Issue Station
Sign In
Tools hereToolbox

vllm - ✅(Solved) Fix [Bug]: DeepSeek-V4-Flash MTP hangs with `vllm bench serve` when concurrency > 1 on vLLM v0.20.0 [1 pull requests, 1 comments, 2 participants]

Official PRs (…)
ON THIS PAGE

Recommended Tools

×6

Utilities matched from this issue’s tags and category — try them while you read without losing context.

GitHub issue graph ai analysis

Paste a GitHub issue URL. We fetch that issue, discover linked issues from bodies/comments/timeline, collect linked pull requests, and produce a structured English report.

The report is written in English Markdown for sharing and archival.

Helpful · Quick feedback

Loading…
GitHub stats
vllm-project/vllm#41402•Fetched 2026-05-01 05:33:50
View on GitHub
Comments
1
Participants
2
Timeline
6
Reactions
0
Author
jasljasl
Participants
jasljaslwojciech-waiswojciech-wais
Timeline (top)
commented ×1cross-referenced ×1labeled ×1mentioned ×1

Error Message

The C=4 run starts, but remains stuck at 0/32:

Fix Action

Fix / Workaround

============================== CPU Info

Architecture: x86_64 CPU op-mode(s): 32-bit, 64-bit Address sizes: 46 bits physical, 57 bits virtual Byte Order: Little Endian CPU(s): 256 On-line CPU(s) list: 0-255 Vendor ID: GenuineIntel Model name: Intel(R) Xeon(R) 6767P CPU family: 6 Model: 173 Thread(s) per core: 2 Core(s) per socket: 64 Socket(s): 2 Stepping: 1 CPU max MHz: 3900.0000 CPU min MHz: 800.0000 BogoMIPS: 4800.00 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 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 intel_ppin cdp_l2 ssbd mba ibrs ibpb stibp ibrs_enhanced tpr_shadow flexpriority ept vpid ept_ad fsgsbase tsc_adjust sgx 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 user_shstk avx_vnni avx512_bf16 wbnoinvd dtherm ida arat pln pts hwp hwp_act_window hwp_epp hwp_pkg_req hfi vnmi avx512vbmi umip pku ospke waitpkg avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg tme avx512_vpopcntdq la57 rdpid bus_lock_detect cldemote movdiri movdir64b enqcmd sgx_lc fsrm md_clear serialize tsxldtrk pconfig arch_lbr ibt amx_bf16 avx512_fp16 amx_tile amx_int8 flush_l1d arch_capabilities Virtualization: VT-x L1d cache: 6 MiB (128 instances) L1i cache: 8 MiB (128 instances) L2 cache: 256 MiB (128 instances) L3 cache: 672 MiB (2 instances) NUMA node(s): 2 NUMA node0 CPU(s): 0-63,128-191 NUMA node1 CPU(s): 64-127,192-255 Vulnerability Gather data sampling: Not affected Vulnerability Ghostwrite: Not affected Vulnerability Indirect target selection: Not affected Vulnerability Itlb multihit: Not affected Vulnerability L1tf: Not affected Vulnerability Mds: Not affected Vulnerability Meltdown: Not affected Vulnerability Mmio stale data: Not affected Vulnerability Old microcode: Not affected Vulnerability Reg file data sampling: Not affected Vulnerability Retbleed: Not affected 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; PBRSB-eIBRS Not affected; BHI BHI_DIS_S Vulnerability Srbds: Not affected Vulnerability Tsa: Not affected Vulnerability Tsx async abort: Not affected Vulnerability Vmscape: Mitigation; IBPB before exit to userspace

This was reproduced with the official vLLM v0.20.0 installed package restored, before applying any local MTP draft-probability patches.

PR fix notes

PR #41404: [Bugfix] Fix DeepSeek-V4 MTP deadlock with batch size limitation

Description (problem / solution / changelog)

Fixes #41402: DeepSeek-V4-Flash MTP hangs at concurrency > 1

Summary

This PR fixes a critical deadlock issue affecting DeepSeek-V4-Flash models with Multi-Token Prediction (MTP) when running with concurrency > 1 and tensor parallelism > 1. The deadlock causes complete throughput collapse from ~120 tokens/s to 0.

