFAULT_PORT"
|
|
server_cmd:
|
|
- "--quantization"
|
|
- "ascend"
|
|
- "--allowed-local-media-path"
|
|
- "/"
|
|
- "--trust-remote-code"
|
|
- "--tensor-parallel-size"
|
|
- "2"
|
|
- "--enable-expert-parallel"
|
|
- "--port"
|
|
- "$SERVER_PORT"
|
|
- "--max-num-seqs"
|
|
- "50"
|
|
- "--max-model-len"
|
|
- "65536"
|
|
- "--max-num-batched-tokens"
|
|
- "8192"
|
|
- "--gpu-memory-utilization"
|
|
- "0.9"
|
|
- "--seed"
|
|
- "1024"
|
|
- "--no-enable-prefix-caching"
|
|
- "--async-scheduling"
|
|
- "--compilation-config"
|
|
- '{"cudagraph_capture_sizes":[1,8,16,24,28,40,52,64,76,88,96,100,112,120,132,144,160,176,184,192,196,200,204,208,212,216,220],"cudagraph_mode":"FULL_DECODE_ONLY"}'
|
|
- "--speculative-config"
|
|
- '{"method": "qwen3_5_mtp","num_speculative_tokens": 3}'
|
|
- "--additional-config"
|
|
- '{"enable_cpu_binding":true}'
|
|
- "--profiler-config"
|
|
- '{"profiler": "torch","torch_profiler_dir": "./profiling","torch_profiler_with_stack": false}'
|
|
- "--mm-processor-cache-gb"
|
|
- "0"
|
|
- "--mm-encoder-tp-mode"
|
|
- "data"
|
|
benchmarks:
|
|
acc:
|
|
case_type: accuracy
|
|
dataset_path: vllm-ascend/aime2025
|
|
request_conf: vllm_api_general_chat
|
|
dataset_conf: aime2025/aime2025_gen_0_shot_chat_prompt
|
|
max_out_len: 32768
|
|
batch_size: 32
|
|
baseline: 90
|
|
threshold: 10
|
|
temperature: 1.0
|
|
top_p: 0.95
|
|
top_k: 20
|
|
min_p: 0.0
|
|
presence_penalty: 1.5
|
|
repetition_penalty: 1.0 |