[Bug]: init_mm_limits_per_prompt not been called when using V1 + TensorSplit + Qwen2VL

[Leoyzen]

Your current environment

The output of `python collect_env.py`

Collecting environment information...
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: Debian GNU/Linux 11 (bullseye) (x86_64)
GCC version: (Debian 10.2.1-6) 10.2.1 20210110
Clang version: Could not collect
CMake version: Could not collect
Libc version: glibc-2.31

Python version: 3.10.16 (main, Jan  5 2025, 05:32:43) [Clang 19.1.6 ] (64-bit runtime)
Python platform: Linux-5.10.135.bsk.6-amd64-x86_64-with-glibc2.31
Is CUDA available: True
CUDA runtime version: Could not collect
CUDA_MODULE_LOADING set to: LAZY
GPU models and configuration:
GPU 0: NVIDIA H20
GPU 1: NVIDIA H20

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:                   52 bits physical, 57 bits virtual
Byte Order:                      Little Endian
CPU(s):                          192
On-line CPU(s) list:             0-191
Vendor ID:                       GenuineIntel
Model name:                      Intel(R) Xeon(R) Platinum 8457C
CPU family:                      6
Model:                           143
Thread(s) per core:              2
Core(s) per socket:              48
Socket(s):                       2
Stepping:                        8
CPU(s) scaling MHz:              82%
CPU max MHz:                     3800.0000
CPU min MHz:                     800.0000
BogoMIPS:                        5200.00
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 tsc_known_freq pni pclmulqdq dtes64 monitor ds_cpl vmx 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 cat_l2 cdp_l3 invpcid_single intel_ppin cdp_l2 ssbd mba ibrs ibpb stibp ibrs_enhanced tpr_shadow vnmi flexpriority ept vpid ept_ad 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 avx_vnni avx512_bf16 wbnoinvd dtherm ida arat pln pts hwp hwp_act_window hwp_epp hwp_pkg_req avx512vbmi umip pku ospke waitpkg avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg tme avx512_vpopcntdq rdpid bus_lock_detect cldemote movdiri movdir64b enqcmd fsrm md_clear serialize tsxldtrk pconfig arch_lbr amx_bf16 avx512_fp16 amx_tile amx_int8 flush_l1d arch_capabilities
Virtualization:                  VT-x
L1d cache:                       4.5 MiB (96 instances)
L1i cache:                       3 MiB (96 instances)
L2 cache:                        192 MiB (96 instances)
L3 cache:                        195 MiB (2 instances)
NUMA node(s):                    2
NUMA node0 CPU(s):               0-47,96-143
NUMA node1 CPU(s):               48-95,144-191
Vulnerability Itlb multihit:     Not affected
Vulnerability L1tf:              Not affected
Vulnerability Mds:               Not affected
Vulnerability Meltdown:          Not affected
Vulnerability Mmio stale data:   Not affected
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
Vulnerability Srbds:             Not affected
Vulnerability Tsx async abort:   Not affected

Versions of relevant libraries:
[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+cu124
[pip3] torchvision==0.20.1+cu124
[pip3] transformers==4.47.1
[pip3] transformers-stream-generator==0.0.5
[pip3] triton==3.1.0
[conda] Could not collect
ROCM Version: Could not collect
Neuron SDK Version: N/A
vLLM Version: 0.6.6.post2.dev274+g81763c58
vLLM Build Flags:
CUDA Archs: Not Set; ROCm: Disabled; Neuron: Disabled
GPU Topology:
GPU0    GPU1    NIC0    NIC1    NIC2    NIC3    NIC4    NIC5    NIC6    NIC7    CPU Affinity    NUMA Affinity   GPU NUMA ID
GPU0     X      NV18    NODE    NODE    PIX     NODE    SYS     SYS     SYS     SYS     0-47,96-143     0               N/A
GPU1    NV18     X      NODE    NODE    NODE    PIX     SYS     SYS     SYS     SYS     0-47,96-143     0               N/A
NIC0    NODE    NODE     X      NODE    NODE    NODE    SYS     SYS     SYS     SYS
NIC1    NODE    NODE    NODE     X      NODE    NODE    SYS     SYS     SYS     SYS
NIC2    PIX     NODE    NODE    NODE     X      NODE    SYS     SYS     SYS     SYS
NIC3    NODE    PIX     NODE    NODE    NODE     X      SYS     SYS     SYS     SYS
NIC4    SYS     SYS     SYS     SYS     SYS     SYS      X      NODE    NODE    NODE
NIC5    SYS     SYS     SYS     SYS     SYS     SYS     NODE     X      NODE    NODE
NIC6    SYS     SYS     SYS     SYS     SYS     SYS     NODE    NODE     X      NODE
NIC7    SYS     SYS     SYS     SYS     SYS     SYS     NODE    NODE    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_1
  NIC1: mlx5_2
  NIC2: mlx5_3
  NIC3: mlx5_4
  NIC4: mlx5_5
  NIC5: mlx5_6
  NIC6: mlx5_7
  NIC7: mlx5_8