Root Cause

The deadlock occurs in shared memory broadcast coordination during speculative decoding:

  1. Multiple tensor-parallel workers coordinate simultaneously via shared memory broadcast
  2. Ring buffer metadata becomes inconsistent across processes during MTP target hidden state coordination
  3. Some workers finish while others wait indefinitely for broadcast data ("No available shared memory broadcast block found in 60 seconds")
  4. Buffer pool gets exhausted with pending operations

Solution

This fix implements batch size limitation for DeepSeek-V4 MTP models when tensor parallelism is enabled, following the existing disable_padded_drafter_batch pattern:

  • Detects DeepSeek-V4 MTP models via hasattr(model, 'get_mtp_target_hidden_states')
  • Disables speculative decoding when batch_size > 1 AND tensor_parallel_size > 1
  • Preserves normal speculative decoding for single requests and non-parallel configurations

Changes

  • Modified: vllm/v1/worker/gpu_model_runner.py lines 4233-4241
  • Added: mtp_deadlock_avoidance condition with clear documentation
  • Logic: Sets input_fits_in_drafter = False to skip speculative decoding when conditions are met

Testing

  • ✅ Unit tests pass for all logic scenarios
  • ✅ Syntax validation passes
  • ✅ Pre-commit hooks pass
  • ✅ Follows existing architectural patterns

Impact

  • Fixes: Production deadlock affecting DeepSeek-V4 users with high concurrency
  • Performance: Better to have working regular decoding than broken speculative decoding
  • Compatibility: No breaking changes, graceful degradation
  • Risk: Minimal - conservative approach that only affects the speculative decoding decision

This enables reliable operation of DeepSeek-V4 MTP models in production environments.

Changed files

  • vllm/v1/worker/gpu_model_runner.py (modified, +10/-0)

Code Example

Collecting environment information...
==============================
        System Info
==============================
OS                           : Ubuntu 22.04.5 LTS (x86_64)
GCC version                  : (Ubuntu 11.4.0-1ubuntu1~22.04.3) 11.4.0
Clang version                : Could not collect
CMake version                : version 3.22.1
Libc version                 : glibc-2.35

==============================
       PyTorch Info
==============================
PyTorch version              : 2.11.0+cu130
Is debug build               : False
CUDA used to build PyTorch   : 13.0
ROCM used to build PyTorch   : N/A
XPU used to build PyTorch    : N/A

==============================
      Python Environment
==============================
Python version               : 3.12.13 (main, Mar  4 2026, 09:23:07) [GCC 11.4.0] (64-bit runtime)
Python platform              : Linux-6.17.0-14-generic-x86_64-with-glibc2.35
    
==============================
       CUDA / GPU Info
==============================
Is CUDA available            : True
CUDA runtime version         : 13.0.88
CUDA_MODULE_LOADING set to   : 
GPU models and configuration : 
GPU 0: NVIDIA B200
GPU 1: NVIDIA B200
GPU 2: NVIDIA B200
GPU 3: NVIDIA B200

Nvidia driver version        : 595.58.03
cuDNN version                : Could not collect
HIP runtime version          : N/A
MIOpen runtime version       : N/A
Is XNNPACK available         : True

==============================
          CPU Info
==============================
Architecture:                            x86_64
CPU op-mode(s):                          32-bit, 64-bit
Address sizes:                           46 bits physical, 57 bits virtual
Byte Order:                              Little Endian
CPU(s):                                  256
On-line CPU(s) list:                     0-255
Vendor ID:                               GenuineIntel
Model name:                              Intel(R) Xeon(R) 6767P
CPU family:                              6
Model:                                   173
Thread(s) per core:                      2
Core(s) per socket:                      64
Socket(s):                               2
Stepping:                                1
CPU max MHz:                             3900.0000
CPU min MHz:                             800.0000
BogoMIPS:                                4800.00
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 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 intel_ppin cdp_l2 ssbd mba ibrs ibpb stibp ibrs_enhanced tpr_shadow flexpriority ept vpid ept_ad fsgsbase tsc_adjust sgx 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 user_shstk avx_vnni avx512_bf16 wbnoinvd dtherm ida arat pln pts hwp hwp_act_window hwp_epp hwp_pkg_req hfi vnmi avx512vbmi umip pku ospke waitpkg avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg tme avx512_vpopcntdq la57 rdpid bus_lock_detect cldemote movdiri movdir64b enqcmd sgx_lc fsrm md_clear serialize tsxldtrk pconfig arch_lbr ibt amx_bf16 avx512_fp16 amx_tile amx_int8 flush_l1d arch_capabilities
Virtualization:                          VT-x
L1d cache:                               6 MiB (128 instances)
L1i cache:                               8 MiB (128 instances)
L2 cache:                                256 MiB (128 instances)
L3 cache:                                672 MiB (2 instances)
NUMA node(s):                            2
NUMA node0 CPU(s):                       0-63,128-191
NUMA node1 CPU(s):                       64-127,192-255
Vulnerability Gather data sampling:      Not affected
Vulnerability Ghostwrite:                Not affected
Vulnerability Indirect target selection: Not affected
Vulnerability Itlb multihit:             Not affected
Vulnerability L1tf:                      Not affected
Vulnerability Mds:                       Not affected
Vulnerability Meltdown:                  Not affected
Vulnerability Mmio stale data:           Not affected
Vulnerability Old microcode:             Not affected
Vulnerability Reg file data sampling:    Not affected
Vulnerability Retbleed:                  Not affected
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; PBRSB-eIBRS Not affected; BHI BHI_DIS_S
Vulnerability Srbds:                     Not affected
Vulnerability Tsa:                       Not affected
Vulnerability Tsx async abort:           Not affected
Vulnerability Vmscape:                   Mitigation; IBPB before exit to userspace

