search

vllm-project / vllm

A high-throughput and memory-efficient inference and serving engine for LLMs
https://docs.vllm.ai
Apache License 2.0
29.94k stars 4.52k forks source link

[Bug]: gptq model fails on pascal gpu with long prompt #6567

Closed shesung closed 2 months ago

shesung commented 3 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 22.04.2 LTS (x86_64)
GCC version: (Ubuntu 11.3.0-1ubuntu1~22.04.1) 11.3.0
Clang version: Could not collect
CMake version: version 3.30.0
Libc version: glibc-2.35

Python version: 3.10.6 (main, May 29 2023, 11:10:38) [GCC 11.3.0] (64-bit runtime)
Python platform: Linux-4.14.83-x86_64-with-glibc2.35
Is CUDA available: True
CUDA runtime version: 12.1.105
CUDA_MODULE_LOADING set to: LAZY
GPU models and configuration:
GPU 0: Tesla P40
GPU 1: Tesla P40

Nvidia driver version: 535.129.03
cuDNN version: Probably one of the following:
/usr/lib/x86_64-linux-gnu/libcudnn.so.8.9.3
/usr/lib/x86_64-linux-gnu/libcudnn_adv_infer.so.8.9.3
/usr/lib/x86_64-linux-gnu/libcudnn_adv_train.so.8.9.3
/usr/lib/x86_64-linux-gnu/libcudnn_cnn_infer.so.8.9.3
/usr/lib/x86_64-linux-gnu/libcudnn_cnn_train.so.8.9.3
/usr/lib/x86_64-linux-gnu/libcudnn_ops_infer.so.8.9.3
/usr/lib/x86_64-linux-gnu/libcudnn_ops_train.so.8.9.3
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):                          24
On-line CPU(s) list:             0-23
Vendor ID:                       GenuineIntel
Model name:                      Intel(R) Xeon(R) CPU E5-2678 v3 @ 2.50GHz
CPU family:                      6
Model:                           63
Thread(s) per core:              2
Core(s) per socket:              12
Socket(s):                       1
Stepping:                        2
CPU max MHz:                     3300.0000
CPU min MHz:                     1200.0000
BogoMIPS:                        5000.19
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 $
m constant_tsc arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid dc$
 sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm cpuid_fault epb invpcid_single pti intel_ppin ibrs ibpb stibp tpr_shadow vnmi flexprior$
ty ept vpid fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid cqm xsaveopt cqm_llc cqm_occup_llc dtherm ida arat pln pts
Virtualization:                  VT-x
L1d cache:                       384 KiB (12 instances)
L1i cache:                       384 KiB (12 instances)
L2 cache:                        3 MiB (12 instances)
L3 cache:                        30 MiB (1 instance)
NUMA node(s):                    1
NUMA node0 CPU(s):               0-23
Vulnerability L1tf:              Mitigation; PTE Inversion; VMX conditional cache flushes, SMT vulnerable
Vulnerability Meltdown:          Mitigation; PTI
Vulnerability Spec store bypass: Vulnerable
Vulnerability Spectre v1:        Mitigation; __user pointer sanitization
Vulnerability Spectre v2:        Mitigation; Full generic retpoline, IBPB, IBRS_FW

Versions of relevant libraries:
[pip3] numpy==1.23.5
[pip3] nvidia-nccl-cu12==2.20.5
[pip3] torch==2.3.0
[pip3] torchvision==0.18.0
[pip3] transformers==4.42.4
[pip3] transformers-stream-generator==0.0.4
[pip3] triton==2.3.0
[conda] Could not collect
ROCM Version: Could not collect
Neuron SDK Version: N/A
vLLM Version: 0.5.1
vLLM Build Flags:
CUDA Archs: Not Set; ROCm: Disabled; Neuron: Disabled
GPU Topology:
GPU0    GPU1    CPU Affinity    NUMA Affinity   GPU NUMA ID
GPU0     X      PHB     0-23    0               N/A
GPU1    PHB      X      0-23    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