LD_LIBRARY_PATH=/mnt/bn/nas-develop-lyc/mlx/users/yuchen.lyc/src/ui_vllm_deploy/.venv/lib/python3.10/site-packages/cv2/../../lib64:/opt/tiger/yarn_deploy/hadoop/lib/native:/opt/tiger/yarn_deploy/hadoop_current/lib/native:/opt/tiger/yarn_deploy/hadoop_current/lzo/lib:
NVIDIA_VISIBLE_DEVICES=GPU-80bf8a83-cec5-bb7c-3b4f-cd7cbc1dd155,GPU-60c4fea0-31e5-aa88-dfe7-5e570abb533c
NVIDIA_REQUIRE_CUDA=cuda>=10.1
CUDA_HOME=/usr/local/cuda
CUDA_HOME=/usr/local/cuda
NVIDIA_DRIVER_CAPABILITIES=all
CUDA_MODULE_LOADING=LAZY
 

Model Input Dumps

No input.

🐛 Describe the bug

V1 engine works for qwen2-vl only when single gpu, but not tensor-split(multi-gpu).

MAX_PIXELS=1003520
MIN_PIXELS=250880
INFO 01-21 13:45:38 __init__.py:179] Automatically detected platform cuda.
INFO 01-21 13:45:42 api_server.py:768] vLLM API server version 0.6.6.post2.dev274+g81763c58
INFO 01-21 13:45:42 api_server.py:769] args: Namespace(subparser='serve', model_tag='/mnt/bn/nas-develop-lyc/mlx/users/yuchen.lyc/src/ui_vllm_deploy/bin/../model/Qwen/Qwen2-VL-7B-Instruct', config='', host='::', port=11111, uvicorn_log_level='info', allow_credentials=False, allowed_origins=['*'], allowed_methods=['*'], allowed_headers=['*'], api_key=None, lora_modules=None, prompt_adapters=None, chat_template=None, chat_template_content_format='auto', response_role='assistant', ssl_keyfile=None, ssl_certfile=None, ssl_ca_certs=None, ssl_cert_reqs=0, root_path=None, middleware=[], return_tokens_as_token_ids=False, disable_frontend_multiprocessing=False, enable_request_id_headers=False, enable_auto_tool_choice=False, tool_call_parser=None, tool_parser_plugin='', model='/mnt/bn/nas-develop-lyc/mlx/users/yuchen.lyc/src/ui_vllm_deploy/bin/../model/Qwen/Qwen2-VL-7B-Instruct', task='auto', tokenizer=None, skip_tokenizer_init=False, revision=None, code_revision=None, tokenizer_revision=None, tokenizer_mode='auto', trust_remote_code=False, allowed_local_media_path=None, download_dir=None, load_format='auto', config_format=<ConfigFormat.AUTO: 'auto'>, dtype='auto', kv_cache_dtype='auto', quantization_param_path=None, max_model_len=32768, guided_decoding_backend='lm-format-enforcer', logits_processor_pattern=None, distributed_executor_backend=None, worker_use_ray=False, pipeline_parallel_size=1, tensor_parallel_size=2, max_parallel_loading_workers=None, ray_workers_use_nsight=False, block_size=None, enable_prefix_caching=True, disable_sliding_window=False, use_v2_block_manager=True, num_lookahead_slots=0, seed=0, swap_space=4, cpu_offload_gb=0, gpu_memory_utilization=0.9, num_gpu_blocks_override=None, max_num_batched_tokens=None, max_num_seqs=None, max_logprobs=20, disable_log_stats=False, quantization=None, rope_scaling=None, rope_theta=None, hf_overrides=None, enforce_eager=False, max_seq_len_to_capture=8192, disable_custom_all_reduce=False, tokenizer_pool_size=0, tokenizer_pool_type='ray', tokenizer_pool_extra_config=None, limit_mm_per_prompt={'image': 5}, mm_processor_kwargs={'min_pixels': 250880, 'max_pixels': 