==============================
Versions of relevant libraries
==============================
[pip3] flashinfer-python==0.6.8.post1
[pip3] numpy==2.2.6
[pip3] nvidia-cublas==13.1.0.3
[pip3] nvidia-cuda-cupti==13.0.85
[pip3] nvidia-cuda-nvrtc==13.0.88
[pip3] nvidia-cuda-runtime==13.0.96
[pip3] nvidia-cudnn-cu13==9.19.0.56
[pip3] nvidia-cudnn-frontend==1.18.0
[pip3] nvidia-cufft==12.0.0.61
[pip3] nvidia-cufile==1.15.1.6
[pip3] nvidia-curand==10.4.0.35
[pip3] nvidia-cusolver==12.0.4.66
[pip3] nvidia-cusparse==12.6.3.3
[pip3] nvidia-cusparselt-cu13==0.8.0
[pip3] nvidia-cutlass-dsl==4.4.2
[pip3] nvidia-cutlass-dsl-libs-base==4.4.2
[pip3] nvidia-ml-py==13.595.45
[pip3] nvidia-nccl-cu13==2.28.9
[pip3] nvidia-nvjitlink==13.0.88
[pip3] nvidia-nvshmem-cu13==3.4.5
[pip3] nvidia-nvtx==13.0.85
[pip3] pyzmq==27.1.0
[pip3] torch==2.11.0+cu130
[pip3] torch_c_dlpack_ext==0.1.5
[pip3] torchaudio==2.11.0+cu130
[pip3] torchvision==0.26.0+cu130
[pip3] transformers==5.6.2
[pip3] triton==3.6.0
[conda] Could not collect

==============================
         vLLM Info
==============================
ROCM Version                 : Could not collect
vLLM Version                 : 0.20.0
vLLM Build Flags:
  CUDA Archs: 7.5 8.0 8.6 8.9 9.0 10.0 12.0+PTX; ROCm: Disabled; XPU: Disabled
GPU Topology:
  	GPU0	GPU1	GPU2	GPU3	NIC0	NIC1	NIC2	NIC3	CPU Affinity	NUMA Affinity	GPU NUMA ID
GPU0	 X 	NV18	NV18	NV18	NODE	NODE	SYS	SYS	0-63,128-191	0		N/A
GPU1	NV18	 X 	NV18	NV18	NODE	NODE	SYS	SYS	0-63,128-191	0		N/A
GPU2	NV18	NV18	 X 	NV18	SYS	SYS	NODE	NODE	64-127,192-255	1		N/A
GPU3	NV18	NV18	NV18	 X 	SYS	SYS	NODE	NODE	64-127,192-255	1		N/A
NIC0	NODE	NODE	SYS	SYS	 X 	PIX	SYS	SYS				
NIC1	NODE	NODE	SYS	SYS	PIX	 X 	SYS	SYS				
NIC2	SYS	SYS	NODE	NODE	SYS	SYS	 X 	PIX				
NIC3	SYS	SYS	NODE	NODE	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_6
  NIC1: mlx5_7
  NIC2: mlx5_12
  NIC3: mlx5_13

