282 lines
8.9 KiB
YAML
282 lines
8.9 KiB
YAML
# ==========================================
|
|
# ACTUAL TEST CASES
|
|
# ==========================================
|
|
|
|
_envs: &envs
|
|
PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
|
|
HCCL_OP_EXPANSION_MODE: "AIV"
|
|
HCCL_BUFFSIZE: "1024"
|
|
OMP_NUM_THREADS: "1"
|
|
TASK_QUEUE_ENABLE: "1"
|
|
VLLM_ASCEND_ENABLE_FUSED_MC2: "1"
|
|
SERVER_PORT: "DEFAULT_PORT"
|
|
|
|
_server_cmd: &server_cmd
|
|
- "--port"
|
|
- "$SERVER_PORT"
|
|
- "--data-parallel-size"
|
|
- "1"
|
|
- "--tensor-parallel-size"
|
|
- "4"
|
|
- "--max-num-seqs"
|
|
- "128"
|
|
- "--gpu-memory-utilization"
|
|
- "0.9"
|
|
- "--no-enable-prefix-caching"
|
|
- "--trust-remote-code"
|
|
- "--allowed-local-media-path"
|
|
- "/"
|
|
- "--quantization"
|
|
- "ascend"
|
|
- "--mm-processor-cache-gb"
|
|
- "0"
|
|
- "--additional-config"
|
|
- '{"enable_cpu_binding": true, "enable_shared_expert_dp": true}'
|
|
- "--api-server-count"
|
|
- "1"
|
|
|
|
_benchmark_acc: &_benchmark_acc
|
|
acc_GPQA:
|
|
case_type: accuracy
|
|
dataset_path: vllm-ascend/gpqa
|
|
request_conf: vllm_api_general_chat
|
|
dataset_conf: gpqa/gpqa_gen_0_shot_cot_chat_prompt
|
|
max_out_len: 65536
|
|
batch_size: 32
|
|
baseline: 85
|
|
threshold: 5
|
|
thinking: true
|
|
temperature: 1.0
|
|
top_p: 0.95
|
|
top_k: 20
|
|
min_p: 0.0
|
|
presence_penalty: 1.5
|
|
repetition_penalty: 1.0
|
|
ignore_eos: false
|
|
acc_aime2025:
|
|
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: 65536
|
|
batch_size: 32
|
|
baseline: 90
|
|
threshold: 10
|
|
thinking: true
|
|
temperature: 1.0
|
|
top_p: 0.95
|
|
top_k: 20
|
|
min_p: 0.0
|
|
presence_penalty: 1.5
|
|
repetition_penalty: 1.0
|
|
ignore_eos: false
|
|
|
|
_benchmark_TPOT50_3_5k_1_5K: &_benchmark_TPOT50_3_5k_1_5K
|
|
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: 320
|
|
max_out_len: 1500
|
|
batch_size: 80
|
|
request_rate: 0
|
|
baseline: 1052.9
|
|
threshold: 0.95
|
|
|
|
_benchmark_TPOT50_16K_1K: &_benchmark_TPOT50_16K_1K
|
|
case_type: performance
|
|
dataset_path: vllm-ascend/GSM8K_prefix0_in16384_bs200_qwen
|
|
request_conf: vllm_api_stream_chat
|
|
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
|
|
num_prompts: 96
|
|
max_out_len: 1024
|
|
batch_size: 24
|
|
request_rate: 0
|
|
baseline: 400.183
|
|
threshold: 0.95
|
|
|
|
_benchmark_TPOT20_16K_1K: &_benchmark_TPOT20_16K_1K
|
|
case_type: performance
|
|
dataset_path: vllm-ascend/GSM8K_prefix0_in16384_bs200_qwen
|
|
request_conf: vllm_api_stream_chat
|
|
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
|
|
num_prompts: 32
|
|
max_out_len: 1024
|
|
batch_size: 8
|
|
request_rate: 0
|
|
baseline: 325.5
|
|
threshold: 0.95
|
|
|
|
_benchmark_TPOT50_32K_0_5K: &_benchmark_TPOT50_32K_0_5K
|
|
case_type: performance
|
|
dataset_path: vllm-ascend/GSM8K_prefix0_in32768_bs100_qwen
|
|
request_conf: vllm_api_stream_chat
|
|
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
|
|
num_prompts: 36
|
|
max_out_len: 512
|
|
batch_size: 9
|
|
request_rate: 0
|
|
baseline: 123.8