1003520}, disable_mm_preprocessor_cache=False, enable_lora=False, enable_lora_bias=False, max_loras=1, max_lora_rank=16, lora_extra_vocab_size=256, lora_dtype='auto', long_lora_scaling_factors=None, max_cpu_loras=None, fully_sharded_loras=False, enable_prompt_adapter=False, max_prompt_adapters=1, max_prompt_adapter_token=0, device='auto', num_scheduler_steps=1, multi_step_stream_outputs=True, scheduler_delay_factor=0.0, enable_chunked_prefill=None, speculative_model=None, speculative_model_quantization=None, num_speculative_tokens=None, speculative_disable_mqa_scorer=False, speculative_draft_tensor_parallel_size=None, speculative_max_model_len=None, speculative_disable_by_batch_size=None, ngram_prompt_lookup_max=None, ngram_prompt_lookup_min=None, spec_decoding_acceptance_method='rejection_sampler', typical_acceptance_sampler_posterior_threshold=None, typical_acceptance_sampler_posterior_alpha=None, disable_logprobs_during_spec_decoding=None, model_loader_extra_config=None, ignore_patterns=[], preemption_mode=None, served_model_name=['qwen2-vl-7b-instruct'], qlora_adapter_name_or_path=None, otlp_traces_endpoint=None, collect_detailed_traces=None, disable_async_output_proc=False, scheduling_policy='fcfs', override_neuron_config=None, override_pooler_config=None, compilation_config=None, kv_transfer_config=None, worker_cls='auto', generation_config=None, disable_log_requests=False, max_log_len=None, disable_fastapi_docs=False, enable_prompt_tokens_details=False, dispatch_function=<function serve at 0x7fd55eb38e50>)
WARNING 01-21 13:45:42 arg_utils.py:1283] Setting max_num_batched_tokens to 2048 for OPENAI_API_SERVER usage context.
INFO 01-21 13:45:56 config.py:520] This model supports multiple tasks: {'reward', 'classify', 'generate', 'score', 'embed'}. Defaulting to 'generate'.
INFO 01-21 13:45:56 config.py:1327] Defaulting to use mp for distributed inference
INFO 01-21 13:45:56 config.py:1482] Chunked prefill is enabled with max_num_batched_tokens=2048.
INFO 01-21 13:46:04 __init__.py:179] Automatically detected platform cuda.
INFO 01-21 13:46:08 core.py:45] Initializing an LLM engine (v0.6.6.post2.dev274+g81763c58) with config: model='/mnt/bn/nas-develop-lyc/mlx/users/yuchen.lyc/src/ui_vllm_deploy/bin/../model/Qwen/Qwen2-VL-7B-Instruct', speculative_config=None, tokenizer='/mnt/bn/nas-develop-lyc/mlx/users/yuchen.lyc/src/ui_vllm_deploy/bin/../model/Qwen/Qwen2-VL-7B-Instruct', skip_tokenizer_init=False, tokenizer_mode=auto, revision=None, override_neuron_config=None, tokenizer_revision=None, trust_remote_code=False, dtype=torch.bfloat16, max_seq_len=32768, download_dir=None, load_format=auto, tensor_parallel_size=2, pipeline_parallel_size=1, disable_custom_all_reduce=False, quantization=None, enforce_eager=False, kv_cache_dtype=auto, quantization_param_path=None, device_config=cuda, decoding_config=DecodingConfig(guided_decoding_backend='lm-format-enforcer'), observability_config=ObservabilityConfig(otlp_traces_endpoint=None, collect_model_forward_time=False, collect_model_execute_time=False), seed=0, served_model_name=qwen2-vl-7b-instruct, num_scheduler_steps=1, multi_step_stream_outputs=True, enable_prefix_caching=True, chunked_prefill_enabled=True, use_async_output_proc=True, disable_mm_preprocessor_cache=False, mm_processor_kwargs={'min_pixels': 