==============================
     Environment Variables
==============================
NVIDIA_VISIBLE_DEVICES=void
NVIDIA_REQUIRE_CUDA=cuda>=13.0 brand=unknown,driver>=535,driver<536 brand=grid,driver>=535,driver<536 brand=tesla,driver>=535,driver<536 brand=nvidia,driver>=535,driver<536 brand=quadro,driver>=535,driver<536 brand=quadrortx,driver>=535,driver<536 brand=nvidiartx,driver>=535,driver<536 brand=vapps,driver>=535,driver<536 brand=vpc,driver>=535,driver<536 brand=vcs,driver>=535,driver<536 brand=vws,driver>=535,driver<536 brand=cloudgaming,driver>=535,driver<536 brand=unknown,driver>=550,driver<551 brand=grid,driver>=550,driver<551 brand=tesla,driver>=550,driver<551 brand=nvidia,driver>=550,driver<551 brand=quadro,driver>=550,driver<551 brand=quadrortx,driver>=550,driver<551 brand=nvidiartx,driver>=550,driver<551 brand=vapps,driver>=550,driver<551 brand=vpc,driver>=550,driver<551 brand=vcs,driver>=550,driver<551 brand=vws,driver>=550,driver<551 brand=cloudgaming,driver>=550,driver<551 brand=unknown,driver>=565,driver<566 brand=grid,driver>=565,driver<566 brand=tesla,driver>=565,driver<566 brand=nvidia,driver>=565,driver<566 brand=quadro,driver>=565,driver<566 brand=quadrortx,driver>=565,driver<566 brand=nvidiartx,driver>=565,driver<566 brand=vapps,driver>=565,driver<566 brand=vpc,driver>=565,driver<566 brand=vcs,driver>=565,driver<566 brand=vws,driver>=565,driver<566 brand=cloudgaming,driver>=565,driver<566 brand=unknown,driver>=570,driver<571 brand=grid,driver>=570,driver<571 brand=tesla,driver>=570,driver<571 brand=nvidia,driver>=570,driver<571 brand=quadro,driver>=570,driver<571 brand=quadrortx,driver>=570,driver<571 brand=nvidiartx,driver>=570,driver<571 brand=vapps,driver>=570,driver<571 brand=vpc,driver>=570,driver<571 brand=vcs,driver>=570,driver<571 brand=vws,driver>=570,driver<571 brand=cloudgaming,driver>=570,driver<571 brand=unknown,driver>=575,driver<576 brand=grid,driver>=575,driver<576 brand=tesla,driver>=575,driver<576 brand=nvidia,driver>=575,driver<576 brand=quadro,driver>=575,driver<576 brand=quadrortx,driver>=575,driver<576 brand=nvidiartx,driver>=575,driver<576 brand=vapps,driver>=575,driver<576 brand=vpc,driver>=575,driver<576 brand=vcs,driver>=575,driver<576 brand=vws,driver>=575,driver<576 brand=cloudgaming,driver>=575,driver<576
TORCH_CUDA_ARCH_LIST=7.5 8.0 8.6 8.9 9.0 10.0 12.0+PTX
VLLM_CACHE_ROOT=/workspace/.vllm_cache
NVIDIA_DRIVER_CAPABILITIES=all
VLLM_USAGE_SOURCE=production-docker-image
CUDA_VERSION=13.0.2
VLLM_ENABLE_CUDA_COMPATIBILITY=0
LD_LIBRARY_PATH=/usr/local/nvidia/lib64:/usr/local/nvidia/lib:/usr/local/nvidia/lib64
NVIDIA_CTK_LIBCUDA_DIR=/usr/lib/x86_64-linux-gnu
CUDA_HOME=/usr/local/cuda
CUDA_HOME=/usr/local/cuda
PYTORCH_NVML_BASED_CUDA_CHECK=1
TORCHINDUCTOR_COMPILE_THREADS=1
TORCHINDUCTOR_CACHE_DIR=/tmp/torchinductor_root

---

export HF_HOME=/workspace/.hf_home
export CUDA_VISIBLE_DEVICES=0,1,2,3
export VLLM_LOGGING_LEVEL=DEBUG
export VLLM_TRACE_FUNCTION=1

vllm serve deepseek-ai/DeepSeek-V4-Flash \
  --host 127.0.0.1 \
  --port 8080 \
  --served-model-name deepseek-ai/DeepSeek-V4-Flash \
  --trust-remote-code \
  --kv-cache-dtype fp8 \
  --block-size 256 \
  --tensor-parallel-size 4 \
  --enable-expert-parallel \
  --max-model-len 4096 \
  --gpu-memory-utilization 0.88 \
  --speculative-config '{"method":"mtp","num_speculative_tokens":2}' \
  --compilation-config '{"cudagraph_mode":"FULL_AND_PIECEWISE", "custom_ops":["all"]}' \
  --tokenizer-mode deepseek_v4 \
  --tool-call-parser deepseek_v4 \
  --enable-auto-tool-choice \
  --reasoning-parser deepseek_v4 \
  --generation-config vllm

---

vllm bench serve \
  --backend openai \
  --model deepseek-ai/DeepSeek-V4-Flash \
  --served-model-name deepseek-ai/DeepSeek-V4-Flash \
  --host 127.0.0.1 \
  --port 8080 \
  --endpoint /v1/completions \
  --dataset-name random \
  --random-input-len 1024 \
  --random-output-len 256 \
  --num-prompts 32 \
  --max-concurrency 4 \
  --temperature 1.0 \
  --top-p 1.0 \
  --ignore-eos

---

INFO 04-30 17:35:55 [utils.py:90] Sampling input_len from [1024, 1024] and output_len from [256, 256]
Starting initial single prompt test run...
Skipping endpoint ready check.
Starting main benchmark run...
Traffic request rate: inf
Burstiness factor: 1.0 (Poisson process)
Maximum request concurrency: 4
  0%|          | 0/32 [00:00<?, ?it/s]

---

(APIServer pid=7968) INFO 04-30 17:34:33 [loggers.py:271] Engine 000: Avg prompt throughput: 204.8 tokens/s, Avg generation throughput: 39.9 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 0.0%, Prefix cache hit rate: 0.0%
(APIServer pid=7968) INFO 04-30 17:34:43 [loggers.py:271] Engine 000: Avg prompt throughput: 409.5 tokens/s, Avg generation throughput: 113.3 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 0.0%, Prefix cache hit rate: 0.0%
(APIServer pid=7968) INFO 04-30 17:34:53 [loggers.py:271] Engine 000: Avg prompt throughput: 511.9 tokens/s, Avg generation throughput: 121.0 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 0.0%, Prefix cache hit rate: 0.0%
(APIServer pid=7968) INFO 04-30 17:35:03 [loggers.py:271] Engine 000: Avg prompt throughput: 511.9 tokens/s, Avg generation throughput: 113.8 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 0.0%, Prefix cache hit rate: 0.0%

