search

vllm-project / vllm

A high-throughput and memory-efficient inference and serving engine for LLMs
https://docs.vllm.ai
Apache License 2.0
28.86k stars 4.28k forks source link

[Performance] In v0.6.2, when tp=1, TPOT becomes very slow for batch sizes of 10 or so. (not happened in v0.5.5) #9113

Open ashgold opened 2 weeks ago

ashgold commented 2 weeks ago

Your current environment

The output of `python collect_env.py` ```text Your output of `python collect_env.py` here ``` Defaulted container "kserve-container" out of: kserve-container, queue-proxy root@llm-isvc-predictor-00001-deployment-76cb455f5c-jxhzd:/vllm-workspace# vi test.py root@llm-isvc-predictor-00001-deployment-76cb455f5c-jxhzd:/vllm-workspace# python3 test.py Collecting environment information... PyTorch version: 2.4.0+cu121 Is debug build: False CUDA used to build PyTorch: 12.1 ROCM used to build PyTorch: N/A OS: Ubuntu 20.04.6 LTS (x86_64) GCC version: (Ubuntu 9.4.0-1ubuntu1~20.04.2) 9.4.0 Clang version: Could not collect CMake version: Could not collect Libc version: glibc-2.31 Python version: 3.12.6 (main, Sep 10 2024, 00:05:17) [GCC 9.4.0] (64-bit runtime) Python platform: Linux-5.15.0-25-generic-x86_64-with-glibc2.31 Is CUDA available: True CUDA runtime version: Could not collect CUDA_MODULE_LOADING set to: LAZY GPU models and configuration: GPU 0: NVIDIA H100 80GB HBM3 Nvidia driver version: 535.86.10 cuDNN version: Could not collect HIP runtime version: N/A MIOpen runtime version: N/A Is XNNPACK available: True CPU: Architecture: x86_64 CPU op-mode(s): 32-bit, 64-bit Byte Order: Little Endian Address sizes: 52 bits physical, 57 bits virtual CPU(s): 96 On-line CPU(s) list: 0-95 Thread(s) per core: 1 Core(s) per socket: 48 Socket(s): 2 NUMA node(s): 8 Vendor ID: GenuineIntel CPU family: 6 Model: 143 Model name: Intel(R) Xeon(R) Platinum 8468 Stepping: 8 CPU MHz: 800.000 CPU max MHz: 2100.0000 CPU min MHz: 800.0000 BogoMIPS: 4200.00 L1d cache: 4.5 MiB L1i cache: 3 MiB L2 cache: 192 MiB L3 cache: 210 MiB NUMA node0 CPU(s): 0-11 NUMA node1 CPU(s): 12-23 NUMA node2 CPU(s): 24-35 NUMA node3 CPU(s): 36-47 NUMA node4 CPU(s): 48-59 NUMA node5 CPU(s): 60-71 NUMA node6 CPU(s): 72-83 NUMA node7 CPU(s): 84-95 Vulnerability Itlb multihit: Not affected Vulnerability L1tf: Not affected Vulnerability Mds: Not affected Vulnerability Meltdown: Not affected Vulnerability Spec store bypass: Mitigation; Speculative Store Bypass disabled via prctl and seccomp Vulnerability Spectre v1: Mitigation; usercopy/swapgs barriers and __user pointer sanitization Vulnerability Spectre v2: Mitigation; Enhanced IBRS, IBPB conditional, RSB filling Vulnerability Srbds: Not affected Vulnerability Tsx async abort: Not affected 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 cpuid aperfmperf tsc_known_freq pni pclmulqdq dtes64 ds_cpl 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 cpuid_fault epb cat_l3 cat_l2 cdp_l3 invpcid_single