I try to use openai_api_server to serve Qwen2-7B-Instruct on P40. When serving gptq models, vllm comes into errors with long prompt.

fp16 commond:

python -m vllm.entrypoints.openai.api_server --served-model-name Qwen2-7B-Instruct --model llm_models/qwen/Qwen2-7B-Instruct/ --gpu-memory-utilization 0.95 --max-model-len 4096 --max-num-seqs 1 --dtype half

fp16 works fine with short or long prompt:

INFO 07-19 15:44:53 metrics.py:295] Avg prompt throughput: 107.4 tokens/s, Avg generation throughput: 2.2 tokens/s, Running: 1 reqs, Swapped: 0 reqs, Pending: 0 reqs, GPU KV cache
usage: 0.4%, CPU KV cache usage: 0.0%.
INFO 07-19 15:44:58 metrics.py:295] Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 17.6 tokens/s, Running: 1 reqs, Swapped: 0 reqs, Pending: 0 reqs, GPU KV cache u
sage: 0.5%, CPU KV cache usage: 0.0%.
INFO 07-19 15:45:03 metrics.py:295] Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 17.6 tokens/s, Running: 1 reqs, Swapped: 0 reqs, Pending: 0 reqs, GPU KV cache u
sage: 0.6%, CPU KV cache usage: 0.0%.

gptq-int8 command:

 python -m vllm.entrypoints.openai.api_server --served-model-name Qwen2-7B-Instruct --model llm_models/qwen/Qwen2-7B-Instruct-GPTQ-Int8/ --gpu-memory-utilization 0.95 --max-model-len 4096 --max-num-seqs 1

short prompt works fine:

INFO 07-19 16:18:29 metrics.py:295] Avg prompt throughput: 5.2 tokens/s, Avg generation throughput: 3.2 tokens/s, Running: 1 reqs, Swapped: 0 reqs, Pending: 0 reqs, GPU KV cache usage: 0.0%, CPU KV cache usage: 0.0%.
INFO 07-19 16:18:34 metrics.py:295] Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 31.4 tokens/s, Running: 1 reqs, Swapped: 0 reqs, Pending: 0 reqs, GPU KV cache usage: 0.1%, CPU KV cache usage: 0.0%.
INFO 07-19 16:18:39 metrics.py:295] Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 31.1 tokens/s, Running: 1 reqs, Swapped: 0 reqs, Pending: 0 reqs, GPU KV cache usage: 0.2%, CPU KV cache usage: 0.0%

long prompt go into error:

Traceback (most recent call last):
  File "/world/data-gpu-107/github/vllm/vllm/engine/async_llm_engine.py", line 604, in run_engine_loop
    done, _ = await asyncio.wait(
  File "/usr/lib/python3.10/asyncio/tasks.py", line 384, in wait
    return await _wait(fs, timeout, return_when, loop)
  File "/usr/lib/python3.10/asyncio/tasks.py", line 491, in _wait
    await waiter
asyncio.exceptions.CancelledError

