Qwen3-32B
Qwen3-32B
이 페이지는 Ascend NPU에서 Qwen3-32B의 최적 구성과 벤치마크 결과에 집중해요. 환경 설정, 모델 가중치 다운로드, 기능 구성, 배포 지침 등은 Qwen3-32B 모델 튜토리얼을 참고하세요.
A3 시리즈에서는 각 카드에 2개의 die가 있어서 --tp-size가 카드 수의 2배예요. 자세한 내용은 Ascend NPU 참조를 확인하세요.
출처: 문서
본문
저지연 (Low Latency)
| 모델 | 하드웨어 | 카드 | 배포 모드 | 데이터셋 | TPOT | 양자화 | 구성 |
|---|---|---|---|---|---|---|---|
| Qwen3-32B | Ascend A3 Series Products | 8 | PD Mixed | 18k+4k | 6ms | BF16 | 최적 구성 |
고처리량 (High Throughput)
| 모델 | 하드웨어 | 카드 | 배포 모드 | 데이터셋 | TPOT | 양자화 | 구성 |
|---|---|---|---|---|---|---|---|
| Qwen3-32B | Ascend A3 Series Products | 2 | PD Mixed | 3.5k+1.5k | 50ms | W8A8 INT8 | 최적 구성 |
| Qwen3-32B | Ascend A2 Series Products | 2 | PD Mixed | 3.5k+1.5k | 55ms | W8A8 INT8 | 최적 구성 |
최적 구성 (Optimal Configuration)
Qwen3-32B BF16 8P IN18K OUT4K 6ms
모델: Qwen3-32B
하드웨어: Ascend A3 Series Products
카드: 8
배포 모드: PD Mixed
양자화: BF16
데이터셋: 18k+4k
TPOT: 6ms
모델 배포 (Model Deployment)
# ============================================================
# Before running, update the following variables:
# MODEL_PATH: path to the model weights directory
# DRAFT_MODEL_PATH: path to the draft model weights directory
# HCCL_SOCKET_IFNAME: network interface name for HCCL
# GLOO_SOCKET_IFNAME: network interface name for Gloo
# ============================================================
MODEL_PATH=/path/to/model-weights
DRAFT_MODEL_PATH=/path/to/draft-model-weights
echo performance | tee /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor
sysctl -w vm.swappiness=0
sysctl -w kernel.numa_balancing=0
sysctl -w kernel.sched_migration_cost_ns=50000
unset https_proxy
unset http_proxy
unset HTTPS_PROXY
unset HTTP_PROXY
unset ASCEND_LAUNCH_BLOCKING
source /usr/local/Ascend/ascend-toolkit/set_env.sh
source /usr/local/Ascend/nnal/atb/set_env.sh
export GLOO_SOCKET_IFNAME=<network-interface>
export HCCL_OP_EXPANSION_MODE=AIV
export HCCL_SOCKET_IFNAME=<network-interface>
export PYTORCH_NPU_ALLOC_CONF=expandable_segments:True
export SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT=600
export SGLANG_ENABLE_OVERLAP_PLAN_STREAM=1
export SGLANG_PREFILL_DELAYER_MAX_DELAY_PASSES=200
export SGLANG_SCHEDULER_DECREASE_PREFILL_IDLE=1
python3 -m sglang.launch_server \
--model-path $MODEL_PATH \
--host 127.0.0.1 --port 6688 \
--trust-remote-code \
--nnodes 1 \
--node-rank 0 \
--attention-backend ascend \
--device npu \
--max-running-requests 1 \
--disable-radix-cache \
--speculative-draft-model-quantization unquant \
--chunked-prefill-size -1 \
--max-prefill-tokens 65536 \
--speculative-algorithm EAGLE3 \
--speculative-draft-model-path $DRAFT_MODEL_PATH \
--speculative-num-steps 4 \
--speculative-eagle-topk 1 \
--speculative-num-draft-tokens 5 \
--tp-size 16 \
--mem-fraction-static 0.72 \
--cuda-graph-bs-decode 1 \
--dtype bfloat16 \
--reasoning-parser qwen3 \
--tool-call-parser qwen
벤치마크 (Benchmark)
RANDOM 데이터셋을 기반으로 테스트했어요.