---

(APIServer pid=7968) INFO 04-30 17:36:03 [loggers.py:271] Engine 000: Avg prompt throughput: 25.6 tokens/s, Avg generation throughput: 0.2 tokens/s, Running: 4 reqs, Waiting: 0 reqs, GPU KV cache usage: 0.4%, Prefix cache hit rate: 8.3%
(APIServer pid=7968) INFO 04-30 17:36:13 [loggers.py:271] Engine 000: Avg prompt throughput: 76.8 tokens/s, Avg generation throughput: 0.4 tokens/s, Running: 4 reqs, Waiting: 0 reqs, GPU KV cache usage: 0.1%, Prefix cache hit rate: 8.3%
(APIServer pid=7968) INFO 04-30 17:36:23 [loggers.py:271] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 0.0 tokens/s, Running: 4 reqs, Waiting: 0 reqs, GPU KV cache usage: 0.1%, Prefix cache hit rate: 8.3%

---

(EngineCore pid=8431) INFO 04-30 17:37:11 [shm_broadcast.py:681] No available shared memory broadcast block found in 60 seconds. This typically happens when some processes are hanging or doing some time-consuming work (e.g. compilation, weight/kv cache quantization).
(EngineCore pid=8431) INFO 04-30 17:38:11 [shm_broadcast.py:681] No available shared memory broadcast block found in 60 seconds. This typically happens when some processes are hanging or doing some time-consuming work (e.g. compilation, weight/kv cache quantization).
(EngineCore pid=8431) INFO 04-30 17:39:11 [shm_broadcast.py:681] No available shared memory broadcast block found in 60 seconds. This typically happens when some processes are hanging or doing some time-consuming work (e.g. compilation, weight/kv cache quantization).
RAW_BUFFERClick to expand / collapse

Your current environment

<details> <summary>The output of <code>python collect_env.py</code></summary> 
Collecting environment information...
==============================
        System Info
==============================
OS                           : Ubuntu 22.04.5 LTS (x86_64)
GCC version                  : (Ubuntu 11.4.0-1ubuntu1~22.04.3) 11.4.0
Clang version                : Could not collect
CMake version                : version 3.22.1
Libc version                 : glibc-2.35

==============================
       PyTorch Info
==============================
PyTorch version              : 2.11.0+cu130
Is debug build               : False
CUDA used to build PyTorch   : 13.0
ROCM used to build PyTorch   : N/A
XPU used to build PyTorch    : N/A

==============================
      Python Environment
==============================
Python version               : 3.12.13 (main, Mar  4 2026, 09:23:07) [GCC 11.4.0] (64-bit runtime)
Python platform              : Linux-6.17.0-14-generic-x86_64-with-glibc2.35
    
==============================
       CUDA / GPU Info
==============================
Is CUDA available            : True
CUDA runtime version         : 13.0.88
CUDA_MODULE_LOADING set to   : 
GPU models and configuration : 
GPU 0: NVIDIA B200
GPU 1: NVIDIA B200
GPU 2: NVIDIA B200
GPU 3: NVIDIA B200

Nvidia driver version        : 595.58.03
cuDNN version                : Could not collect
HIP runtime version          : N/A
MIOpen runtime version       : N/A
Is XNNPACK available         : True

