[Performance]: For LoRA models, performance drops in versions other than 0.8.2.

[zhangyuqi-1]

Proposal to improve performance

On LoRA models, It's very strange. The performance difference between vllm version 0.8.2 and other versions is extremely significant. With the same input and model, the outputs of vllm 0.8.2 and 0.9.1 are completely different—0.9.1 directly produces garbled output. Based on current testing, 0.8.2 is the most stable and delivers the best results.

Report of performance regression

No response

Misc discussion on performance

No response

Your current environment (if you think it is necessary)

vllm 0.9.1
==============================
        System Info
==============================
OS                           : Ubuntu 22.04.3 LTS (x86_64)
GCC version                  : (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0
Clang version                : Could not collect
CMake version                : Could not collect
Libc version                 : glibc-2.35

==============================
       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               : 3.10.0 | packaged by conda-forge | (default, Nov 20 2021, 02:24:10) [GCC 9.4.0] (64-bit runtime)
Python platform              : Linux-5.14.0-587.el9.x86_64-x86_64-with-glibc2.35

==============================
       CUDA / GPU Info
==============================
Is CUDA available            : True
CUDA runtime version         : Could not collect
CUDA_MODULE_LOADING set to   : LAZY
GPU models and configuration : 
GPU 0: NVIDIA L20
GPU 1: NVIDIA L20
GPU 2: NVIDIA L20
GPU 3: NVIDIA L20
GPU 4: NVIDIA L20
GPU 5: NVIDIA L20
GPU 6: NVIDIA L20
GPU 7: NVIDIA L20

Nvidia driver version        : 550.163.01
cuDNN version                : Probably one of the following:
/usr/lib/x86_64-linux-gnu/libcudnn.so.8.9.6
/usr/lib/x86_64-linux-gnu/libcudnn_adv_infer.so.8.9.6
/usr/lib/x86_64-linux-gnu/libcudnn_adv_train.so.8.9.6
/usr/lib/x86_64-linux-gnu/libcudnn_cnn_infer.so.8.9.6
/usr/lib/x86_64-linux-gnu/libcudnn_cnn_train.so.8.9.6
/usr/lib/x86_64-linux-gnu/libcudnn_ops_infer.so.8.9.6
/usr/lib/x86_64-linux-gnu/libcudnn_ops_train.so.8.9.6
HIP runtime version          : N/A
MIOpen runtime version       : N/A
Is XNNPACK available         : True

==============================
          CPU Info
==============================
架构：                                x86_64
CPU 运行模式：                        32-bit, 64-bit
Address sizes:                        52 bits physical, 57 bits virtual
字节序：                              Little Endian
CPU:                                  128
在线 CPU 列表：                       0-127
厂商 ID：                             GenuineIntel
型号名称：                            INTEL(R) XEON(R) GOLD 6530
CPU 系列：                            6
型号：                                207
每个核的线程数：                      2
每个座的核数：                        32
座：                                  2
步进：                                2
CPU 最大 MHz��                        4000.0000
CPU 最小 MHz：                        800.0000
BogoMIPS：                            4200.00
标记：                                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 monitor 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 cpuid_fault epb cat_l3 cat_l2 cdp_l3 intel_ppin cdp_l2 ssbd mba ibrs ibpb stibp ibrs_enhanced tpr_shadow flexpriority ept vpid ept_ad 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 user_shstk avx_vnni avx512_bf16 wbnoinvd dtherm ida arat pln pts hwp hwp_act_window hwp_epp hwp_pkg_req vnmi 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 ibt amx_bf16 avx512_fp16 amx_tile amx_int8 flush_l1d arch_capabilities
虚拟化：                              VT-x
L1d 缓存：                            3 MiB (64 instances)
L1i 缓存：                            2 MiB (64 instances)
L2 缓存：                             128 MiB (64 instances)
L3 缓存：                             320 MiB (2 instances)
NUMA 节点：                           2
NUMA 节点0 CPU：                      0-31,64-95
NUMA 节点1 CPU：                      32-63,96-127
Vulnerability Gather data sampling:   Not affected
Vulnerability Itlb multihit:          Not affected
Vulnerability L1tf:                   Not affected
Vulnerability Mds:                    Not affected
Vulnerability Meltdown:               Not affected
Vulnerability Mmio stale data:        Not affected
Vulnerability Reg file data sampling: Not affected
Vulnerability Retbleed:               Not affected
Vulnerability Spec rstack overflow:   Not affected
Vulnerability Spec store bypass:      Mitigation; Speculative Store Bypass disabled via prctl
Vulnerability Spectre v1:             Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2:             Mitigation; Enhanced / Automatic IBRS; IBPB conditional; RSB filling; PBRSB-eIBRS SW sequence; BHI BHI_DIS_S
Vulnerability Srbds:                  Not affected
Vulnerability Tsx async abort:        Not affected