intel_ppin cdp_l2 ssbd mba ibrs ibpb stibp ibrs_enhanced fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid cqm rdt_a avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb intel_pt avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local split_lock_detect avx_vnni avx512_bf16 wbnoinvd dtherm arat pln pts hwp hwp_act_window hwp_epp hwp_pkg_req avx512vbmi umip pku ospke waitpkg avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg tme avx512_vpopcntdq la57 rdpid bus_lock_detect cldemote movdiri movdir64b enqcmd fsrm md_clear serialize tsxldtrk pconfig arch_lbr avx512_fp16 flush_l1d arch_capabilities Versions of relevant libraries: [pip3] flashinfer==0.1.6+cu121torch2.4 [pip3] numpy==1.26.4 [pip3] nvidia-cublas-cu12==12.1.3.1 [pip3] nvidia-cuda-cupti-cu12==12.1.105 [pip3] nvidia-cuda-nvrtc-cu12==12.1.105 [pip3] nvidia-cuda-runtime-cu12==12.1.105 [pip3] nvidia-cudnn-cu12==9.1.0.70 [pip3] nvidia-cufft-cu12==11.0.2.54 [pip3] nvidia-curand-cu12==10.3.2.106 [pip3] nvidia-cusolver-cu12==11.4.5.107 [pip3] nvidia-cusparse-cu12==12.1.0.106 [pip3] nvidia-ml-py==12.560.30 [pip3] nvidia-nccl-cu12==2.20.5 [pip3] nvidia-nvjitlink-cu12==12.6.68 [pip3] nvidia-nvtx-cu12==12.1.105 [pip3] pyzmq==26.2.0 [pip3] torch==2.4.0 [pip3] torchvision==0.19.0 [pip3] transformers==4.45.0 [pip3] triton==3.0.0 [conda] Could not collect ROCM Version: Could not collect Neuron SDK Version: N/A vLLM Version: 0.6.1.dev238+ge2c6e0a82 vLLM Build Flags: CUDA Archs: Not Set; ROCm: Disabled; Neuron: Disabled GPU Topology: GPU0 NIC0 NIC1 NIC2 NIC3 NIC4 NIC5 NIC6 NIC7 CPU Affinity NUMA Affinity GPU NUMA ID GPU0 X SYS PIX PIX PXB SYS SYS SYS SYS 24-35 2 N/A NIC0 SYS X SYS SYS SYS SYS SYS SYS SYS NIC1 PIX SYS X PIX PXB SYS SYS SYS SYS NIC2 PIX SYS PIX X PXB SYS SYS SYS SYS NIC3 PXB SYS PXB PXB X SYS SYS SYS SYS NIC4 SYS SYS SYS SYS SYS X PIX PXB SYS NIC5 SYS SYS SYS SYS SYS PIX X PXB SYS NIC6 SYS SYS SYS SYS SYS PXB PXB X SYS NIC7 SYS SYS SYS SYS SYS SYS SYS SYS X 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 NIC Legend: NIC0: mlx5_0 NIC1: mlx5_1 NIC2: mlx5_2 NIC3: mlx5_3 NIC4: mlx5_4 NIC5: mlx5_5 NIC6: mlx5_6 NIC7: mlx5_7

Model Input Dumps

No response

🐛 Describe the bug

In vLLM v0.6.2, when tp=1, TPOT becomes very slow for batch sizes of 10 or so.

The vLLM arguments are as follows.

      --model
      - /data/models/llama-3-8b-instruct/base
      --tensor-parallel-size
      - “1”
      --load-format
      - “auto”
      --max-model-len
      - “8192”
      --disable-log-requests
      --uvicorn-log-level
      - “warning”
      image: aspcr01-queffmyz.scr.skr-west.scp-in.com/serving/vllm:v0.6.2

I tested with the same load in v0.5.5, and it handled fine.

When I looked at why it was slowing down, I realized that I was using very little GPU resources and only using 100% of the CPU cores.

Before submitting a new issue...

rafvasq commented 4 days ago

@DarkLight1337, I can investigate this one! Can I be assigned?