Why does the accuracy decrease as the concurrency increases?

General
1

==============================
System Info

OS : Debian GNU/Linux 12 (bookworm) (x86_64)
GCC version : (Debian 12.2.0-14) 12.2.0
Clang version : Could not collect
CMake version : version 3.25.1
Libc version : glibc-2.36

==============================
PyTorch Info

PyTorch version : 2.7.0+cu126
Is debug build : False
CUDA used to build PyTorch : 12.6
ROCM used to build PyTorch : N/A

==============================
Python Environment

Python version : 3.10.15 (main, Nov 12 2024, 02:24:06) [GCC 12.2.0] (64-bit runtime)
Python platform : Linux-3.10.0-1160.119.1.el7.x86_64-x86_64-with-glibc2.36

==============================
CUDA / GPU Info

Is CUDA available : True
CUDA runtime version : 12.1.66
CUDA_MODULE_LOADING set to : LAZY
GPU models and configuration :
GPU 0: NVIDIA RTX A6000
GPU 1: NVIDIA RTX A6000
GPU 2: NVIDIA RTX A6000
GPU 3: NVIDIA RTX A6000

Nvidia driver version : 550.78
cuDNN version : Could not collect
HIP runtime version : N/A
MIOpen runtime version : N/A
Is XNNPACK available : True

==============================
CPU Info

Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
Address sizes: 46 bits physical, 57 bits virtual
Byte Order: Little Endian
CPU(s): 64
On-line CPU(s) list: 0-63
Vendor ID: GenuineIntel
Model name: Intel(R) Xeon(R) Silver 4314 CPU @ 2.40GHz
CPU family: 6
Model: 106
Thread(s) per core: 2
Core(s) per socket: 16
Socket(s): 2
Stepping: 6
Frequency boost: enabled
CPU(s) scaling MHz: 37%
CPU max MHz: 2401.0000
CPU min MHz: 800.0000
BogoMIPS: 4800.00
Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm constant_tsc art arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc aperfmperf eagerfpu pni pclmulqdq dtes64 ds_cpl vmx smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid dca sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch epb cat_l3 invpcid_single ssbd mba rsb_ctxsw ibrs ibpb stibp ibrs_enhanced tpr_shadow vnmi flexpriority ept vpid fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm cqm rdt_a avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb intel_pt avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local dtherm ida arat pln pts avx512vbmi umip pku ospke avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg avx512_vpopcntdq md_clear pconfig spec_ctrl intel_stibp flush_l1d arch_capabilities
Virtualization: VT-x
L1d cache: 1.5 MiB (32 instances)
L1i cache: 1 MiB (32 instances)
L2 cache: 40 MiB (32 instances)
L3 cache: 48 MiB (2 instances)
NUMA node(s): 2
NUMA node0 CPU(s): 0-15,32-47
NUMA node1 CPU(s): 16-31,48-63
Vulnerability Gather data sampling: Mitigation; Microcode
Vulnerability Itlb multihit: Not affected
Vulnerability L1tf: Not affected
Vulnerability Mds: Not affected
Vulnerability Meltdown: Not affected
Vulnerability Mmio stale data: Mitigation; Clear CPU buffers; SMT vulnerable
Vulnerability Retbleed: Not affected
Vulnerability Spec store bypass: Mitigation; Speculative Store Bypass disabled via prctl and seccomp
Vulnerability Spectre v1: Mitigation; Load fences, usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2: Mitigation; Enhanced IBRS, IBPB
Vulnerability Srbds: Not affected
Vulnerability Tsx async abort: Not affected

==============================
Versions of relevant libraries

[pip3] fast_pytorch_kmeans==0.2.2
[pip3] mypy==1.15.0
[pip3] mypy-extensions==1.0.0
[pip3] numpy==1.26.4
[pip3] nvidia-cublas-cu11==11.11.3.6
[pip3] nvidia-cublas-cu12==12.6.4.1
[pip3] nvidia-cuda-cupti-cu11==11.8.87
[pip3] nvidia-cuda-cupti-cu12==12.6.80
[pip3] nvidia-cuda-nvrtc-cu11==11.8.89
[pip3] nvidia-cuda-nvrtc-cu12==12.6.77
[pip3] nvidia-cuda-runtime-cu11==11.8.89
[pip3] nvidia-cuda-runtime-cu12==12.6.77
[pip3] nvidia-cudnn-cu11==8.7.0.84
[pip3] nvidia-cudnn-cu12==9.5.1.17
[pip3] nvidia-cufft-cu11==10.9.0.58
[pip3] nvidia-cufft-cu12==11.3.0.4
[pip3] nvidia-cufile-cu12==1.11.1.6
[pip3] nvidia-curand-cu11==10.3.0.86
[pip3] nvidia-curand-cu12==10.3.7.77
[pip3] nvidia-cusolver-cu11==11.4.1.48
[pip3] nvidia-cusolver-cu12==11.7.1.2
[pip3] nvidia-cusparse-cu11==11.7.5.86
[pip3] nvidia-cusparse-cu12==12.5.4.2
[pip3] nvidia-cusparselt-cu12==0.6.3
[pip3] nvidia-ml-py==12.570.86
[pip3] nvidia-modelopt==0.29.0
[pip3] nvidia-modelopt-core==0.29.0
[pip3] nvidia-nccl-cu11==2.19.3
[pip3] nvidia-nccl-cu12==2.26.2
[pip3] nvidia-nvjitlink-cu12==12.6.85
[pip3] nvidia-nvtx-cu11==11.8.86
[pip3] nvidia-nvtx-cu12==12.6.77
[pip3] onnx==1.18.0
[pip3] onnx_graphsurgeon==0.5.8
[pip3] pynvml==12.0.0
[pip3] pyzmq==26.4.0
[pip3] torch==2.7.0
[pip3] torchac_cuda==0.2.5
[pip3] torchaudio==2.7.0
[pip3] torchprofile==0.0.4
[pip3] torchvision==0.22.0
[pip3] transformers==4.51.3
[pip3] triton==3.3.0
[conda] Could not collect

