Open takagi97 opened 4 days ago
DarkLight1337
commented
4 days ago Personally, I resolved this via full build (i.e. python -m pip install -e .). I think some of the kernels/ops have been changed in this update, so they have to be rebuilt. @dtrifiro is this intended behavior?
takagi97
commented
4 days ago Is my foundational env (including driver, nccl, etc) corrupted? I am training LLMs which is not related to VLLM, and it is showing new warnings. @DarkLight1337
[WARNING] async_io requires the dev libaio .so object and headers but these were not found.
[WARNING] async_io: please install the libaio-dev package with apt
[WARNING] If libaio is already installed (perhaps from source), try setting the CFLAGS and LDFLAGS environment variables to where it can be found.
[WARNING] Please specify the CUTLASS repo directory as environment variable $CUTLASS_PATH
[WARNING] sparse_attn requires a torch version >= 1.5 and < 2.0 but detected 2.4
[WARNING] using untested triton version (3.0.0), only 1.0.0 is known to be compatibleAre you using the same conda environment for multiple applications? If yes, that may indeed be the case.
dtrifiro
commented
4 days ago @DarkLight1337 I'm not sure about the specific error, but it looks like something is wrong with the nvidia-nccl-cu12 dependency.
@takagi97 could you reinstall vllm in a fresh venv (or force reinstalling with pip install --force-reinstall nvidia-nccl-cu12)?
takagi97
commented
3 days ago @DarkLight1337 I'm not sure about the specific error, but it looks like something is wrong with the
nvidia-nccl-cu12dependency.@takagi97 could you reinstall
vllmin a fresh venv (or force reinstalling withpip install --force-reinstall nvidia-nccl-cu12)?
I run pip install --force-reinstall nvidia-nccl-cu12 but still got a core dumped. For reinstalling vllm in a new conda env, I got a NCCL bug:
INFO 10-18 14:42:48 utils.py:1009] Found nccl from library libnccl.so.2
ERROR 10-18 14:42:48 pynccl_wrapper.py:196] Failed to load NCCL library from libnccl.so.2 .It is expected if you are not running on NVIDIA/AMD GPUs.Otherwise, the nccl library might not exist, be corrupted or it does not support the current platform Linux-5.15.0-50-generic-x86_64-with-glibc2.31.If you already have the library, please set the environment variable VLLM_NCCL_SO_PATH to point to the correct nccl library path.I run python -c "import torch; print(torch.cuda.nccl.version())", return (2, 20, 5), seeming nccl can work?
takagi97
commented
3 days ago I have tested the similar conda env on a new server. My conclusion is that the following error comes from conda env itself:
[WARNING] async_io requires the dev libaio .so object and headers but these were not found.
[WARNING] async_io: please install the libaio-dev package with apt
[WARNING] If libaio is already installed (perhaps from source), try setting the CFLAGS and LDFLAGS environment variables to where it can be found.
[WARNING] Please specify the CUTLASS repo directory as environment variable $CUTLASS_PATH
[WARNING] sparse_attn requires a torch version >= 1.5 and < 2.0 but detected 2.4
[WARNING] using untested triton version (3.0.0), only 1.0.0 is known to be compatibleHowever, I don't know which package is responsible. And another error is caused by the new feature of VLLM, i.e., 0.6.3:
AttributeError: 'PyNcclCommunicator' object has no attribute 'device'My foundational environment, such as NCCL, is functioning normally.
cermeng
commented
1 day ago I don't think simply reinstalling nvidia-nccl-cu12 will solve the issue. Based on the code in the vLLM repository https://github.com/vllm-project/vllm/blob/main/vllm/utils.py#L988-L1010 and https://github.com/vllm-project/vllm/blob/main/vllm/distributed/device_communicators/pynccl_wrapper.py#L186-L205 , it means vLLM prioritizes torch's built-in nccl.
It appears that the nccl in your torch installation may be corrupted. Either of the following may help:
torch, or completely reinstall all vLLM dependencies by pip install --force-reinstall -r requirements-cuda.txt.nvidia-nccl-cu12, you can manually locate the nccl .so file, then you set the env VLLM_NCCL_SO_PATH= to that path.dtrifiro
commented
22 hours ago @cermeng nvidia-nccl-cu12 is a torch dependency. torch itself does not bundle libnccl.so.2.
