search

vllm-project / vllm

A high-throughput and memory-efficient inference and serving engine for LLMs
https://docs.vllm.ai
Apache License 2.0
31.16k stars 4.73k forks source link

[Bug]: AsyncEngineDeadError: Task finished unexpectedly with qwen2 72b #6208

Open thomZ1 opened 4 months ago

thomZ1 commented 4 months ago

Your current environment

PyTorch version: 2.3.0+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: version 3.29.3
Libc version: glibc-2.31

Python version: 3.10.14 (main, May  6 2024, 19:42:50) [GCC 11.2.0] (64-bit runtime)
Python platform: Linux-5.4.119-19-0009.11-x86_64-with-glibc2.31
Is CUDA available: True
CUDA runtime version: 12.2.140
CUDA_MODULE_LOADING set to: LAZY
GPU models and configuration: 
GPU 0: NVIDIA A800-SXM4-80GB
GPU 1: NVIDIA A800-SXM4-80GB
GPU 2: NVIDIA A800-SXM4-80GB
GPU 3: NVIDIA A800-SXM4-80GB
GPU 4: NVIDIA A800-SXM4-80GB
GPU 5: NVIDIA A800-SXM4-80GB
GPU 6: NVIDIA A800-SXM4-80GB
GPU 7: NVIDIA A800-SXM4-80GB

Nvidia driver version: 535.129.03
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:                   48 bits physical, 48 bits virtual
CPU(s):                          232
On-line CPU(s) list:             0-231
Thread(s) per core:              2
Core(s) per socket:              58
Socket(s):                       2
NUMA node(s):                    2
Vendor ID:                       AuthenticAMD
CPU family:                      25
Model:                           1
Model name:                      AMD EPYC 7K83 64-Core Processor
Stepping:                        1
CPU MHz:                         2445.404
BogoMIPS:                        4890.80
Hypervisor vendor:               KVM
Virtualization type:             full
L1d cache:                       3.6 MiB
L1i cache:                       3.6 MiB
L2 cache:                        58 MiB
L3 cache:                        448 MiB
NUMA node0 CPU(s):               0-115
NUMA node1 CPU(s):               116-231
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; Full AMD retpoline, IBPB conditional, IBRS_FW, STIBP 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 mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm constant_tsc rep_good nopl nonstop_tsc cpuid extd_apicid amd_dcm tsc_known_freq pni pclmulqdq monitor ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand hypervisor lahf_lm cmp_legacy cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw topoext perfctr_core invpcid_single ssbd ibrs ibpb stibp vmmcall fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid rdseed adx smap clflushopt clwb sha_ni xsaveopt xsavec xgetbv1 xsaves clzero xsaveerptr wbnoinvd arat umip pku ospke vaes vpclmulqdq rdpid fsrm