==============================
          CPU Info
==============================
Architecture:                            x86_64
CPU op-mode(s):                          32-bit, 64-bit
Address sizes:                           46 bits physical, 57 bits virtual
Byte Order:                              Little Endian
CPU(s):                                  256
On-line CPU(s) list:                     0-255
Vendor ID:                               GenuineIntel
Model name:                              Intel(R) Xeon(R) 6767P
CPU family:                              6
Model:                                   173
Thread(s) per core:                      2
Core(s) per socket:                      64
Socket(s):                               2
Stepping:                                1
CPU max MHz:                             3900.0000
CPU min MHz:                             800.0000
BogoMIPS:                                4800.00
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 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 intel_ppin cdp_l2 ssbd mba ibrs ibpb stibp ibrs_enhanced tpr_shadow flexpriority ept vpid ept_ad fsgsbase tsc_adjust sgx 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 user_shstk avx_vnni avx512_bf16 wbnoinvd dtherm ida arat pln pts hwp hwp_act_window hwp_epp hwp_pkg_req hfi vnmi avx512vbmi umip pku ospke waitpkg avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg tme avx512_vpopcntdq la57 rdpid bus_lock_detect cldemote movdiri movdir64b enqcmd sgx_lc fsrm md_clear serialize tsxldtrk pconfig arch_lbr ibt amx_bf16 avx512_fp16 amx_tile amx_int8 flush_l1d arch_capabilities
Virtualization:                          VT-x
L1d cache:                               6 MiB (128 instances)
L1i cache:                               8 MiB (128 instances)
L2 cache:                                256 MiB (128 instances)
L3 cache:                                672 MiB (2 instances)
NUMA node(s):                            2
NUMA node0 CPU(s):                       0-63,128-191
NUMA node1 CPU(s):                       64-127,192-255
Vulnerability Gather data sampling:      Not affected
Vulnerability Ghostwrite:                Not affected
Vulnerability Indirect target selection: Not affected
Vulnerability Itlb multihit:             Not affected
Vulnerability L1tf:                      Not affected
Vulnerability Mds:                       Not affected
Vulnerability Meltdown:                  Not affected
Vulnerability Mmio stale data:           Not affected
Vulnerability Old microcode:             Not affected
Vulnerability Reg file data sampling:    Not affected
Vulnerability Retbleed:                  Not affected
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; PBRSB-eIBRS Not affected; BHI BHI_DIS_S
Vulnerability Srbds:                     Not affected
Vulnerability Tsa:                       Not affected
Vulnerability Tsx async abort:           Not affected
Vulnerability Vmscape:                   Mitigation; IBPB before exit to userspace

==============================
Versions of relevant libraries
==============================
[pip3] flashinfer-python==0.6.8.post1
[pip3] numpy==2.2.6
[pip3] nvidia-cublas==13.1.0.3
[pip3] nvidia-cuda-cupti==13.0.85
[pip3] nvidia-cuda-nvrtc==13.0.88
[pip3] nvidia-cuda-runtime==13.0.96
[pip3] nvidia-cudnn-cu13==9.19.0.56
[pip3] nvidia-cudnn-frontend==1.18.0
[pip3] nvidia-cufft==12.0.0.61
[pip3] nvidia-cufile==1.15.1.6
[pip3] nvidia-curand==10.4.0.35
[pip3] nvidia-cusolver==12.0.4.66
[pip3] nvidia-cusparse==12.6.3.3
[pip3] nvidia-cusparselt-cu13==0.8.0
[pip3] nvidia-cutlass-dsl==4.4.2
[pip3] nvidia-cutlass-dsl-libs-base==4.4.2
[pip3] nvidia-ml-py==13.595.45
[pip3] nvidia-nccl-cu13==2.28.9
[pip3] nvidia-nvjitlink==13.0.88
[pip3] nvidia-nvshmem-cu13==3.4.5
[pip3] nvidia-nvtx==13.0.85
[pip3] pyzmq==27.1.0
[pip3] torch==2.11.0+cu130
[pip3] torch_c_dlpack_ext==0.1.5
[pip3] torchaudio==2.11.0+cu130
[pip3] torchvision==0.26.0+cu130
[pip3] transformers==5.6.2
[pip3] triton==3.6.0
[conda] Could not collect

==============================
         vLLM Info
==============================
ROCM Version                 : Could not collect
vLLM Version                 : 0.20.0
vLLM Build Flags:
  CUDA Archs: 7.5 8.0 8.6 8.9 9.0 10.0 12.0+PTX; ROCm: Disabled; XPU: Disabled
GPU Topology:
  	GPU0	GPU1	GPU2	GPU3	NIC0	NIC1	NIC2	NIC3	CPU Affinity	NUMA Affinity	GPU NUMA ID
GPU0	 X 	NV18	NV18	NV18	NODE	NODE	SYS	SYS	0-63,128-191	0		N/A
GPU1	NV18	 X 	NV18	NV18	NODE	NODE	SYS	SYS	0-63,128-191	0		N/A
GPU2	NV18	NV18	 X 	NV18	SYS	SYS	NODE	NODE	64-127,192-255	1		N/A
GPU3	NV18	NV18	NV18	 X 	SYS	SYS	NODE	NODE	64-127,192-255	1		N/A
NIC0	NODE	NODE	SYS	SYS	 X 	PIX	SYS	SYS				
NIC1	NODE	NODE	SYS	SYS	PIX	 X 	SYS	SYS				
NIC2	SYS	SYS	NODE	NODE	SYS	SYS	 X 	PIX				
NIC3	SYS	SYS	NODE	NODE	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_6
  NIC1: mlx5_7
  NIC2: mlx5_12
  NIC3: mlx5_13

==============================
     Environment Variables
==============================
NVIDIA_VISIBLE_DEVICES=void
NVIDIA_REQUIRE_CUDA=cuda>=13.0 brand=unknown,driver>=535,driver<536 brand=grid,driver>=535,driver<536 brand=tesla,driver>=535,driver<536 brand=nvidia,driver>=535,driver<536 brand=quadro,driver>=535,driver<536 brand=quadrortx,driver>=535,driver<536 brand=nvidiartx,driver>=535,driver<536 brand=vapps,driver>=535,driver<536 brand=vpc,driver>=535,driver<536 brand=vcs,driver>=535,driver<536 brand=vws,driver>=535,driver<536 brand=cloudgaming,driver>=535,driver<536 brand=unknown,driver>=550,driver<551 brand=grid,driver>=550,driver<551 brand=tesla,driver>=550,driver<551 brand=nvidia,driver>=550,driver<551 brand=quadro,driver>=550,driver<551 brand=quadrortx,driver>=550,driver<551 brand=nvidiartx,driver>=550,driver<551 brand=vapps,driver>=550,driver<551 brand=vpc,driver>=550,driver<551 brand=vcs,driver>=550,driver<551 brand=vws,driver>=550,driver<551 brand=cloudgaming,driver>=550,driver<551 brand=unknown,driver>=565,driver<566 brand=grid,driver>=565,driver<566 brand=tesla,driver>=565,driver<566 brand=nvidia,driver>=565,driver<566 brand=quadro,driver>=565,driver<566 brand=quadrortx,driver>=565,driver<566 brand=nvidiartx,driver>=565,driver<566 brand=vapps,driver>=565,driver<566 brand=vpc,driver>=565,driver<566 brand=vcs,driver>=565,driver<566 brand=vws,driver>=565,driver<566 brand=cloudgaming,driver>=565,driver<566 brand=unknown,driver>=570,driver<571 brand=grid,driver>=570,driver<571 brand=tesla,driver>=570,driver<571 brand=nvidia,driver>=570,driver<571 brand=quadro,driver>=570,driver<571 brand=quadrortx,driver>=570,driver<571 brand=nvidiartx,driver>=570,driver<571 brand=vapps,driver>=570,driver<571 brand=vpc,driver>=570,driver<571 brand=vcs,driver>=570,driver<571 brand=vws,driver>=570,driver<571 brand=cloudgaming,driver>=570,driver<571 brand=unknown,driver>=575,driver<576 brand=grid,driver>=575,driver<576 brand=tesla,driver>=575,driver<576 brand=nvidia,driver>=575,driver<576 brand=quadro,driver>=575,driver<576 brand=quadrortx,driver>=575,driver<576 brand=nvidiartx,driver>=575,driver<576 brand=vapps,driver>=575,driver<576 brand=vpc,driver>=575,driver<576 brand=vcs,driver>=575,driver<576 brand=vws,driver>=575,driver<576 brand=cloudgaming,driver>=575,driver<576
TORCH_CUDA_ARCH_LIST=7.5 8.0 8.6 8.9 9.0 10.0 12.0+PTX
VLLM_CACHE_ROOT=/workspace/.vllm_cache
NVIDIA_DRIVER_CAPABILITIES=all
VLLM_USAGE_SOURCE=production-docker-image
CUDA_VERSION=13.0.2
VLLM_ENABLE_CUDA_COMPATIBILITY=0
LD_LIBRARY_PATH=/usr/local/nvidia/lib64:/usr/local/nvidia/lib:/usr/local/nvidia/lib64
NVIDIA_CTK_LIBCUDA_DIR=/usr/lib/x86_64-linux-gnu
CUDA_HOME=/usr/local/cuda
CUDA_HOME=/usr/local/cuda
PYTORCH_NVML_BASED_CUDA_CHECK=1
TORCHINDUCTOR_COMPILE_THREADS=1
TORCHINDUCTOR_CACHE_DIR=/tmp/torchinductor_root
</details>

🐛 Describe the bug

On official vLLM v0.20.0, DeepSeek-V4-Flash with MTP (num_speculative_tokens=2) makes progress at serving benchmark concurrency 1, but appears to hang at concurrency 4. The benchmark remains at 0/32; the server stays alive with 4 running requests, generation throughput drops to 0, and the engine later reports repeated No available shared memory broadcast block found in 60 seconds messages.

This was reproduced with the official vLLM v0.20.0 installed package restored, before applying any local MTP draft-probability patches.

Minimal reproduction

This uses only vLLM's built-in random benchmark dataset, so no external dataset is required. The model weights still need to be available from Hugging Face or local cache.