250880, 'max_pixels': 1003520}, pooler_config=None, compilation_config={"level":3,"custom_ops":["none"],"splitting_ops":["vllm.unified_attention","vllm.unified_attention_with_output"],"use_inductor":true,"candidate_compile_sizes":[],"use_cudagraph":true,"cudagraph_num_of_warmups":1,"compile_sizes":[],"capture_sizes":[512,504,496,488,480,472,464,456,448,440,432,424,416,408,400,392,384,376,368,360,352,344,336,328,320,312,304,296,288,280,272,264,256,248,240,232,224,216,208,200,192,184,176,168,160,152,144,136,128,120,112,104,96,88,80,72,64,56,48,40,32,24,16,8,4,2,1],"max_capture_size":512}
WARNING 01-21 13:46:08 multiproc_worker_utils.py:298] Reducing Torch parallelism from 96 threads to 1 to avoid unnecessary CPU contention. Set OMP_NUM_THREADS in the external environment to tune this value as needed.
INFO 01-21 13:46:08 custom_cache_manager.py:17] Setting Triton cache manager to: vllm.triton_utils.custom_cache_manager:CustomCacheManager
INFO 01-21 13:46:08 shm_broadcast.py:256] vLLM message queue communication handle: Handle(connect_ip='127.0.0.1', local_reader_ranks=[0, 1], buffer_handle=(2, 10485760, 10, 'psm_abae8bd3'), local_subscribe_port=36709, remote_subscribe_port=None)
INFO 01-21 13:46:16 __init__.py:179] Automatically detected platform cuda.
(VllmWorker rank=0 pid=7701) INFO 01-21 13:46:19 shm_broadcast.py:256] vLLM message queue communication handle: Handle(connect_ip='127.0.0.1', local_reader_ranks=[0], buffer_handle=(1, 10485760, 10, 'psm_e2892ec4'), local_subscribe_port=63511, remote_subscribe_port=None)
INFO 01-21 13:46:27 __init__.py:179] Automatically detected platform cuda.
(VllmWorker rank=1 pid=7789) INFO 01-21 13:46:31 shm_broadcast.py:256] vLLM message queue communication handle: Handle(connect_ip='127.0.0.1', local_reader_ranks=[0], buffer_handle=(1, 10485760, 10, 'psm_c1c0dd5b'), local_subscribe_port=47903, remote_subscribe_port=None)
(VllmWorker rank=0 pid=7701) INFO 01-21 13:46:32 utils.py:938] Found nccl from library libnccl.so.2
(VllmWorker rank=1 pid=7789) INFO 01-21 13:46:32 utils.py:938] Found nccl from library libnccl.so.2
(VllmWorker rank=0 pid=7701) INFO 01-21 13:46:32 pynccl.py:67] vLLM is using nccl==2.21.5
(VllmWorker rank=1 pid=7789) INFO 01-21 13:46:32 pynccl.py:67] vLLM is using nccl==2.21.5
(VllmWorker rank=0 pid=7701) INFO 01-21 13:46:32 custom_all_reduce_utils.py:242] reading GPU P2P access cache from /root/.cache/vllm/gpu_p2p_access_cache_for_0,1.json
(VllmWorker rank=1 pid=7789) INFO 01-21 13:46:32 custom_all_reduce_utils.py:242] reading GPU P2P access cache from /root/.cache/vllm/gpu_p2p_access_cache_for_0,1.json
(VllmWorker rank=0 pid=7701) INFO 01-21 13:46:32 shm_broadcast.py:256] vLLM message queue communication handle: Handle(connect_ip='127.0.0.1', local_reader_ranks=[1], buffer_handle=(1, 4194304, 6, 'psm_0220bb80'), local_subscribe_port=50433, remote_subscribe_port=None)
(VllmWorker rank=1 pid=7789) INFO 01-21 13:46:34 gpu_model_runner.py:815] Starting to load model /mnt/bn/nas-develop-lyc/mlx/users/yuchen.lyc/src/ui_vllm_deploy/bin/../model/Qwen/Qwen2-VL-7B-Instruct...
(VllmWorker rank=0 pid=7701) INFO 01-21 13:46:34 gpu_model_runner.py:815] Starting to load model /mnt/bn/nas-develop-lyc/mlx/users/yuchen.lyc/src/ui_vllm_deploy/bin/../model/Qwen/Qwen2-VL-7B-Instruct...
(VllmWorker rank=0 pid=7701) INFO 01-21 13:46:34 cuda.py:157] Using Flash Attention backend on V1 engine.
(VllmWorker rank=1 pid=7789) INFO 01-21 13:46:34 cuda.py:157] Using Flash Attention backend on V1 engine.
(VllmWorker rank=1 pid=7789) WARNING 01-21 13:46:35 topk_topp_sampler.py:44] FlashInfer is not available. Falling back to the PyTorch-native implementation of top-p & top-k sampling. For the best performance, please install FlashInfer.
(VllmWorker rank=0 pid=7701) WARNING 01-21 13:46:35 topk_topp_sampler.py:44] FlashInfer is not available. Falling back to the PyTorch-native implementation of top-p & top-k sampling. For the best performance, please install FlashInfer.
Loading safetensors checkpoint shards:   0% Completed | 0/5 [00:00<?, ?it/s]
Loading safetensors checkpoint shards:  20% Completed | 1/5 [00:00<00:01,  2.06it/s]
Loading safetensors checkpoint shards:  40% Completed | 2/5 [00:01<00:01,  1.86it/s]
Loading safetensors checkpoint shards:  60% Completed | 3/5 [00:01<00:01,  1.80it/s]
Loading safetensors checkpoint shards:  80% Completed | 4/5 [00:02<00:00,  1.73it/s]
Loading safetensors checkpoint shards: 100% Completed | 5/5 [00:02<00:00,  2.43it/s]
Loading safetensors checkpoint shards: 100% Completed | 5/5 [00:02<00:00,  2.11it/s]
(VllmWorker rank=0 pid=7701)
(VllmWorker rank=0 pid=7701) INFO 01-21 13:46:37 gpu_model_runner.py:820] Loading model weights took 7.7541 GB
(VllmWorker rank=1 pid=7789) INFO 01-21 13:46:37 gpu_model_runner.py:820] Loading model weights took 7.7541 GB
(VllmWorker rank=0 pid=7701) INFO 01-21 13:46:37 gpu_model_runner.py:897] Encoder cache will be initialized with a budget of 9800 tokens, and profiled with 1 video items of the maximum feature size.
(VllmWorker rank=1 pid=7789) INFO 01-21 13:46:37 gpu_model_runner.py:897] Encoder cache will be initialized with a budget of 9800 tokens, and profiled with 1 video items of the maximum feature size.
(VllmWorker rank=1 pid=7789) Keyword argument `min_pixels` is not a valid argument for this processor and will be ignored.
(VllmWorker rank=1 pid=7789) Keyword argument `max_pixels` is not a valid argument for this processor and will be ignored.
(VllmWorker rank=0 pid=7701) Keyword argument `min_pixels` is not a valid argument for this processor and will be ignored.
(VllmWorker rank=0 pid=7701) Keyword argument `max_pixels` is not a valid argument for this processor and will be ignored.
(VllmWorker rank=0 pid=7701) It looks like you are trying to rescale already rescaled images. If the input images have pixel values between 0 and 1, set `do_rescale=False` to avoid rescaling them again.
(VllmWorker rank=1 pid=7789) It looks like you are trying to rescale already rescaled images. If the input images have pixel values between 0 and 1, set `do_rescale=False` to avoid rescaling them again.
