[Bug]: Gemma 27B crashes on GCP A100

[noamgat]

Your current environment

PyTorch version: 2.3.1+cu121
Is debug build: False
CUDA used to build PyTorch: 12.1
ROCM used to build PyTorch: N/A

OS: Ubuntu 22.04.4 LTS (x86_64)
GCC version: (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0
Clang version: Could not collect
CMake version: version 3.30.0
Libc version: glibc-2.35

Python version: 3.10.12 (main, Nov 20 2023, 15:14:05) [GCC 11.4.0] (64-bit runtime)
Python platform: Linux-6.1.85+-x86_64-with-glibc2.35
Is CUDA available: True
CUDA runtime version: Could not collect
CUDA_MODULE_LOADING set to: LAZY
GPU models and configuration: GPU 0: NVIDIA A100-SXM4-80GB
Nvidia driver version: 535.161.08
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
Address sizes:                        46 bits physical, 48 bits virtual
Byte Order:                           Little Endian
CPU(s):                               96
On-line CPU(s) list:                  0-95
Vendor ID:                            GenuineIntel
Model name:                           Intel(R) Xeon(R) CPU @ 2.20GHz
CPU family:                           6
Model:                                85
Thread(s) per core:                   2
Core(s) per socket:                   24
Socket(s):                            2
Stepping:                             7
BogoMIPS:                             4400.34
Flags:                                fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ss ht syscall nx pdpe1gb rdtscp lm constant_tsc rep_good nopl xtopology nonstop_tsc cpuid tsc_known_freq pni pclmulqdq ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand hypervisor lahf_lm abm 3dnowprefetch invpcid_single ssbd ibrs ibpb stibp ibrs_enhanced fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm mpx avx512f avx512dq rdseed adx smap clflushopt clwb avx512cd avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves arat avx512_vnni md_clear arch_capabilities
Hypervisor vendor:                    KVM
Virtualization type:                  full
L1d cache:                            1.5 MiB (48 instances)
L1i cache:                            1.5 MiB (48 instances)
L2 cache:                             48 MiB (48 instances)
L3 cache:                             77 MiB (2 instances)
NUMA node(s):                         2
NUMA node0 CPU(s):                    0-23,48-71
NUMA node1 CPU(s):                    24-47,72-95
Vulnerability Gather data sampling:   Not affected
Vulnerability Itlb multihit:          Not affected
Vulnerability L1tf:                   Not affected
Vulnerability Mds:                    Mitigation; Clear CPU buffers; SMT Host state unknown
Vulnerability Meltdown:               Not affected
Vulnerability Mmio stale data:        Vulnerable: Clear CPU buffers attempted, no microcode; SMT Host state unknown
Vulnerability Reg file data sampling: Not affected
Vulnerability Retbleed:               Mitigation; Enhanced IBRS
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 SW loop, KVM SW loop
Vulnerability Srbds:                  Not affected
Vulnerability Tsx async abort:        Mitigation; Clear CPU buffers; SMT Host state unknown

Versions of relevant libraries:
[pip3] flashinfer==0.0.9+cu121torch2.3
[pip3] numpy==1.26.4
[pip3] nvidia-nccl-cu12==2.20.5
[pip3] torch==2.3.1
[pip3] torchvision==0.18.1
[pip3] transformers==4.42.4
[pip3] triton==2.3.1
[conda] Could not collect
ROCM Version: Could not collect
Neuron SDK Version: N/A
vLLM Version: 0.5.2
vLLM Build Flags:
CUDA Archs: Not Set; ROCm: Disabled; Neuron: Disabled
GPU Topology:
GPU0	CPU Affinity	NUMA Affinity	GPU NUMA ID
GPU0	 X 	0-23,48-71	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

🐛 Describe the bug

When using the OpenAI image on docker on GCP and launching Gemma-2-27B-it, we get the following error after a few requests:

    output = await make_async(self.driver_worker.execute_model
  File "/usr/lib/python3.10/concurrent/futures/thread.py", line 58, in run
    result = self.fn(*self.args, **self.kwargs)
  File "/usr/local/lib/python3.10/dist-packages/vllm/worker/worker_base.py", line 260, in execute_model
    self.execute_worker(worker_input)
  File "/usr/local/lib/python3.10/dist-packages/torch/utils/_contextlib.py", line 115, in decorate_context
    return func(*args, **kwargs)
  File "/usr/local/lib/python3.10/dist-packages/vllm/worker/worker.py", line 291, in execute_worker
    self.cache_engine[virtual_engine].copy(worker_input.blocks_to_copy)
  File "/usr/local/lib/python3.10/dist-packages/vllm/worker/cache_engine.py", line 102, in copy
    self.attn_backend.copy_blocks(self.gpu_cache, src_to_dsts)
  File "/usr/local/lib/python3.10/dist-packages/vllm/attention/backends/flashinfer.py", line 56, in copy_blocks
    raise NotImplementedError
NotImplementedError

This is when using either 0.5.1 or 0.5.2, using FlashInfer attention backend and disabling sliding window attention.

[noamgat]

Maybe related to #6361

It seems that indeed FlashInfer does not implement copy_blocks. I see several other implementations just pass the call to PagedAttention.copy_blocks(). Can that work in this case?

[HelloElwin]

Same issue here. I was able to fix it by importing PagedAttention and implementing these methods.

    @staticmethod
    def swap_blocks(
        src_kv_cache: torch.Tensor,
        dst_kv_cache: torch.Tensor,
        src_to_dst: torch.Tensor,
    ) -> None:
        PagedAttention.swap_blocks(src_kv_cache, dst_kv_cache, src_to_dst)
        # raise NotImplementedError

    @staticmethod
    def copy_blocks(
        kv_caches: List[torch.Tensor],
        src_to_dists: torch.Tensor,
    ) -> None:
        PagedAttention.copy_blocks(kv_caches, src_to_dists)
        # raise NotImplementedError