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/usr/local/lib/python3.10/dist-packages/uvicorn/protocols/http/httptools_impl.py", line 399, in run_asgi
    result = await app(  # type: ignore[func-returns-value]
  File "/usr/local/lib/python3.10/dist-packages/uvicorn/middleware/proxy_headers.py", line 70, in __call__
    return await self.app(scope, receive, send)
  File "/usr/local/lib/python3.10/dist-packages/fastapi/applications.py", line 1054, in __call__
    await super().__call__(scope, receive, send)
  File "/usr/local/lib/python3.10/dist-packages/starlette/applications.py", line 123, in __call__
    await self.middleware_stack(scope, receive, send)
  File "/usr/local/lib/python3.10/dist-packages/starlette/middleware/errors.py", line 186, in __call__
    raise exc
  File "/usr/local/lib/python3.10/dist-packages/starlette/middleware/errors.py", line 164, in __call__
    await self.app(scope, receive, _send)
  File "/usr/local/lib/python3.10/dist-packages/starlette/middleware/cors.py", line 85, in __call__
    await self.app(scope, receive, send)
  File "/usr/local/lib/python3.10/dist-packages/starlette/middleware/exceptions.py", line 65, in __call__
    await wrap_app_handling_exceptions(self.app, conn)(scope, receive, send)
  File "/usr/local/lib/python3.10/dist-packages/starlette/_exception_handler.py", line 64, in wrapped_app
   await app(scope, receive, sender)
  File "/usr/local/lib/python3.10/dist-packages/starlette/routing.py", line 756, in __call__
    await self.middleware_stack(scope, receive, send)
  File "/usr/local/lib/python3.10/dist-packages/starlette/routing.py", line 776, in app
    await route.handle(scope, receive, send)
  File "/usr/local/lib/python3.10/dist-packages/starlette/routing.py", line 297, in handle
    await self.app(scope, receive, send)
  File "/usr/local/lib/python3.10/dist-packages/starlette/routing.py", line 77, in app
    await wrap_app_handling_exceptions(app, request)(scope, receive, send)
  File "/usr/local/lib/python3.10/dist-packages/starlette/_exception_handler.py", line 64, in wrapped_app
    raise exc
  File "/usr/local/lib/python3.10/dist-packages/starlette/_exception_handler.py", line 53, in wrapped_app
    await app(scope, receive, sender)
  File "/usr/local/lib/python3.10/dist-packages/starlette/routing.py", line 72, in app
    response = await func(request)
  File "/usr/local/lib/python3.10/dist-packages/fastapi/routing.py", line 278, in app
    raw_response = await run_endpoint_function(
  File "/usr/local/lib/python3.10/dist-packages/fastapi/routing.py", line 191, in run_endpoint_function
    return await dependant.call(**values)
  File "/world/data-gpu-107/github/vllm/vllm/entrypoints/openai/api_server.py", line 132, in create_chat_completion
    generator = await openai_serving_chat.create_chat_completion(
  File "/world/data-gpu-107/github/vllm/vllm/entrypoints/openai/serving_chat.py", line 305, in create_chat_completion
    return await self.chat_completion_full_generator(
  File "/world/data-gpu-107/github/vllm/vllm/entrypoints/openai/serving_chat.py", line 505, in chat_completion_full_generator
    async for res in result_generator:
  File "/world/data-gpu-107/github/vllm/vllm/engine/async_llm_engine.py", line 765, in generate
    async for output in self._process_request(
  File "/world/data-gpu-107/github/vllm/vllm/engine/async_llm_engine.py", line 881, in _process_request
    raise e
  File "/world/data-gpu-107/github/vllm/vllm/engine/async_llm_engine.py", line 877, in _process_request
    async for request_output in stream:
  File "/world/data-gpu-107/github/vllm/vllm/engine/async_llm_engine.py", line 91, in __anext__
    raise result
  File "/world/data-gpu-107/github/vllm/vllm/engine/async_llm_engine.py", line 44, in _log_task_completion
    return_value = task.result()
  File "/world/data-gpu-107/github/vllm/vllm/engine/async_llm_engine.py", line 603, in run_engine_loop
    async with asyncio_timeout(ENGINE_ITERATION_TIMEOUT_S):
  File "/world/data-gpu-107/github/vllm/vllm/engine/async_timeout.py", line 95, in __aexit__
    self._do_exit(exc_type)
  File "/world/data-gpu-107/github/vllm/vllm/engine/async_timeout.py", line 178, in _do_exit
    raise asyncio.TimeoutError
asyncio.exceptions.TimeoutError
INFO 07-19 16:09:21 metrics.py:295] Avg prompt throughput: 84.8 tokens/s, Avg generation throughput: 0.1 tokens/s, Running: 1 reqs, Swapped: 0 reqs, Pending: 0 reqs, GPU KV cache usage: 0.3%, CPU KV cache usage: 0.0%.
shesung commented 2 months ago

I found the reason may be prefill cost a long time. After setting VLLM_ENGINE_ITERATION_TIMEOUT_S to larger value, the inference goes well.