==============================
vLLM Info

ROCM Version : Could not collect
Neuron SDK Version : N/A
vLLM Version : 0.9.1
vLLM Build Flags:
CUDA Archs: 8.6; ROCm: Disabled; Neuron: Disabled
GPU Topology:
GPU0 GPU1 GPU2 GPU3 CPU Affinity NUMA Affinity GPU NUMA ID
GPU0 X PIX PIX PIX 0-15,32-47 0 N/A
GPU1 PIX X PIX PIX 0-15,32-47 0 N/A
GPU2 PIX PIX X PIX 0-15,32-47 0 N/A
GPU3 PIX PIX PIX X 0-15,32-47 0 N/A

Legend:

X = Self
SYS = Connection traversing PCIe as well as the SMP interconnect between NUMA nodes (e.g., QPI/UPI)
NODE = Connection traversing PCIe as well as the interconnect between PCIe Host Bridges within a NUMA node
PHB = Connection traversing PCIe as well as a PCIe Host Bridge (typically the CPU)
PXB = Connection traversing multiple PCIe bridges (without traversing the PCIe Host Bridge)
PIX = Connection traversing at most a single PCIe bridge
NV# = Connection traversing a bonded set of # NVLinks

==============================
Environment Variables

TORCH_CUDA_ARCH_LIST=8.6
LD_LIBRARY_PATH=/usr/local/cuda/lib64:
NCCL_CUMEM_ENABLE=1
PYTORCH_NVML_BASED_CUDA_CHECK=1
TORCHINDUCTOR_COMPILE_THREADS=1
TORCHINDUCTOR_CACHE_DIR=/tmp/torchinductor_root
VLLM_WORKER_MULTIPROC_METHOD=spawn
CUDA_MODULE_LOADING=LAZY

GitHub address: learning_project/benchmarks at main · moguizhizi/learning_project · GitHub

start serving model：

vllm serve /home/temp/llm_model/nm-testing/Qwen2.5-VL-72B-Instruct-quantized.w8a8 --load-format runai_streamer --tensor-parallel-size 4 --max-model-len 8192 --max-num-seqs 2048 --kv-cache auto --gpu-memory-utilization 0.5 --disable-custom-all-reduce --served-model-name qwen2_5_vl_72B_quant --disable-log-requests

Send request:

python3 learning_project/benchmarks/benchmark_serving.py --backend openai-chat --model /home/temp/llm_model/nm-testing/Qwen2___5-VL-72B-Instruct-quantized___w8a8 --served-model-name qwen2_5_vl_72B_quant --endpoint /v1/chat/completions --dataset-name phonetest --dataset-path /home/project/dataset/phonetest/web_nj_action_0426_grpo.json --num-prompts 128 --result_dir /home/project/learning_project/benchmarks/output/qwen2_5_vl_72B --save-result

llm model url：

accuracy_vs_num_prompts
accuracy_vs_num_prompts1000×600 31 KB

2

You are running vLLM with a 72B quantized Qwen2.5-VL model using Run:ai Model Streamer and tensor parallelism across 4 NVIDIA RTX A6000 GPUs. Your command and setup are correct for loading large quantized models with streaming, and your benchmarking script is compatible with this setup. The use of --load-format runai_streamer is supported and recommended for large models, and you can further tune loading performance with --model-loader-extra-config parameters like concurrency and memory_limit if needed. For sharded models, use --load-format runai_streamer_sharded and adjust the file pattern as required. See the documentation for more details and tuning options[^1][^2][^3].

If you are experiencing issues such as worker crashes or throughput drops, these may be related to GPU memory limits, concurrency settings, or model quantization compatibility. The logs and configuration you provided are consistent with supported vLLM usage, but if you encounter errors like “Worker proc VllmWorker-2 died unexpectedly,” consider increasing gpu-memory-utilization, reducing max-model-len or max-num-seqs, or tuning Run:ai streamer parameters. There are no unsupported commands or APIs in your workflow, and your hardware/software stack is compatible with vLLM 0.9.1[^1][^2][^3][^4]. Would you like more detail on troubleshooting or performance tuning?

Sources: