[Bug]: vllm failed to run two instance with one gpu

[pandada8]

Your current environment

The output of `python collect_env.py`

PyTorch version: 2.5.1+cu124
Is debug build: False
CUDA used to build PyTorch: 12.4
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: Could not collect
Libc version: glibc-2.35

Python version: 3.12.7 (main, Oct  1 2024, 08:52:12) [GCC 11.4.0] (64-bit runtime)
Python platform: Linux-5.15.0-73-generic-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: 535.113.01
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):                          16
On-line CPU(s) list:             0-15
Vendor ID:                       GenuineIntel
Model name:                      Intel(R) Xeon(R) Platinum 8369B CPU @ 2.90GHz
CPU family:                      6
Model:                           106
Thread(s) per core:              2
Core(s) per socket:              8
Socket(s):                       1
Stepping:                        6
BogoMIPS:                        5799.99
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 monitor ssse3 fma cx16 pdcm pcid sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand hypervisor lahf_lm abm 3dnowprefetch cpuid_fault invpcid_single ssbd ibrs ibpb stibp ibrs_enhanced fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves wbnoinvd arat avx512vbmi avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg avx512_vpopcntdq rdpid fsrm arch_capabilities
Hypervisor vendor:               KVM
Virtualization type:             full
L1d cache:                       384 KiB (8 instances)
L1i cache:                       256 KiB (8 instances)
L2 cache:                        10 MiB (8 instances)
L3 cache:                        48 MiB (1 instance)
NUMA node(s):                    1
NUMA node0 CPU(s):               0-15
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 Host state unknown
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

Versions of relevant libraries:
[pip3] flashinfer==0.1.6+cu121torch2.4
[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-ml-py==12.560.30
[pip3] nvidia-nccl-cu12==2.21.5
[pip3] nvidia-nvjitlink-cu12==12.4.127
[pip3] nvidia-nvtx-cu12==12.4.127
[pip3] pyzmq==26.2.0
[pip3] torch==2.5.1
[pip3] torchvision==0.20.1
[pip3] transformers==4.46.2
[pip3] triton==3.1.0
[conda] Could not collect
ROCM Version: Could not collect
Neuron SDK Version: N/A
vLLM Version: 0.6.4.post1
vLLM Build Flags:
CUDA Archs: Not Set; ROCm: Disabled; Neuron: Disabled
GPU Topology:
GPU0    CPU Affinity    NUMA Affinity   GPU NUMA ID
GPU0     X      0-15            N/A             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

NVIDIA_VISIBLE_DEVICES=all
NVIDIA_REQUIRE_CUDA=cuda>=12.4 brand=tesla,driver>=470,driver<471 brand=unknown,driver>=470,driver<471 brand=nvidia,driver>=470,driver<471 brand=nvidiartx,driver>=470,driver<471 brand=geforce,driver>=470,driver<471 brand=geforcertx,driver>=470,driver<471 brand=quadro,driver>=470,driver<471 brand=quadrortx,driver>=470,driver<471 brand=titan,driver>=470,driver<471 brand=titanrtx,driver>=470,driver<471 brand=tesla,driver>=525,driver<526 brand=unknown,driver>=525,driver<526 brand=nvidia,driver>=525,driver<526 brand=nvidiartx,driver>=525,driver<526 brand=geforce,driver>=525,driver<526 brand=geforcertx,driver>=525,driver<526 brand=quadro,driver>=525,driver<526 brand=quadrortx,driver>=525,driver<526 brand=titan,driver>=525,driver<526 brand=titanrtx,driver>=525,driver<526 brand=tesla,driver>=535,driver<536 brand=unknown,driver>=535,driver<536 brand=nvidia,driver>=535,driver<536 brand=nvidiartx,driver>=535,driver<536 brand=geforce,driver>=535,driver<536 brand=geforcertx,driver>=535,driver<536 brand=quadro,driver>=535,driver<536 brand=quadrortx,driver>=535,driver<536 brand=titan,driver>=535,driver<536 brand=titanrtx,driver>=535,driver<536
NVIDIA_DRIVER_CAPABILITIES=compute,utility
VLLM_USAGE_SOURCE=production-docker-image
CUDA_VERSION=12.4.1
LD_LIBRARY_PATH=/usr/local/lib/python3.12/dist-packages/cv2/../../lib64:/usr/local/nvidia/lib:/usr/local/nvidia/lib64
CUDA_MODULE_LOADING=LAZY