(VllmWorker rank=0 pid=7701) INFO 01-21 13:46:48 backends.py:571] Using cache directory: /root/.cache/vllm/torch_compile_cache/9f30b8927e/rank_0 for vLLM's torch.compile
(VllmWorker rank=0 pid=7701) INFO 01-21 13:46:48 backends.py:579] Dynamo bytecode transform time: 8.61 s
(VllmWorker rank=1 pid=7789) INFO 01-21 13:46:48 backends.py:571] Using cache directory: /root/.cache/vllm/torch_compile_cache/9f30b8927e/rank_1 for vLLM's torch.compile
(VllmWorker rank=1 pid=7789) INFO 01-21 13:46:48 backends.py:579] Dynamo bytecode transform time: 8.61 s
(VllmWorker rank=0 pid=7701) INFO 01-21 13:46:52 backends.py:309] Cache the graph of shape None for later use
(VllmWorker rank=1 pid=7789) INFO 01-21 13:46:52 backends.py:309] Cache the graph of shape None for later use
(VllmWorker rank=1 pid=7789) INFO 01-21 13:47:13 backends.py:321] Compiling a graph for general shape takes 24.17 s
(VllmWorker rank=0 pid=7701) INFO 01-21 13:47:13 backends.py:321] Compiling a graph for general shape takes 24.34 s
(VllmWorker rank=1 pid=7789) INFO 01-21 13:47:15 monitor.py:31] torch.compile takes 32.78 s in total
(VllmWorker rank=0 pid=7701) INFO 01-21 13:47:15 monitor.py:31] torch.compile takes 32.95 s in total
INFO 01-21 13:47:15 kv_cache_utils.py:395] # GPU blocks: 174004
(VllmWorker rank=1 pid=7789) INFO 01-21 13:47:34 custom_all_reduce.py:224] Registering 3752 cuda graph addresses
(VllmWorker rank=0 pid=7701) INFO 01-21 13:47:36 custom_all_reduce.py:224] Registering 3752 cuda graph addresses
(VllmWorker rank=1 pid=7789) INFO 01-21 13:47:36 gpu_model_runner.py:990] Graph capturing finished in 21 secs, took 1.77 GiB
(VllmWorker rank=0 pid=7701) INFO 01-21 13:47:36 gpu_model_runner.py:990] Graph capturing finished in 21 secs, took 1.77 GiB
INFO 01-21 13:47:36 core.py:89] init engine (profile, create kv cache, warmup model) took 58.62 seconds
ERROR 01-21 13:47:36 core.py:205] EngineCore hit an exception: Traceback (most recent call last):
ERROR 01-21 13:47:36 core.py:205]   File "/mnt/bn/nas-develop-lyc/mlx/users/yuchen.lyc/src/ui_vllm_deploy/.venv/lib/python3.10/site-packages/vllm/multimodal/registry.py", line 94, in __getitem__
ERROR 01-21 13:47:36 core.py:205]     return super().__getitem__(key)
ERROR 01-21 13:47:36 core.py:205]   File "/root/.local/share/uv/python/cpython-3.10.16-linux-x86_64-gnu/lib/python3.10/collections/__init__.py", line 1106, in __getitem__
ERROR 01-21 13:47:36 core.py:205]     raise KeyError(key)
ERROR 01-21 13:47:36 core.py:205] KeyError: <vllm.config.ModelConfig object at 0x7fec41821900>
ERROR 01-21 13:47:36 core.py:205]
ERROR 01-21 13:47:36 core.py:205] The above exception was the direct cause of the following exception:
ERROR 01-21 13:47:36 core.py:205]
ERROR 01-21 13:47:36 core.py:205] Traceback (most recent call last):
ERROR 01-21 13:47:36 core.py:205]   File "/mnt/bn/nas-develop-lyc/mlx/users/yuchen.lyc/src/ui_vllm_deploy/.venv/lib/python3.10/site-packages/vllm/v1/engine/core.py", line 197, in run_engine_core
ERROR 01-21 13:47:36 core.py:205]     engine_core = EngineCoreProc(*args, **kwargs)
