86 lines
2.1 KiB
YAML
86 lines
2.1 KiB
YAML
|
# ==========================================
|
||
|
|
# Shared Configurations
|
||
|
|
# ==========================================
|
||
|
|
|
||
|
|
_envs: &envs
|
||
|
|
OMP_NUM_THREADS: "10"
|
||
|
|
OMP_PROC_BIND: "false"
|
||
|
|
HCCL_BUFFSIZE: "1024"
|
||
|
|
PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
|
||
|
|
VLLM_ASCEND_ENABLE_FLASHCOMM1: "1"
|
||
|
|
SERVER_PORT: "DEFAULT_PORT"
|
||
|
|
|
||
|
|
_server_cmd: &server_cmd
|
||
|
|
- "--quantization"
|
||
|
|
- "ascend"
|
||
|
|
- "--async-scheduling"
|
||
|
|
- "--data-parallel-size"
|
||
|
|
- "4"
|
||
|
|
- "--tensor-parallel-size"
|
||
|
|
- "4"
|
||
|
|
- "--enable-expert-parallel"
|
||
|
|
- "--port"
|
||
|
|
- "$SERVER_PORT"
|
||
|
|
- "--max-model-len"
|
||
|
|
- "40960"
|
||
|
|
- "--max-num-batched-tokens"
|
||
|
|
- "8192"
|
||
|
|
- "--max-num-seqs"
|
||
|
|
- "12"
|
||
|
|
- "--trust-remote-code"
|
||
|
|
- "--gpu-memory-utilization"
|
||
|
|
- "0.9"
|
||
|
|
|
||
|
|
_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_cot_chat_prompt
|
||
|
|
max_out_len: 32768
|
||
|
|
batch_size: 32
|
||
|
|
top_k: 20
|
||
|
|
baseline: 95
|
||
|
|
threshold: 5
|
||
|
|
|
||
|
|
# ==========================================
|
||
|
|
# ACTUAL TEST CASES
|
||
|
|
# ==========================================
|
||
|
|
|
||
|
|
test_cases:
|
||
|
|
- name: "Qwen3-235B-A22B-W8A8-full_graph"
|
||
|
|
model: "vllm-ascend/Qwen3-235B-A22B-W8A8"
|
||
|
|
envs:
|
||
|
|
<<: *envs
|
||
|
|
server_cmd: *server_cmd
|
||
|
|
server_cmd_extra:
|
||
|
|
- "--compilation-config"
|
||
|
|
- '{"cudagraph_mode": "FULL_DECODE_ONLY"}'
|
||
|
|
benchmarks:
|
||
|
|
<<: *benchmarks
|
||
|
|
|
||
|
|
- name: "Qwen3-235B-A22B-W8A8-piecewise"
|
||
|
|
model: "vllm-ascend/Qwen3-235B-A22B-W8A8"
|
||
|
|
envs:
|
||
|
|
<<: *envs
|
||
|
|
server_cmd: *server_cmd
|
||
|
|
server_cmd_extra:
|
||
|
|
- "--compilation-config"
|
||
|
|
- '{"cudagraph_mode": "PIECEWISE"}'
|
||
|
|
benchmarks:
|
||
|
|
<<: *benchmarks
|
||
|
|
|
||
|
|
- name: "Qwen3-235B-A22B-W8A8-EPLB"
|
||
|
|
model: "vllm-ascend/Qwen3-235B-A22B-W8A8"
|
||
|
|
envs:
|
||
|
|
<<: *envs
|
||
|
|
DYNAMIC_EPLB: "true"
|
||
|
|
server_cmd: *server_cmd
|
||
|
|
server_cmd_extra:
|
||
|
|
- "--additional-config"
|
||
|
|
- '{"eplb_config": {"dynamic_eplb": "true", "expert_heat_collection_interval": 600, "algorithm_execution_interval": 50, "num_redundant_experts": 16, "eplb_policy_type": 2}}'
|
||
|
|
- "--compilation-config"
|
||
|
|
- '{"cudagraph_mode": "FULL_DECODE_ONLY"}'
|
||
|
|
benchmarks:
|
||
|
|
<<: *benchmarks
|