@takagi97
Could you try doing the following (not using conda for now, but just the system python
python -m venv tmp_venv
source tmp_venv/bin/activate
pip install vllm==0.6.3.post1and then either running the code you were trying earlier or vllm serve facebook/opt-125m and see if the server starts up and you can query localhost:8000/v1/completions
Your current environment
The output of `python collect_env.py`
```text PyTorch version: 2.4.0 Is debug build: False CUDA used to build PyTorch: 12.4 ROCM used to build PyTorch: N/A OS: Ubuntu 20.04.5 LTS (x86_64) GCC version: (Ubuntu 9.4.0-1ubuntu1~20.04.1) 9.4.0 Clang version: Could not collect CMake version: Could not collect Libc version: glibc-2.31 Python version: 3.10.15 (main, Oct 3 2024, 07:27:34) [GCC 11.2.0] (64-bit runtime) Python platform: Linux-5.15.0-50-generic-x86_64-with-glibc2.31 Is CUDA available: True CUDA runtime version: 12.1.66 CUDA_MODULE_LOADING set to: LAZY GPU models and configuration: GPU 0: NVIDIA A800 80GB PCIe GPU 1: NVIDIA A800 80GB PCIe GPU 2: NVIDIA A800 80GB PCIe GPU 3: NVIDIA A800 80GB PCIe GPU 4: NVIDIA A800 80GB PCIe GPU 5: NVIDIA A800 80GB PCIe GPU 6: NVIDIA A800 80GB PCIe GPU 7: NVIDIA A800 80GB PCIe Nvidia driver version: 530.30.02 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: 46 bits physical, 57 bits virtual CPU(s): 128 On-line CPU(s) list: 0-127 Thread(s) per core: 2 Core(s) per socket: 32 Socket(s): 2 NUMA node(s): 4 Vendor ID: GenuineIntel CPU family: 6 Model: 106 Model name: Intel(R) Xeon(R) Platinum 8358P CPU @ 2.60GHz Stepping: 6 CPU MHz: 877.468 CPU max MHz: 3400.0000 CPU min MHz: 800.0000 BogoMIPS: 5200.00 L1d cache: 3 MiB L1i cache: 2 MiB L2 cache: 80 MiB L3 cache: 96 MiB NUMA node0 CPU(s): 0-15,64-79 NUMA node1 CPU(s): 16-31,80-95 NUMA node2 CPU(s): 32-47,96-111 NUMA node3 CPU(s): 48-63,112-127 Vulnerability Itlb multihit: Not affected Vulnerability L1tf: Not affected Vulnerability Mds: Not affected Vulnerability Meltdown: Not affected Vulnerability Mmio stale data: Vulnerable: Clear CPU buffers attempted, no microcode; SMT vulnerable Vulnerability Retbleed: 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, PBRSB-eIBRS SW sequence 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 pni pclmulqdq dtes64 monitor 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 invpcid_single intel_ppin 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 wbnoinvd dtherm ida arat pln pts avx512vbmi umip pku ospke avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg tme avx512_vpopcntdq la57 rdpid fsrm md_clear pconfig flush_l1d arch_capabilities Versions of relevant libraries: [pip3] numpy==1.26.4 [pip3] nvidia-ml-py==12.560.30 [pip3] pyzmq==26.2.0 [pip3] torch==2.4.0 [pip3] torchaudio==2.4.0 [pip3] torchvision==0.19.0 [pip3] transformers==4.45.0 [pip3] triton==3.0.0 [conda] blas 1.0 mkl defaults [conda] ffmpeg 4.3 hf484d3e_0 pytorch [conda] libjpeg-turbo 2.0.0 h9bf148f_0 pytorch [conda] mkl 2023.1.0 h213fc3f_46344 defaults [conda] mkl-service 2.4.0 py310h5eee18b_1 defaults [conda] mkl_fft 1.3.10 py310h5eee18b_0 defaults [conda] mkl_random 1.2.7 py310h1128e8f_0 defaults [conda] numpy 1.26.4 pypi_0 pypi [conda] nvidia-ml-py 12.560.30 pypi_0 pypi [conda] pytorch 2.4.0 py3.10_cuda12.4_cudnn9.1.0_0 pytorch [conda] pytorch-cuda 12.4 hc786d27_7 pytorch [conda] pytorch-mutex 1.0 cuda pytorch [conda] pyzmq 26.2.0 pypi_0 pypi [conda] torchaudio 2.4.0 py310_cu124 pytorch [conda] torchtriton 3.0.0 py310 pytorch [conda] torchvision 0.19.0 py310_cu124 pytorch [conda] transformers 4.45.0 pypi_0 pypi ROCM Version: Could not collect Neuron SDK Version: N/A vLLM Version: N/A vLLM Build Flags: CUDA Archs: Not Set; ROCm: Disabled; Neuron: Disabled GPU Topology: GPU0 GPU1 GPU2 GPU3 GPU4 GPU5 GPU6 GPU7 NIC0 NIC1 CPU Affinity NUMA Affinity GPU0 X PIX NV8 PIX SYS SYS SYS SYS NODE PIX 0-15,64-79 0 GPU1 PIX X PIX NV8 SYS SYS SYS SYS NODE PIX 0-15,64-79 0 GPU2 NV8 PIX X PIX SYS SYS SYS SYS NODE PIX 0-15,64-79 0 GPU3 PIX NV8 PIX X SYS SYS SYS SYS NODE PIX 0-15,64-79 0 GPU4 SYS SYS SYS SYS X PIX PIX NV6 SYS SYS 32-47,96-111 2 GPU5 SYS SYS SYS SYS PIX X NV4 PIX SYS SYS 32-47,96-111 2 GPU6 SYS SYS SYS SYS PIX NV4 X PIX SYS SYS 32-47,96-111 2 GPU7 SYS SYS SYS SYS NV6 PIX PIX X SYS SYS 32-47,96-111 2 NIC0 NODE NODE NODE NODE SYS SYS SYS SYS X NODE NIC1 PIX PIX PIX PIX SYS SYS SYS SYS NODE 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 ```Model Input Dumps
No response
🐛 Describe the bug
I want to update vllm 0.6.2 to 0.6.3. Followed by https://docs.vllm.ai/en/latest/getting_started/installation.html, I run
python python_only_dev.py --quit-devbut got an error. So I runpip uninstall vllm,pip install https://vllm-wheels.s3.us-west-2.amazonaws.com/nightly/vllm-1.0.0.dev-cp38-abi3-manylinux1_x86_64.whl+clone the latest code +python python_only_dev.py. After that, I tried to run the demo vllm code but got "Aborted (core dumped)“. I rebuilt my conda env, and the "core dumped" error disappeared. However, I got a NCCL error:Then I check my driver by running
CUDA_VISIBLE_DEVICES=6,7 NCCL_DEBUG=TRACE torchrun --nproc-per-node=2 test_vllm_env.py. It seems that NCCL is not found. Indeed, I checked this before updating vllm 0.6.2 to 0.6.3, everything is ok.Before submitting a new issue...