Versions of relevant libraries:
[pip3] numpy==1.26.4
[pip3] nvidia-nccl-cu12==2.20.5
[pip3] onnxruntime==1.18.0
[pip3] sentence-transformers==3.0.0
[pip3] torch==2.3.0
[pip3] transformers==4.41.2
[pip3] transformers-stream-generator==0.0.5
[pip3] triton==2.3.0
[pip3] vllm_nccl_cu12==2.18.1.0.4.0
[conda] numpy                     1.26.4                   pypi_0    pypi
[conda] nvidia-nccl-cu12          2.20.5                   pypi_0    pypi
[conda] sentence-transformers     3.0.0                    pypi_0    pypi
[conda] torch                     2.3.0                    pypi_0    pypi
[conda] transformers              4.41.2                   pypi_0    pypi
[conda] transformers-stream-generator 0.0.5                    pypi_0    pypi
[conda] triton                    2.3.0                    pypi_0    pypi
[conda] vllm-nccl-cu12            2.18.1.0.4.0             pypi_0    pypi
ROCM Version: Could not collect
Neuron SDK Version: N/A
vLLM Version: 0.5.0.post1
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    NIC4    NIC5    NIC6    NIC7    CPU Affinity    NUMA Affinity   GPU NUMA ID
GPU0     X      NV8     NV8     NV8     NV8     NV8     NV8     NV8     PIX     PIX     SYS     SYS     SYS     SYS     SYS     SYS     0-115   0               N/A
GPU1    NV8      X      NV8     NV8     NV8     NV8     NV8     NV8     PIX     PIX     SYS     SYS     SYS     SYS     SYS     SYS     0-115   0               N/A
GPU2    NV8     NV8      X      NV8     NV8     NV8     NV8     NV8     SYS     SYS     PIX     PIX     SYS     SYS     SYS     SYS     0-115   0               N/A
GPU3    NV8     NV8     NV8      X      NV8     NV8     NV8     NV8     SYS     SYS     PIX     PIX     SYS     SYS     SYS     SYS     0-115   0               N/A
GPU4    NV8     NV8     NV8     NV8      X      NV8     NV8     NV8     SYS     SYS     SYS     SYS     PIX     PIX     SYS     SYS     116-231 1               N/A
GPU5    NV8     NV8     NV8     NV8     NV8      X      NV8     NV8     SYS     SYS     SYS     SYS     PIX     PIX     SYS     SYS     116-231 1               N/A
GPU6    NV8     NV8     NV8     NV8     NV8     NV8      X      NV8     SYS     SYS     SYS     SYS     SYS     SYS     PIX     PIX     116-231 1               N/A
GPU7    NV8     NV8     NV8     NV8     NV8     NV8     NV8      X      SYS     SYS     SYS     SYS     SYS     SYS     PIX     PIX     116-231 1               N/A
NIC0    PIX     PIX     SYS     SYS     SYS     SYS     SYS     SYS      X      PIX     SYS     SYS     SYS     SYS     SYS     SYS
NIC1    PIX     PIX     SYS     SYS     SYS     SYS     SYS     SYS     PIX      X      SYS     SYS     SYS     SYS     SYS     SYS
NIC2    SYS     SYS     PIX     PIX     SYS     SYS     SYS     SYS     SYS     SYS      X      PIX     SYS     SYS     SYS     SYS
NIC3    SYS     SYS     PIX     PIX     SYS     SYS     SYS     SYS     SYS     SYS     PIX      X      SYS     SYS     SYS     SYS
NIC4    SYS     SYS     SYS     SYS     PIX     PIX     SYS     SYS     SYS     SYS     SYS     SYS      X      PIX     SYS     SYS
NIC5    SYS     SYS     SYS     SYS     PIX     PIX     SYS     SYS     SYS     SYS     SYS     SYS     PIX      X      SYS     SYS
NIC6    SYS     SYS     SYS     SYS     SYS     SYS     PIX     PIX     SYS     SYS     SYS     SYS     SYS     SYS      X      PIX
NIC7    SYS     SYS     SYS     SYS     SYS     SYS     PIX     PIX     SYS     SYS     SYS     SYS     SYS     SYS     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_bond_0
  NIC1: mlx5_bond_1
  NIC2: mlx5_bond_2
  NIC3: mlx5_bond_3
  NIC4: mlx5_bond_4
  NIC5: mlx5_bond_5
  NIC6: mlx5_bond_6
  NIC7: mlx5_bond_7

🐛 Describe the bug

I am using the eval-scope to test the concurrent throughput of the Qwen2 72B Instruct model deployed with VLLM. When running with 8 concurrent sessions, inputting 8k tokens, and outputting 2k tokens for a period of time, the VLLM service becomes inaccessible.