==============================
Versions of relevant libraries
==============================
[pip3] numpy==2.2.6
[pip3] nvidia-cublas-cu12==12.6.4.1
[pip3] nvidia-cuda-cupti-cu12==12.6.80
[pip3] nvidia-cuda-nvrtc-cu12==12.6.77
[pip3] nvidia-cuda-runtime-cu12==12.6.77
[pip3] nvidia-cudnn-cu12==9.5.1.17
[pip3] nvidia-cufft-cu12==11.3.0.4
[pip3] nvidia-cufile-cu12==1.11.1.6
[pip3] nvidia-curand-cu12==10.3.7.77
[pip3] nvidia-cusolver-cu12==11.7.1.2
[pip3] nvidia-cusparse-cu12==12.5.4.2
[pip3] nvidia-cusparselt-cu12==0.6.3
[pip3] nvidia-ml-py==12.575.51
[pip3] nvidia-nccl-cu12==2.26.2
[pip3] nvidia-nvjitlink-cu12==12.6.85
[pip3] nvidia-nvtx-cu12==12.6.77
[pip3] onnx==1.18.0
[pip3] onnxruntime==1.22.0
[pip3] onnxruntime-gpu==1.22.0
[pip3] pynvml==12.0.0
[pip3] pytorch-crf==0.7.2
[pip3] pyzmq==26.4.0
[pip3] sentence-transformers==4.1.0
[pip3] torch==2.7.0
[pip3] torchaudio==2.7.0
[pip3] torchvision==0.22.0
[pip3] transformers==4.52.4
[pip3] transformers-stream-generator==0.0.5
[pip3] triton==3.3.0
[conda] numpy                     2.2.6                    pypi_0    pypi
[conda] nvidia-cublas-cu12        12.6.4.1                 pypi_0    pypi
[conda] nvidia-cuda-cupti-cu12    12.6.80                  pypi_0    pypi
[conda] nvidia-cuda-nvrtc-cu12    12.6.77                  pypi_0    pypi
[conda] nvidia-cuda-runtime-cu12  12.6.77                  pypi_0    pypi
[conda] nvidia-cudnn-cu12         9.5.1.17                 pypi_0    pypi
[conda] nvidia-cufft-cu12         11.3.0.4                 pypi_0    pypi
[conda] nvidia-cufile-cu12        1.11.1.6                 pypi_0    pypi
[conda] nvidia-curand-cu12        10.3.7.77                pypi_0    pypi
[conda] nvidia-cusolver-cu12      11.7.1.2                 pypi_0    pypi
[conda] nvidia-cusparse-cu12      12.5.4.2                 pypi_0    pypi
[conda] nvidia-cusparselt-cu12    0.6.3                    pypi_0    pypi
[conda] nvidia-ml-py              12.575.51                pypi_0    pypi
[conda] nvidia-nccl-cu12          2.26.2                   pypi_0    pypi
[conda] nvidia-nvjitlink-cu12     12.6.85                  pypi_0    pypi
[conda] nvidia-nvtx-cu12          12.6.77                  pypi_0    pypi
[conda] pynvml                    12.0.0                   pypi_0    pypi
[conda] pytorch-crf               0.7.2                    pypi_0    pypi
[conda] pyzmq                     26.4.0                   pypi_0    pypi
[conda] sentence-transformers     4.1.0                    pypi_0    pypi
[conda] torch                     2.7.0                    pypi_0    pypi
[conda] torchaudio                2.7.0                    pypi_0    pypi
[conda] torchvision               0.22.0                   pypi_0    pypi
[conda] transformers              4.52.4                   pypi_0    pypi
[conda] transformers-stream-generator 0.0.5                    pypi_0    pypi
[conda] triton                    3.3.0                    pypi_0    pypi

==============================
         vLLM Info
==============================
ROCM Version                 : Could not collect
Neuron SDK Version           : N/A
vLLM Version                 : 0.9.1
vLLM Build Flags:
  CUDA Archs: Not Set; ROCm: Disabled; Neuron: Disabled
GPU Topology:
        GPU0    GPU1    GPU2    GPU3    GPU4    GPU5    GPU6    GPU7    NIC0    NIC1    NIC2    NIC3    CPU Affinity    NUMA Affinity   GPU NUMA ID
GPU0     X      PIX     PXB     PXB     PXB     PXB     PXB     PXB     NODE    NODE    NODE    NODE    0-31,64-95      0               N/A
GPU1    PIX      X      PXB     PXB     PXB     PXB     PXB     PXB     NODE    NODE    NODE    NODE    0-31,64-95      0               N/A
GPU2    PXB     PXB      X      PXB     PXB     PXB     PXB     PXB     NODE    NODE    NODE    NODE    0-31,64-95      0               N/A
GPU3    PXB     PXB     PXB      X      PXB     PXB     PXB     PXB     NODE    NODE    NODE    NODE    0-31,64-95      0               N/A
GPU4    PXB     PXB     PXB     PXB      X      PXB     PXB     PXB     NODE    NODE    NODE    NODE    0-31,64-95      0               N/A
GPU5    PXB     PXB     PXB     PXB     PXB      X      PXB     PXB     NODE    NODE    NODE    NODE    0-31,64-95      0               N/A
GPU6    PXB     PXB     PXB     PXB     PXB     PXB      X      PIX     NODE    NODE    NODE    NODE    0-31,64-95      0               N/A
GPU7    PXB     PXB     PXB     PXB     PXB     PXB     PIX      X      NODE    NODE    NODE    NODE    0-31,64-95      0               N/A
NIC0    NODE    NODE    NODE    NODE    NODE    NODE    NODE    NODE     X      PIX     NODE    NODE                            
NIC1    NODE    NODE    NODE    NODE    NODE    NODE    NODE    NODE    PIX      X      NODE    NODE                            
NIC2    NODE    NODE    NODE    NODE    NODE    NODE    NODE    NODE    NODE    NODE     X      PIX                             
NIC3    NODE    NODE    NODE    NODE    NODE    NODE    NODE    NODE    NODE    NODE    PIX      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

==============================
     Environment Variables
==============================
NCCL_CUMEM_ENABLE=0
PYTORCH_NVML_BASED_CUDA_CHECK=1
TORCHINDUCTOR_COMPILE_THREADS=1
CUDA_MODULE_LOADING=LAZY

vllm 0.8.2
==============================
        System Info
==============================
OS                           : Ubuntu 22.04.3 LTS (x86_64)
GCC version                  : (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0
Clang version                : Could not collect
CMake version                : Could not collect
Libc version                 : glibc-2.35

==============================
       PyTorch Info
==============================
PyTorch version              : 2.6.0+cu124
Is debug build               : False
CUDA used to build PyTorch   : 12.4
ROCM used to build PyTorch   : N/A

==============================
      Python Environment
==============================
Python version               : 3.10.18 | packaged by conda-forge | (main, Jun  4 2025, 14:45:41) [GCC 13.3.0] (64-bit runtime)
Python platform              : Linux-5.14.0-587.el9.x86_64-x86_64-with-glibc2.35

==============================
       CUDA / GPU Info
==============================
Is CUDA available            : True
CUDA runtime version         : Could not collect
CUDA_MODULE_LOADING set to   : LAZY
GPU models and configuration : 
GPU 0: NVIDIA L20
GPU 1: NVIDIA L20
GPU 2: NVIDIA L20
GPU 3: NVIDIA L20
GPU 4: NVIDIA L20
GPU 5: NVIDIA L20
GPU 6: NVIDIA L20
GPU 7: NVIDIA L20