Model Input Dumps

No response

🐛 Describe the bug

I want to run Qwen2.5-14B-Instruct-GPTQ-Int4 and Qwen2.5-72B-Instruct-GPTQ-Int4 with one 80G A100.

with following command

python3 -m vllm.entrypoints.openai.api_server \
        --model /models/Qwen2.5-72B-Instruct-GPTQ-Int4 \
        --gpu-memory-utilization 0.75 \
        --max-model-len=16384 \
        --served-model-name qwen2.5-14b \
        --quantization=gptq \
        --enforce-eager \
        --enable-chunked-prefill \
        --enable-prefix-caching
python3 -m vllm.entrypoints.openai.api_server \
        --model /models/Qwen2.5-14B-Instruct-GPTQ-Int4 \
        --gpu-memory-utilization 0.20 \
        --max-model-len=16384 \
        --served-model-name qwen2.5-14b \
        --quantization=gptq \
        --enforce-eager \
        --enable-chunked-prefill \
        --enable-prefix-caching

Both commands succeed when running separately, consuming approximately 75% and 20% of the vRAM respectively.
However when you start the 72B first and then 14b. the 14b will refused to start with Exception

ValueError: No available memory for the cache blocks. Try increasing `gpu_memory_utilization` when initializing the engine.

Note that the 14B vllm server outputs the following log

Memory profiling results: total_gpu_memory=79.15GiB initial_memory_usage=68.01GiB peak_torch_memory=10.77GiB memory_usage_post_profile=68.04Gib non_torch_memory=58.66GiB kv_cache_size=-53.60GiB gpu_memory_utilization=0.20

which has negative kv_cache_size.

it looks like currently

vllm/vllm/worker/worker.py

Lines 213 to 219 in 1cfde82

	non_torch_allocations = total_allocated_bytes - torch_allocated_bytes
	if non_torch_allocations > 0:
	peak_memory += non_torch_allocations
	available_kv_cache_memory = (
	total_gpu_memory * self.cache_config.gpu_memory_utilization -
	peak_memory)

peak_memory would include the other vllm used vram and result in negative available_kv_cache_memory

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.

[pandada8]

#2248 (comment) suggest counting used vRAM before loading model and exclude it when calculating the available_kv_cache_memory. it seems works well when you staggered startup different vLLM worker in the same card

[pandada8]

proposed change

diff --git a/vllm/worker/worker.py b/vllm/worker/worker.py
index 80fd7bc3..793812e5 100644
--- a/vllm/worker/worker.py
+++ b/vllm/worker/worker.py
@@ -150,6 +150,8 @@ class Worker(LocalOrDistributedWorkerBase):
         set_random_seed(self.model_config.seed)
 
     def load_model(self):
+        free_gpu_memory, total_gpu_memory = torch.cuda.mem_get_info()
+        self.gpu_mem_pre_occupied = total_gpu_memory - free_gpu_memory
         self.model_runner.load_model()
 
     def save_sharded_state(
@@ -214,9 +216,7 @@ class Worker(LocalOrDistributedWorkerBase):
         if non_torch_allocations > 0:
             peak_memory += non_torch_allocations
 
-        available_kv_cache_memory = (
-            total_gpu_memory * self.cache_config.gpu_memory_utilization -
-            peak_memory)
+        available_kv_cache_memory = total_gpu_memory * gpu_memory_utilization - (peak_memory - self.gpu_mem_pre_occupied)
 
         # Calculate the number of blocks that can be allocated with the
         # profiled peak memory.

[DarkLight1337]

Please see #10511