https://github.com/modelscope/eval-scope/tree/main:

llmuses perf --url 'http://localhost:8000/v1/chat/completions' --parallel 8 --model 'qwen' --log-every-n-query 10 --read-timeout=1000 --connect-timeout=1000 -n 500 --max-prompt-length 100000 \
                --tokenizer-path '/mnt/cfs_standard/Model/Qwen2/Qwen2-72B-Instruct/' --api openai --query-template '{"model": "%m", "messages": [{"role": "user","content": "%p"}], "stream": true,"skip_special_tokens": false,"max_tokens":1600,"stop": [], "early_stopping":false ,"ignore_eos":false,"repetition_penalty":0.25}' \
                --dataset openqa --dataset-path '/mnt/cfs_standard/User/zhangyi/Inference/lbl_m16000.jsonl' --name 'vllm-perf-qwen-72b-p8-16k' --wandb-api-key 'c6ff0fb6067a9a68e8be256b29c7f3b69b01b8ce'
INFO 07-07 15:20:11 metrics.py:341] Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 0.0 tokens/s, Running: 8 reqs, Swapped: 0 reqs, Pending: 0 reqs, GPU KV cache usage: 3.5%, CPU KV cache usage: 0.0%.
ERROR 07-07 15:20:11 async_llm_engine.py:535] Engine iteration timed out. This should never happen!
ERROR 07-07 15:20:11 async_llm_engine.py:52] Engine background task failed
ERROR 07-07 15:20:11 async_llm_engine.py:52] Traceback (most recent call last):
ERROR 07-07 15:20:11 async_llm_engine.py:52]   File "/root/miniconda3/lib/python3.10/site-packages/vllm/engine/async_llm_engine.py", line 506, in engine_step
ERROR 07-07 15:20:11 async_llm_engine.py:52]     request_outputs = await self.engine.step_async()
ERROR 07-07 15:20:11 async_llm_engine.py:52]   File "/root/miniconda3/lib/python3.10/site-packages/vllm/engine/async_llm_engine.py", line 235, in step_async
ERROR 07-07 15:20:11 async_llm_engine.py:52]     output = await self.model_executor.execute_model_async(
ERROR 07-07 15:20:11 async_llm_engine.py:52]   File "/root/miniconda3/lib/python3.10/site-packages/vllm/executor/distributed_gpu_executor.py", line 166, in execute_model_async
ERROR 07-07 15:20:11 async_llm_engine.py:52]     return await self._driver_execute_model_async(execute_model_req)
ERROR 07-07 15:20:11 async_llm_engine.py:52]   File "/root/miniconda3/lib/python3.10/site-packages/vllm/executor/multiproc_gpu_executor.py", line 149, in _driver_execute_model_async
ERROR 07-07 15:20:11 async_llm_engine.py:52]     return await self.driver_exec_model(execute_model_req)
ERROR 07-07 15:20:11 async_llm_engine.py:52] asyncio.exceptions.CancelledError
ERROR 07-07 15:20:11 async_llm_engine.py:52] 
ERROR 07-07 15:20:11 async_llm_engine.py:52] During handling of the above exception, another exception occurred:
ERROR 07-07 15:20:11 async_llm_engine.py:52] 