|
|
threshold: 0.95
|
|
|
|
_benchmark_TPOT20_32K_0_5K: &_benchmark_TPOT20_32K_0_5K
|
|
case_type: performance
|
|
dataset_path: vllm-ascend/GSM8K_prefix0_in32768_bs100_qwen
|
|
request_conf: vllm_api_stream_chat
|
|
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
|
|
num_prompts: 16
|
|
max_out_len: 512
|
|
batch_size: 4
|
|
request_rate: 0
|
|
baseline: 113.8
|
|
threshold: 0.95
|
|
|
|
_benchmark_TPOT50_64K_1K: &_benchmark_TPOT50_64K_1K
|
|
case_type: performance
|
|
dataset_path: vllm-ascend/GSM8K_prefix0_in65536_bs100_qwen
|
|
request_conf: vllm_api_stream_chat
|
|
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
|
|
num_prompts: 32
|
|
max_out_len: 1024
|
|
batch_size: 8
|
|
request_rate: 0
|
|
baseline: 114.1
|
|
threshold: 0.95
|
|
|
|
_benchmark_TPOT20_64K_1K: &_benchmark_TPOT20_64K_1K
|
|
case_type: performance
|
|
dataset_path: vllm-ascend/GSM8K_prefix0_in65536_bs100_qwen
|
|
request_conf: vllm_api_stream_chat
|
|
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
|
|
num_prompts: 16
|
|
max_out_len: 1024
|
|
batch_size: 4
|
|
request_rate: 0
|
|
baseline: 99.28
|
|
threshold: 0.95
|
|
|
|
|
|
test_cases:
|
|
- name: "Qwen3.5-122B-A10B-W8A8-A3"
|
|
model: "Eco-Tech/Qwen3.5-122B-A10B-w8a8-mtp"
|
|
envs:
|
|
<<: *envs
|
|
VLLM_ASCEND_ENABLE_FLASHCOMM1: "1"
|
|
server_cmd: *server_cmd
|
|
server_cmd_extra:
|
|
- "--enable-expert-parallel"
|
|
- "--max-model-len"
|
|
- "8000"
|
|
- "--max-num-batched-tokens"
|
|
- "8000"
|
|
- "--compilation-config"
|
|
- '{"cudagraph_capture_sizes":[1,4,8,12,16,24,32,48,56,64,72,84,96,108,112,128,160,172,196,200,212,232,256,160,172,196,200,212,232,256,272,288,312,328,344,360,384,400,416,432,448,480,512], "cudagraph_mode":"FULL_DECODE_ONLY"}'
|
|
- "--speculative_config"
|
|
- '{"method": "qwen3_5_mtp", "num_speculative_tokens": 3, "enforce_eager": true}'
|
|
benchmarks:
|
|
perf: *_benchmark_TPOT50_3_5k_1_5K
|
|
- name: "Qwen3.5-122B-A10B-W8A8-TPOT50-16k-1k"
|
|
model: "Eco-Tech/Qwen3.5-122B-A10B-w8a8-mtp"
|
|
envs:
|
|
<<: *envs
|
|
VLLM_ASCEND_ENABLE_FLASHCOMM1: "1"
|
|
server_cmd: *server_cmd
|
|
server_cmd_extra:
|
|
- "--enable-expert-parallel"
|
|
- "--max-model-len"
|
|
- "18432"
|
|
- "--max-num-batched-tokens"
|
|
- "16384"
|
|
- "--compilation-config"
|
|
- '{"cudagraph_capture_sizes":[1,4,8,12,16,24,32,48,56,64,72,84,96,108,112,128,160,172,196,200,212,232,256,160,172,196,200,212,232,256,272,288,312,328,344,360,384,400,416,432,448,480,512], "cudagraph_mode":"FULL_DECODE_ONLY"}'
|
|
- "--speculative_config"
|
|
- '{"method": "qwen3_5_mtp", "num_speculative_tokens": 3, "enforce_eager": true}'
|
|
benchmarks:
|
|
perf: *_benchmark_TPOT50_16K_1K
|
|
- name: "Qwen3.5-122B-A10B-W8A8-TPOT20-16k-1k"
|
|
model: "Eco-Tech/Qwen3.5-122B-A10B-w8a8-mtp"
|
|
envs:
|
|
<<: *envs
|
|
server_cmd: *server_cmd
|
|
server_cmd_extra:
|
|
- "--max-model-len"
|
|
- "18432"
|
|
- "--max-num-batched-tokens"
|
|
- "16384"
|
|
- "--compilation-config"
|
|
- '{"cudagraph_capture_sizes":[1,6,12,18,24,30,36,42,48,54,72,78,84,90,96,102,108,144,192], "cudagraph_mode":"FULL_DECODE_ONLY"}'