Nvidia driver version        : 550.163.01
cuDNN version                : Probably one of the following:
/usr/lib/x86_64-linux-gnu/libcudnn.so.8.9.6
/usr/lib/x86_64-linux-gnu/libcudnn_adv_infer.so.8.9.6
/usr/lib/x86_64-linux-gnu/libcudnn_adv_train.so.8.9.6
/usr/lib/x86_64-linux-gnu/libcudnn_cnn_infer.so.8.9.6
/usr/lib/x86_64-linux-gnu/libcudnn_cnn_train.so.8.9.6
/usr/lib/x86_64-linux-gnu/libcudnn_ops_infer.so.8.9.6
/usr/lib/x86_64-linux-gnu/libcudnn_ops_train.so.8.9.6
HIP runtime version          : N/A
MIOpen runtime version       : N/A
Is XNNPACK available         : True

==============================
          CPU Info
==============================
架构：                                x86_64
CPU 运行模式：                        32-bit, 64-bit
Address sizes:                        52 bits physical, 57 bits virtual
字节序：                              Little Endian
CPU:                                  128
在线 CPU 列表：                       0-127
厂商 ID：                             GenuineIntel
型号名称：                            INTEL(R) XEON(R) GOLD 6530
CPU 系列：                            6
型号：                                207
每个核的线程数：                      2
每个座的核数：                        32
座：                                  2
步进：                                2
CPU 最大 MHz：                        4000.0000
CPU 最小 MHz：                        800.0000
BogoMIPS：                            4200.00
标记：                                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 monitor 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 cpuid_fault epb cat_l3 cat_l2 cdp_l3 intel_ppin cdp_l2 ssbd mba ibrs ibpb stibp ibrs_enhanced tpr_shadow flexpriority ept vpid ept_ad 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 user_shstk avx_vnni avx512_bf16 wbnoinvd dtherm ida arat pln pts hwp hwp_act_window hwp_epp hwp_pkg_req vnmi 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 ibt amx_bf16 avx512_fp16 amx_tile amx_int8 flush_l1d arch_capabilities
虚拟化：                              VT-x
L1d 缓存：                            3 MiB (64 instances)
L1i 缓存：                            2 MiB (64 instances)
L2 缓存：                             128 MiB (64 instances)
L3 缓存：                             320 MiB (2 instances)
NUMA 节点：                           2
NUMA 节点0 CPU：                      0-31,64-95
NUMA 节点1 CPU：                      32-63,96-127
Vulnerability Gather data sampling:   Not affected
Vulnerability Itlb multihit:          Not affected
Vulnerability L1tf:                   Not affected
Vulnerability Mds:                    Not affected
Vulnerability Meltdown:               Not affected
Vulnerability Mmio stale data:        Not affected
Vulnerability Reg file data sampling: Not affected
Vulnerability Retbleed:               Not affected
Vulnerability Spec rstack overflow:   Not affected
Vulnerability Spec store bypass:      Mitigation; Speculative Store Bypass disabled via prctl
Vulnerability Spectre v1:             Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2:             Mitigation; Enhanced / Automatic IBRS; IBPB conditional; RSB filling; PBRSB-eIBRS SW sequence; BHI BHI_DIS_S
Vulnerability Srbds:                  Not affected
Vulnerability Tsx async abort:        Not affected

==============================
Versions of relevant libraries
==============================
[pip3] numpy==1.26.4
[pip3] nvidia-cublas-cu12==12.4.5.8
[pip3] nvidia-cuda-cupti-cu12==12.4.127
[pip3] nvidia-cuda-nvrtc-cu12==12.4.127
[pip3] nvidia-cuda-runtime-cu12==12.4.127
[pip3] nvidia-cudnn-cu12==9.1.0.70
[pip3] nvidia-cufft-cu12==11.2.1.3
[pip3] nvidia-curand-cu12==10.3.5.147
[pip3] nvidia-cusolver-cu12==11.6.1.9
[pip3] nvidia-cusparse-cu12==12.3.1.170
[pip3] nvidia-cusparselt-cu12==0.6.2
[pip3] nvidia-nccl-cu12==2.21.5
[pip3] nvidia-nvjitlink-cu12==12.4.127
[pip3] nvidia-nvtx-cu12==12.4.127
[pip3] pyzmq==27.0.0
[pip3] torch==2.6.0
[pip3] torchaudio==2.6.0
[pip3] torchvision==0.21.0
[pip3] transformers==4.52.4
[pip3] triton==3.2.0
[conda] numpy                     1.26.4                   pypi_0    pypi
[conda] nvidia-cublas-cu12        12.4.5.8                 pypi_0    pypi
[conda] nvidia-cuda-cupti-cu12    12.4.127                 pypi_0    pypi
[conda] nvidia-cuda-nvrtc-cu12    12.4.127                 pypi_0    pypi
[conda] nvidia-cuda-runtime-cu12  12.4.127                 pypi_0    pypi
[conda] nvidia-cudnn-cu12         9.1.0.70                 pypi_0    pypi
[conda] nvidia-cufft-cu12         11.2.1.3                 pypi_0    pypi
[conda] nvidia-curand-cu12        10.3.5.147               pypi_0    pypi
[conda] nvidia-cusolver-cu12      11.6.1.9                 pypi_0    pypi
[conda] nvidia-cusparse-cu12      12.3.1.170               pypi_0    pypi
[conda] nvidia-cusparselt-cu12    0.6.2                    pypi_0    pypi
[conda] nvidia-nccl-cu12          2.21.5                   pypi_0    pypi
[conda] nvidia-nvjitlink-cu12     12.4.127                 pypi_0    pypi
[conda] nvidia-nvtx-cu12          12.4.127                 pypi_0    pypi
[conda] pyzmq                     27.0.0                   pypi_0    pypi
[conda] torch                     2.6.0                    pypi_0    pypi
[conda] torchaudio                2.6.0                    pypi_0    pypi
[conda] torchvision               0.21.0                   pypi_0    pypi
[conda] transformers              4.52.4                   pypi_0    pypi
[conda] triton                    3.2.0                    pypi_0    pypi

==============================
         vLLM Info
==============================
ROCM Version                 : Could not collect
Neuron SDK Version           : N/A
vLLM Version                 : 0.8.2
vLLM Build Flags:
  CUDA Archs: Not Set; ROCm: Disabled; Neuron: Disabled
GPU Topology:
        GPU0    GPU1    GPU2    GPU3    GPU4    GPU5    GPU6    GPU7    NIC0    NIC1    NIC2    NIC3    CPU Affinity    NUMA Affinity   GPU NUMA ID
GPU0     X      PIX     PXB     PXB     PXB     PXB     PXB     PXB     NODE    NODE    NODE    NODE    0-31,64-95      0               N/A
GPU1    PIX      X      PXB     PXB     PXB     PXB     PXB     PXB     NODE    NODE    NODE    NODE    0-31,64-95      0               N/A
GPU2    PXB     PXB      X      PXB     PXB     PXB     PXB     PXB     NODE    NODE    NODE    NODE    0-31,64-95      0               N/A
GPU3    PXB     PXB     PXB      X      PXB     PXB     PXB     PXB     NODE    NODE    NODE    NODE    0-31,64-95      0               N/A
GPU4    PXB     PXB     PXB     PXB      X      PXB     PXB     PXB     NODE    NODE    NODE    NODE    0-31,64-95      0               N/A
GPU5    PXB     PXB     PXB     PXB     PXB      X      PXB     PXB     NODE    NODE    NODE    NODE    0-31,64-95      0               N/A
GPU6    PXB     PXB     PXB     PXB     PXB     PXB      X      PIX     NODE    NODE    NODE    NODE    0-31,64-95      0               N/A
GPU7    PXB     PXB     PXB     PXB     PXB     PXB     PIX      X      NODE    NODE    NODE    NODE    0-31,64-95      0               N/A
NIC0    NODE    NODE    NODE    NODE    NODE    NODE    NODE    NODE     X      PIX     NODE    NODE                            
NIC1    NODE    NODE    NODE    NODE    NODE    NODE    NODE    NODE    PIX      X      NODE    NODE                            
NIC2    NODE    NODE    NODE    NODE    NODE    NODE    NODE    NODE    NODE    NODE     X      PIX                             
NIC3    NODE    NODE    NODE    NODE    NODE    NODE    NODE    NODE    NODE    NODE    PIX      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

==============================
     Environment Variables
==============================
NCCL_CUMEM_ENABLE=0
TORCHINDUCTOR_COMPILE_THREADS=1
CUDA_MODULE_LOADING=LAZY

Before submitting a new issue...

• Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.

[zhangyuqi-1]

The base models are Qwen-2-7B and Qwen-3-4B, both of which have been tested.

[jeejeelee]

Would it be convenient for you to provide your LoRA weights? I could try to reproduce this issue

[moguizhizi]

Can you provide your method? How can I print the system information?

==============================
        System Info
==============================
OS                           : Ubuntu 22.04.3 LTS (x86_64)
GCC version                  : (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0
Clang version                : Could not collect
CMake version                : Could not collect
Libc version                 : glibc-2.35

==============================
       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

[zhangyuqi-1]

Would it be convenient for you to provide your LoRA weights? I could try to reproduce this issue

Conveniencely， how should I send it to you？

[jeejeelee]

Can you provide your method? How can I print the system information?

==============================
        System Info
==============================
OS                           : Ubuntu 22.04.3 LTS (x86_64)
GCC version                  : (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0
Clang version                : Could not collect
CMake version                : Could not collect
Libc version                 : glibc-2.35

==============================
       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

Running the following script:

python https://github.com/vllm-project/vllm/blob/main/vllm/collect_env.py

[jeejeelee]

Would it be convenient for you to provide your LoRA weights? I could try to reproduce this issue

Conveniencely， how should I send it to you？

Have you joined the vLLM Slack? You can send it to me through Slack.

[zhangyuqi-1]

Would it be convenient for you to provide your LoRA weights? I could try to reproduce this issue

Interestingly, I just tested and found a strong correlation with concurrency. In version vllm 0.9.1, when I reduced the concurrency from 45 to 8, the output became normal. When the concurrency was set to 45 in vllm 0.9.1, the model's output started to repeat certain characters once it reached a certain length. I'm not sure if this is caused by abnormal GPU memory usage when the concurrency is too high.

However, in version vllm 0.8.2, the output was not affected by the concurrency level; even with a concurrency of 45, the output was normal.

My colleague suggested trying a smaller value for max-num-seqs in vllm 0.9.1, which should be helpful. I changed it to 8, and after running it again, vllm 0.9.1 was able to operate normally even with a concurrency of 45.

Here is my startup command:

CUDA_VISIBLE_DEVICES=4 PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True python -m vllm.entrypoints.openai.api_server --model /isilon_ai_data/llm_model_hub/Qwen/Qwen3-4B/ --trust-remote-code --gpu-memory-utilization 0.5 --max-model-len 20480 --max-num-seqs 8 --enable-lora --lora-modules html_table2json_lora=/isilon_ai_data/zhangyq/html-table2json/loras/qwen3-4b/v1-20250623-135815/checkpoint-810 --port 30009 --tensor-parallel-size 1

Since I cannot determine the maximum supported concurrency and there are no error messages, I'm concerned that during online service, a sudden spike in concurrency could lead to unusable model outputs.

[zhangyuqi-1]

In vllm 0.8.2, the output is not affected by the concurrency level; specifically, I mean that it can operate normally even without setting --max-num-seqs to 8. One possibility is that the internal default value for max-num-seqs in version 0.8.2 may be a smaller number, which could be related.

[zhangyuqi-1]

The prediction using engine v0 is normal

[jeejeelee]

It looks like there's a bug in the V1 LoRA. I'll investigate and figure it out ASAP. Have you tried using llama lora to see if there are any issues?

[Xuan2-33]

We also encountered the same problem in the new version 0.9.0. When the concurrency is large, it is very easy to have duplicate output. The proportion is approximately 1%. Hope this problem can be solved as soon as possible

[zhangyuqi-1]

It looks like there's a bug in the V1 LoRA. I'll investigate and figure it out ASAP. Have you tried using llama lora to see if there are any issues?

I have only tried Qwen LoRA

[moguizhizi]

What are the differences between the V0 LLM engine and the V1 LLM engine?