ERROR 07-07 15:20:11 async_llm_engine.py:52] Traceback (most recent call last):
ERROR 07-07 15:20:11 async_llm_engine.py:52]   File "/root/miniconda3/lib/python3.10/asyncio/tasks.py", line 456, in wait_for
ERROR 07-07 15:20:11 async_llm_engine.py:52]     return fut.result()
ERROR 07-07 15:20:11 async_llm_engine.py:52] asyncio.exceptions.CancelledError
ERROR 07-07 15:20:11 async_llm_engine.py:52] 
ERROR 07-07 15:20:11 async_llm_engine.py:52] The above exception was the direct cause of the following exception:
ERROR 07-07 15:20:11 async_llm_engine.py:52] 
ERROR 07-07 15:20:11 async_llm_engine.py:52] Traceback (most recent call last):
ERROR 07-07 15:20:11 async_llm_engine.py:52]   File "/root/miniconda3/lib/python3.10/site-packages/vllm/engine/async_llm_engine.py", line 42, in _log_task_completion
ERROR 07-07 15:20:11 async_llm_engine.py:52]     return_value = task.result()
ERROR 07-07 15:20:11 async_llm_engine.py:52]   File "/root/miniconda3/lib/python3.10/site-packages/vllm/engine/async_llm_engine.py", line 532, in run_engine_loop
ERROR 07-07 15:20:11 async_llm_engine.py:52]     has_requests_in_progress = await asyncio.wait_for(
ERROR 07-07 15:20:11 async_llm_engine.py:52]   File "/root/miniconda3/lib/python3.10/asyncio/tasks.py", line 458, in wait_for
ERROR 07-07 15:20:11 async_llm_engine.py:52]     raise exceptions.TimeoutError() from exc
ERROR 07-07 15:20:11 async_llm_engine.py:52] asyncio.exceptions.TimeoutError
Exception in callback functools.partial(<function _log_task_completion at 0x7f725fb83880>, error_callback=<bound method AsyncLLMEngine._error_callback of <vllm.engine.async_llm_engine.AsyncLLMEngine object at 0x7f73e6b29600>>)
handle: <Handle functools.partial(<function _log_task_completion at 0x7f725fb83880>, error_callback=<bound method AsyncLLMEngine._error_callback of <vllm.engine.async_llm_engine.AsyncLLMEngine object at 0x7f73e6b29600>>)>
Traceback (most recent call last):
  File "/root/miniconda3/lib/python3.10/site-packages/vllm/engine/async_llm_engine.py", line 506, in engine_step
    request_outputs = await self.engine.step_async()
  File "/root/miniconda3/lib/python3.10/site-packages/vllm/engine/async_llm_engine.py", line 235, in step_async
    output = await self.model_executor.execute_model_async(
  File "/root/miniconda3/lib/python3.10/site-packages/vllm/executor/distributed_gpu_executor.py", line 166, in execute_model_async
    return await self._driver_execute_model_async(execute_model_req)
  File "/root/miniconda3/lib/python3.10/site-packages/vllm/executor/multiproc_gpu_executor.py", line 149, in _driver_execute_model_async
    return await self.driver_exec_model(execute_model_req)
asyncio.exceptions.CancelledError