|
|
- "--speculative-config"
|
|
- '{"method": "qwen3_5_mtp", "num_speculative_tokens": 5}'
|
|
benchmarks:
|
|
perf: *_benchmark_TPOT20_16K_1K
|
|
- name: "Qwen3.5-122B-A10B-W8A8-TPOT50-32k-0.5k"
|
|
model: "Eco-Tech/Qwen3.5-122B-A10B-w8a8-mtp"
|
|
envs:
|
|
<<: *envs
|
|
VLLM_ASCEND_ENABLE_FLASHCOMM1: "1"
|
|
server_cmd: *server_cmd
|
|
server_cmd_extra:
|
|
- "--enable-expert-parallel"
|
|
- "--max-model-len"
|
|
- "34304"
|
|
- "--max-num-batched-tokens"
|
|
- "16384"
|
|
- "--compilation-config"
|
|
- '{"cudagraph_capture_sizes":[1,4,8,12,16,24,32,48,56,64,72,84,96,108,112,128,160,172,196,200,212,232,256,160,172,196,200,212,232,256,272,288,312,328,344,360,384,400,416,432,448,480,512], "cudagraph_mode":"FULL_DECODE_ONLY"}'
|
|
- "--speculative-config"
|
|
- '{"method": "qwen3_5_mtp", "num_speculative_tokens": 3, "enforce_eager": true}'
|
|
benchmarks:
|
|
perf: *_benchmark_TPOT50_32K_0_5K
|
|
- name: "Qwen3.5-122B-A10B-W8A8-TPOT20-32k-0.5k"
|
|
model: "Eco-Tech/Qwen3.5-122B-A10B-w8a8-mtp"
|
|
envs:
|
|
<<: *envs
|
|
server_cmd: *server_cmd
|
|
server_cmd_extra:
|
|
- "--max-model-len"
|
|
- "34304"
|
|
- "--max-num-batched-tokens"
|
|
- "16384"
|
|
- "--compilation-config"
|
|
- '{"cudagraph_capture_sizes":[1,4,8,12,16,24,32,48,56,64,72,84,96,108,112,128,160,172,196,200,212,232,256,160,172,196,200,212,232,256,272,288,312,328,344,360,384,400,416,432,448,480,512], "cudagraph_mode":"FULL_DECODE_ONLY"}'
|
|
- "--speculative-config"
|
|
- '{"method": "qwen3_5_mtp", "num_speculative_tokens": 3, "enforce_eager": true}'
|
|
benchmarks:
|
|
perf: *_benchmark_TPOT20_32K_0_5K
|
|
- name: "Qwen3.5-122B-A10B-W8A8-TPOT50-64k-1k"
|
|
model: "Eco-Tech/Qwen3.5-122B-A10B-w8a8-mtp"
|
|
envs:
|
|
<<: *envs
|
|
VLLM_ASCEND_ENABLE_FLASHCOMM1: "1"
|
|
server_cmd: *server_cmd
|
|
server_cmd_extra:
|
|
- "--enable-expert-parallel"
|
|
- "--max-model-len"
|
|
- "67584"
|
|
- "--max-num-batched-tokens"
|
|
- "16384"
|
|
- "--compilation-config"
|
|
- '{"cudagraph_capture_sizes":[1,4,8,12,16,24,32,48,56,64,72,84,96,108,112,128,160,172,196,200,212,232,256,160,172,196,200,212,232,256,272,288,312,328,344,360,384,400,416,432,448,480,512], "cudagraph_mode":"FULL_DECODE_ONLY"}'
|
|
- "--speculative-config"
|
|
- '{"method": "qwen3_5_mtp", "num_speculative_tokens": 3, "enforce_eager": true}'
|
|
benchmarks:
|
|
perf: *_benchmark_TPOT50_64K_1K
|
|
<<: *_benchmark_acc
|
|
- name: "Qwen3.5-122B-A10B-W8A8-TPOT20-64k-1k"
|
|
model: "Eco-Tech/Qwen3.5-122B-A10B-w8a8-mtp"
|
|
envs:
|
|
<<: *envs
|
|
VLLM_ASCEND_ENABLE_FLASHCOMM1: "0"
|
|
server_cmd: *server_cmd
|
|
server_cmd_extra:
|
|
- "--max-model-len"
|
|
- "67584"
|
|
- "--max-num-batched-tokens"
|
|
- "16384"
|
|
- "--compilation-config"
|
|
- '{"cudagraph_capture_sizes":[1,6,12,18,24,30,36,42,48,54,72,78,84,90,96,102,108,144,192], "cudagraph_mode":"FULL_DECODE_ONLY"}'
|
|
- "--speculative-config"
|
|
- '{"method": "qwen3_5_mtp", "num_speculative_tokens": 5}'
|
|
benchmarks:
|
|
perf: *_benchmark_TPOT20_64K_1K
|