ERROR 01-21 13:47:36 core.py:205]   File "/mnt/bn/nas-develop-lyc/mlx/users/yuchen.lyc/src/ui_vllm_deploy/.venv/lib/python3.10/site-packages/vllm/v1/engine/core.py", line 151, in __init__
ERROR 01-21 13:47:36 core.py:205]     super().__init__(vllm_config, executor_class)
ERROR 01-21 13:47:36 core.py:205]   File "/mnt/bn/nas-develop-lyc/mlx/users/yuchen.lyc/src/ui_vllm_deploy/.venv/lib/python3.10/site-packages/vllm/v1/engine/core.py", line 58, in __init__
ERROR 01-21 13:47:36 core.py:205]     self.scheduler = Scheduler(
ERROR 01-21 13:47:36 core.py:205]   File "/mnt/bn/nas-develop-lyc/mlx/users/yuchen.lyc/src/ui_vllm_deploy/.venv/lib/python3.10/site-packages/vllm/v1/core/scheduler.py", line 78, in __init__
ERROR 01-21 13:47:36 core.py:205]     encoder_compute_budget, encoder_cache_size = compute_encoder_budget(
ERROR 01-21 13:47:36 core.py:205]   File "/mnt/bn/nas-develop-lyc/mlx/users/yuchen.lyc/src/ui_vllm_deploy/.venv/lib/python3.10/site-packages/vllm/v1/core/encoder_cache_manager.py", line 83, in compute_encoder_budget
ERROR 01-21 13:47:36 core.py:205]     ) = _compute_encoder_budget_multimodal(model_config, scheduler_config)
ERROR 01-21 13:47:36 core.py:205]   File "/mnt/bn/nas-develop-lyc/mlx/users/yuchen.lyc/src/ui_vllm_deploy/.venv/lib/python3.10/site-packages/vllm/v1/core/encoder_cache_manager.py", line 106, in _compute_encoder_budget_multimodal
ERROR 01-21 13:47:36 core.py:205]     max_tokens_by_modality_dict = MULTIMODAL_REGISTRY.get_max_tokens_per_item_by_nonzero_modality(  # noqa: E501
ERROR 01-21 13:47:36 core.py:205]   File "/mnt/bn/nas-develop-lyc/mlx/users/yuchen.lyc/src/ui_vllm_deploy/.venv/lib/python3.10/site-packages/vllm/multimodal/registry.py", line 285, in get_max_tokens_per_item_by_nonzero_modality
ERROR 01-21 13:47:36 core.py:205]     limits_per_plugin = self._limits_by_model[model_config]
ERROR 01-21 13:47:36 core.py:205]   File "/mnt/bn/nas-develop-lyc/mlx/users/yuchen.lyc/src/ui_vllm_deploy/.venv/lib/python3.10/site-packages/vllm/multimodal/registry.py", line 98, in __getitem__
ERROR 01-21 13:47:36 core.py:205]     raise KeyError(msg) from exc
ERROR 01-21 13:47:36 core.py:205] KeyError: 'Cannot find `mm_limits` for model=/mnt/bn/nas-develop-lyc/mlx/users/yuchen.lyc/src/ui_vllm_deploy/bin/../model/Qwen/Qwen2-VL-7B-Instruct. Did you forget to call `init_mm_limits_per_prompt`?'
ERROR 01-21 13:47:36 core.py:205]
CRITICAL 01-21 13:47:36 core_client.py:146] Got fatal signal from worker processes, shutting down. See stack trace above for root cause issue.
bin/qwen2_vl-deploy_vllm.sh: line 46:  7164 Killed                  vllm serve ${MODEL_DIR}/${MODEL_ID} --served-model-name ${MODEL_TYPE} --host :: --port ${PORT} --guided-decoding-backend lm-format-enforcer --tensor-parallel-size ${NUM_DEVICE} --enable-prefix-caching --gpu-memory-utilization 0.9 --max-model-len ${MAX_MODEL_LEN} --mm-processor-kwargs "{\"min_pixels\": $MIN_PIXELS, \"max_pixels\": $MAX_PIXELS}" --limit-mm-per-prompt image=${NUM_IMAGES_PER_REQUEST}

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.

[DarkLight1337]

Thanks for reporting! I have opened #12252 to fix it, please try it out!