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/root/miniconda3/lib/python3.10/asyncio/tasks.py", line 456, in wait_for
    return fut.result()
asyncio.exceptions.CancelledError

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "/root/miniconda3/lib/python3.10/site-packages/vllm/engine/async_llm_engine.py", line 42, in _log_task_completion
    return_value = task.result()
  File "/root/miniconda3/lib/python3.10/site-packages/vllm/engine/async_llm_engine.py", line 532, in run_engine_loop
    has_requests_in_progress = await asyncio.wait_for(
  File "/root/miniconda3/lib/python3.10/asyncio/tasks.py", line 458, in wait_for
    raise exceptions.TimeoutError() from exc
asyncio.exceptions.TimeoutError

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "uvloop/cbhandles.pyx", line 63, in uvloop.loop.Handle._run
  File "/root/miniconda3/lib/python3.10/site-packages/vllm/engine/async_llm_engine.py", line 54, in _log_task_completion
    raise AsyncEngineDeadError(
vllm.engine.async_llm_engine.AsyncEngineDeadError: Task finished unexpectedly. This should never happen! Please open an issue on Github. See stack trace above for theactual cause.
xbinglzh commented 4 months ago

ERROR 07-10 11:08:14 async_llm_engine.py:483] Engine iteration timed out. This should never happen! ERROR 07-10 11:08:14 async_llm_engine.py:43] Engine background task failed ERROR 07-10 11:08:14 async_llm_engine.py:43] Traceback (most recent call last): ERROR 07-10 11:08:14 async_llm_engine.py:43] File "/home/ray/anaconda3/lib/python3.10/site-packages/vllm/engine/async_llm_engine.py", line 454, in engine_step ERROR 07-10 11:08:14 async_llm_engine.py:43] request_outputs = await self.engine.step_async() ERROR 07-10 11:08:14 async_llm_engine.py:43] File "/home/ray/anaconda3/lib/python3.10/site-packages/vllm/engine/async_llm_engine.py", line 213, in step_async ERROR 07-10 11:08:14 async_llm_engine.py:43] output = await self.model_executor.execute_model_async( ERROR 07-10 11:08:14 async_llm_engine.py:43] File "/home/ray/anaconda3/lib/python3.10/site-packages/vllm/executor/ray_gpu_executor.py", line 418, in execute_model_async ERROR 07-10 11:08:14 async_llm_engine.py:43] all_outputs = await self._run_workers_async( ERROR 07-10 11:08:14 async_llm_engine.py:43] File "/home/ray/anaconda3/lib/python3.10/site-packages/vllm/executor/ray_gpu_executor.py", line 408, in _run_workers_async ERROR 07-10 11:08:14 async_llm_engine.py:43] all_outputs = await asyncio.gather(coros) ERROR 07-10 11:08:14 async_llm_engine.py:43] asyncio.exceptions.CancelledError ERROR 07-10 11:08:14 async_llm_engine.py:43] ERROR 07-10 11:08:14 async_llm_engine.py:43] During handling of the above exception, another exception occurred: ERROR 07-10 11:08:14 async_llm_engine.py:43] ERROR 07-10 11:08:14 async_llm_engine.py:43] Traceback (most recent call last): ERROR 07-10 11:08:14 async_llm_engine.py:43] File "/home/ray/anaconda3/lib/python3.10/asyncio/tasks.py", line 456, in wait_for ERROR 07-10 11:08:14 async_llm_engine.py:43] return fut.result() ERROR 07-10 11:08:14 async_llm_engine.py:43] asyncio.exceptions.CancelledError ERROR 07-10 11:08:14 async_llm_engine.py:43] ERROR 07-10 11:08:14 async_llm_engine.py:43] The above exception was the direct cause of the following exception: ERROR 07-10 11:08:14 async_llm_engine.py:43] ERROR 07-10 11:08:14 async_llm_engine.py:43] Traceback (most recent call last): ERROR 07-10 11:08:14 async_llm_engine.py:43] File "/home/ray/anaconda3/lib/python3.10/site-packages/vllm/engine/async_llm_engine.py", line 38, in _raise_exception_on_finish ERROR 07-10 11:08:14 async_llm_engine.py:43] task.result() ERROR 07-10 11:08:14 async_llm_engine.py:43] File "/home/ray/anaconda3/lib/python3.10/site-packages/vllm/engine/async_llm_engine.py", line 480, in run_engine_loop ERROR 07-10 11:08:14 async_llm_engine.py:43] has_requests_in_progress = await asyncio.wait_for( ERROR 07-10 11:08:14 async_llm_engine.py:43] File "/home/ray/anaconda3/lib/python3.10/asyncio/tasks.py", line 458, in wait_for ERROR 07-10 11:08:14 async_llm_engine.py:43] raise exceptions.TimeoutError() from exc ERROR 07-10 11:08:14 async_llm_engine.py:43] asyncio.exceptions.TimeoutError 2024-07-10 11:08:14,573 - asyncio - ERROR - Exception in callback functools.partial(<function _raise_exception_on_finish at 0x7fcf686829e0>, error_callback=<bound method AsyncLLMEngine._error_callback of <vllm.engine.async_llm_engine.AsyncLLMEngine object at 0x7fd15e5da680>>) handle: <Handle functools.partial(<function _raise_exception_on_finish at 0x7fcf686829e0>, error_callback=<bound method AsyncLLMEngine._error_callback of <vllm.engine.async_llm_engine.AsyncLLMEngine object at 0x7fd15e5da680>>)> Traceback (most recent call last): File "/home/ray/anaconda3/lib/python3.10/site-packages/vllm/engine/async_llm_engine.py", line 454, in engine_step request_outputs = await self.engine.step_async() File "/home/ray/anaconda3/lib/python3.10/site-packages/vllm/engine/async_llm_engine.py", line 213, in step_async output = await self.model_executor.execute_model_async( File "/home/ray/anaconda3/lib/python3.10/site-packages/vllm/executor/ray_gpu_executor.py", line 418, in execute_model_async all_outputs = await self._run_workers_async( File "/home/ray/anaconda3/lib/python3.10/site-packages/vllm/executor/ray_gpu_executor.py", line 408, in _run_workers_async all_outputs = await asyncio.gather(coros) asyncio.exceptions.CancelledError