python -m sglang.bench_serving \
--dataset-name random \
--backend sglang \
--host 127.0.0.1 \
--port 6688 \
--max-concurrency 1 \
--num-prompts 1 \
--random-input-len 18000 \
--random-output-len 4000 \
--random-range-ratio 1 \
--seed 1
Qwen3-32B W8A8 2P IN3K5 OUT1K5 50ms
모델: Qwen3-32B
하드웨어: Ascend A3 Series Products
카드: 2
배포 모드: PD Mixed
양자화: W8A8 INT8
데이터셋: 3.5k+1.5k
TPOT: 50ms
모델 배포 (Model Deployment)
# ============================================================
# Before running, update the following variables:
# MODEL_PATH: path to the model weights directory
# DRAFT_MODEL_PATH: path to the draft model weights directory
# HCCL_SOCKET_IFNAME: network interface name for HCCL
# GLOO_SOCKET_IFNAME: network interface name for Gloo
# ============================================================
MODEL_PATH=/path/to/model-weights
DRAFT_MODEL_PATH=/path/to/draft-model-weights
echo performance | tee /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor
sysctl -w vm.swappiness=0
sysctl -w kernel.numa_balancing=0
sysctl -w kernel.sched_migration_cost_ns=50000
unset https_proxy
unset http_proxy
unset HTTPS_PROXY
unset HTTP_PROXY
unset ASCEND_LAUNCH_BLOCKING
source /usr/local/Ascend/ascend-toolkit/set_env.sh
source /usr/local/Ascend/nnal/atb/set_env.sh
export GLOO_SOCKET_IFNAME=<network-interface>
export HCCL_OP_EXPANSION_MODE=AIV
export HCCL_SOCKET_IFNAME=<network-interface>
export PYTORCH_NPU_ALLOC_CONF=expandable_segments:True
export SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT=600
export SGLANG_ENABLE_OVERLAP_PLAN_STREAM=1
export SGLANG_NPU_USE_DEEPGEMM=1
export SGLANG_PREFILL_DELAYER_MAX_DELAY_PASSES=100
export SGLANG_SCHEDULER_DECREASE_PREFILL_IDLE=1
python3 -m sglang.launch_server \
--model-path $MODEL_PATH \
--host 127.0.0.1 --port 6688 \
--trust-remote-code \
--nnodes 1 \
--node-rank 0 \
--attention-backend ascend \
--device npu \
--quantization modelslim \
--max-running-requests 101 \
--disable-radix-cache \
--speculative-draft-model-quantization unquant \
--chunked-prefill-size -1 \
--max-prefill-tokens 35000 \
--speculative-algorithm EAGLE3 \
--speculative-draft-model-path $DRAFT_MODEL_PATH \
--speculative-num-steps 3 \
--speculative-eagle-topk 1 \
--speculative-num-draft-tokens 4 \
--tp-size 4 \
--mem-fraction-static 0.845 \
--cuda-graph-bs-decode 16 32 64 72 88 90 92 94 96 97 98 99 100 101 \
--dtype bfloat16 \
--reasoning-parser qwen3 \
--tool-call-parser qwen
벤치마크 (Benchmark)
RANDOM 데이터셋을 기반으로 테스트했어요.
python -m sglang.bench_serving \
--dataset-name random \
--backend sglang \
--host 127.0.0.1 \
--port 6688 \
--max-concurrency 100 \
--num-prompts 400 \
--random-input-len 3584 \
--random-output-len 1536 \
--random-range-ratio 1 \
--seed 1
Qwen3-32B W8A8 2P IN3K5 OUT1K5 55ms A2 Series
모델: Qwen3-32B
하드웨어: Ascend A2 Series Products
카드: 2
배포 모드: PD Mixed
양자화: W8A8 INT8
데이터셋: 3.5k+1.5k
TPOT: 55ms
모델 배포 (Model Deployment)
# ============================================================
# Before running, update the following variables:
# MODEL_PATH: path to the model weights directory
# DRAFT_MODEL_PATH: path to the draft model weights directory
# HCCL_SOCKET_IFNAME: network interface name for HCCL
# GLOO_SOCKET_IFNAME: network interface name for Gloo
# ============================================================
MODEL_PATH=/path/to/model-weights
DRAFT_MODEL_PATH=/path/to/draft-model-weights
echo performance | tee /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor
sysctl -w vm.swappiness=0
sysctl -w kernel.numa_balancing=0
sysctl -w kernel.sched_migration_cost_ns=50000
unset https_proxy
unset http_proxy
unset HTTPS_PROXY
unset HTTP_PROXY
unset ASCEND_LAUNCH_BLOCKING
source /usr/local/Ascend/ascend-toolkit/set_env.sh
source /usr/local/Ascend/nnal/atb/set_env.sh
export GLOO_SOCKET_IFNAME=<network-interface>
export HCCL_OP_EXPANSION_MODE=AIV
export HCCL_SOCKET_IFNAME=<network-interface>
export PYTORCH_NPU_ALLOC_CONF=expandable_segments:True
export SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT=600
export SGLANG_ENABLE_OVERLAP_PLAN_STREAM=1
export SGLANG_NPU_USE_DEEPGEMM=1
export SGLANG_PREFILL_DELAYER_MAX_DELAY_PASSES=100
export SGLANG_SCHEDULER_DECREASE_PREFILL_IDLE=1
python3 -m sglang.launch_server \
--model-path $MODEL_PATH \
--host 127.0.0.1 --port 6688 \
--trust-remote-code \
--nnodes 1 \
--node-rank 0 \
--attention-backend ascend \
--device npu \
--quantization modelslim \
--max-running-requests 101 \
--disable-radix-cache \
--speculative-draft-model-quantization unquant \
--chunked-prefill-size -1 \
--max-prefill-tokens 35000 \
--speculative-algorithm EAGLE3 \
--speculative-draft-model-path $DRAFT_MODEL_PATH \
--speculative-num-steps 3 \
--speculative-eagle-topk 1 \
--speculative-num-draft-tokens 4 \
--tp-size 4 \
--mem-fraction-static 0.845 \
--cuda-graph-bs-decode 16 32 64 72 88 90 92 94 96 97 98 99 100 101 \
--dtype bfloat16 \
--reasoning-parser qwen3 \
--tool-call-parser qwen
벤치마크 (Benchmark)
RANDOM 데이터셋을 기반으로 테스트했어요.
python -m sglang.bench_serving \
--dataset-name random \
--backend sglang \
--host 127.0.0.1 \
--port 6688 \
--max-concurrency 100 \
--num-prompts 400 \
--random-input-len 3584 \
--random-output-len 1536 \
--random-range-ratio 1 \
--seed 1