During handling of the above exception, another exception occurred:

Traceback (most recent call last): File "/home/ray/anaconda3/lib/python3.10/asyncio/tasks.py", line 456, in wait_for return fut.result() asyncio.exceptions.CancelledError

The above exception was the direct cause of the following exception:

Traceback (most recent call last): File "/home/ray/anaconda3/lib/python3.10/site-packages/vllm/engine/async_llm_engine.py", line 38, in _raise_exception_on_finish task.result() File "/home/ray/anaconda3/lib/python3.10/site-packages/vllm/engine/async_llm_engine.py", line 480, in run_engine_loop has_requests_in_progress = await asyncio.wait_for( File "/home/ray/anaconda3/lib/python3.10/asyncio/tasks.py", line 458, in wait_for raise exceptions.TimeoutError() from exc asyncio.exceptions.TimeoutError

The above exception was the direct cause of the following exception:

Traceback (most recent call last): File "uvloop/cbhandles.pyx", line 63, in uvloop.loop.Handle._run File "/home/ray/anaconda3/lib/python3.10/site-packages/vllm/engine/async_llm_engine.py", line 45, in _raise_exception_on_finish raise AsyncEngineDeadError( vllm.engine.async_llm_engine.AsyncEngineDeadError: Task finished unexpectedly. This should never happen! Please open an issue on Github. See stack trace above for the actual cause.

Lzhang-hub commented 4 months ago

+1

jessiewiswjc commented 4 months ago

+1

wangye360 commented 4 months ago

+1

DarkLight1337 commented 4 months ago

We have a tracking issue (#5901) for this. Please provide more details there so we can better troubleshoot the underlying cause.

HaIouya commented 4 months ago

when i used the glm4-9b-int8 and qwen2-72b-int4, i met this problem too.

bltcn commented 4 months ago

me too.Once it happen "Engine iteration timed out. This should never happen!", all the server will never response.

ericzhou571 commented 2 months ago

We encountered the same error ("Process group watchdog thread terminated with exception: CUDA error: an illegal memory access was encountered... Engine iteration timed out. This should never happen!") multiple times across v0.4.3, v0.5.5, and v0.6.1 post2, specifically with the A800-80G * 4 setup, tp=4 configuration.

In v0.4.3, using --disable-custom-all-reduce resolved the issue. However, in v0.5.5 and v0.6.1 post2, this flag no longer works.

as mentioned in the discussion here: https://github.com/vllm-project/vllm/issues/8230 @Sekri0 highlighted that --enable-prefix-caching could cause the CUDA illegal memory access error, with the traceback pointing to FlashAttention as a potential source. Although pull requests 7018 and 7142 appeared to address this issue, it persists in vLLM 0.5.5. Given that Flash Attention directly manages and accesses GPU memory, this detailed control could indeed increase the likelihood of encountering illegal memory access errors, especially if Flash Attention is not configured properly.

Based on these clues, we decided to uninstall Flash Attention and switch to xformers. After doing so, we successfully processed thousands of requests with tp=4, without any errors. We also tested aborting hundreds of requests abruptly and resending them in loops. The server remained robust throughout. While the speed is slightly slower compared to Flash Attention (a difference of a few hundred milliseconds), the system stability has significantly improved.

If you're facing similar issues, Maybe uninstall Flash Attention and use xformers instead will help.