Terminal 1:

export HF_HOME=/workspace/.hf_home
export CUDA_VISIBLE_DEVICES=0,1,2,3
export VLLM_LOGGING_LEVEL=DEBUG
export VLLM_TRACE_FUNCTION=1

vllm serve deepseek-ai/DeepSeek-V4-Flash \
  --host 127.0.0.1 \
  --port 8080 \
  --served-model-name deepseek-ai/DeepSeek-V4-Flash \
  --trust-remote-code \
  --kv-cache-dtype fp8 \
  --block-size 256 \
  --tensor-parallel-size 4 \
  --enable-expert-parallel \
  --max-model-len 4096 \
  --gpu-memory-utilization 0.88 \
  --speculative-config '{"method":"mtp","num_speculative_tokens":2}' \
  --compilation-config '{"cudagraph_mode":"FULL_AND_PIECEWISE", "custom_ops":["all"]}' \
  --tokenizer-mode deepseek_v4 \
  --tool-call-parser deepseek_v4 \
  --enable-auto-tool-choice \
  --reasoning-parser deepseek_v4 \
  --generation-config vllm

Terminal 2, after /health is ready:

vllm bench serve \
  --backend openai \
  --model deepseek-ai/DeepSeek-V4-Flash \
  --served-model-name deepseek-ai/DeepSeek-V4-Flash \
  --host 127.0.0.1 \
  --port 8080 \
  --endpoint /v1/completions \
  --dataset-name random \
  --random-input-len 1024 \
  --random-output-len 256 \
  --num-prompts 32 \
  --max-concurrency 4 \
  --temperature 1.0 \
  --top-p 1.0 \
  --ignore-eos

For comparison, changing only --max-concurrency 4 to --max-concurrency 1 completes on the same server.

Expected behavior

The benchmark should make forward progress at concurrency > 1, or fail with a clear error if the current DeepSeek-V4 MTP path does not support this concurrency.

Observed behavior

The C=4 run starts, but remains stuck at 0/32:

INFO 04-30 17:35:55 [utils.py:90] Sampling input_len from [1024, 1024] and output_len from [256, 256]
Starting initial single prompt test run...
Skipping endpoint ready check.
Starting main benchmark run...
Traffic request rate: inf
Burstiness factor: 1.0 (Poisson process)
Maximum request concurrency: 4
  0%|          | 0/32 [00:00<?, ?it/s]

The server had just completed a C=1 benchmark successfully before this. During the C=1 run, normal generation throughput was reported:

(APIServer pid=7968) INFO 04-30 17:34:33 [loggers.py:271] Engine 000: Avg prompt throughput: 204.8 tokens/s, Avg generation throughput: 39.9 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 0.0%, Prefix cache hit rate: 0.0%
(APIServer pid=7968) INFO 04-30 17:34:43 [loggers.py:271] Engine 000: Avg prompt throughput: 409.5 tokens/s, Avg generation throughput: 113.3 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 0.0%, Prefix cache hit rate: 0.0%
(APIServer pid=7968) INFO 04-30 17:34:53 [loggers.py:271] Engine 000: Avg prompt throughput: 511.9 tokens/s, Avg generation throughput: 121.0 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 0.0%, Prefix cache hit rate: 0.0%
(APIServer pid=7968) INFO 04-30 17:35:03 [loggers.py:271] Engine 000: Avg prompt throughput: 511.9 tokens/s, Avg generation throughput: 113.8 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 0.0%, Prefix cache hit rate: 0.0%

When the C=4 benchmark starts, four requests remain running and generation throughput drops to almost zero:

(APIServer pid=7968) INFO 04-30 17:36:03 [loggers.py:271] Engine 000: Avg prompt throughput: 25.6 tokens/s, Avg generation throughput: 0.2 tokens/s, Running: 4 reqs, Waiting: 0 reqs, GPU KV cache usage: 0.4%, Prefix cache hit rate: 8.3%
(APIServer pid=7968) INFO 04-30 17:36:13 [loggers.py:271] Engine 000: Avg prompt throughput: 76.8 tokens/s, Avg generation throughput: 0.4 tokens/s, Running: 4 reqs, Waiting: 0 reqs, GPU KV cache usage: 0.1%, Prefix cache hit rate: 8.3%
(APIServer pid=7968) INFO 04-30 17:36:23 [loggers.py:271] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 0.0 tokens/s, Running: 4 reqs, Waiting: 0 reqs, GPU KV cache usage: 0.1%, Prefix cache hit rate: 8.3%

After that, the engine repeatedly reports shared-memory broadcast stalls:

(EngineCore pid=8431) INFO 04-30 17:37:11 [shm_broadcast.py:681] No available shared memory broadcast block found in 60 seconds. This typically happens when some processes are hanging or doing some time-consuming work (e.g. compilation, weight/kv cache quantization).
(EngineCore pid=8431) INFO 04-30 17:38:11 [shm_broadcast.py:681] No available shared memory broadcast block found in 60 seconds. This typically happens when some processes are hanging or doing some time-consuming work (e.g. compilation, weight/kv cache quantization).
(EngineCore pid=8431) INFO 04-30 17:39:11 [shm_broadcast.py:681] No available shared memory broadcast block found in 60 seconds. This typically happens when some processes are hanging or doing some time-consuming work (e.g. compilation, weight/kv cache quantization).

No Python exception or CUDA traceback was printed before the run was manually terminated.

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.

extent analysis

TL;DR

The issue is likely due to a concurrency limitation in the DeepSeek-V4 MTP path, causing the benchmark to hang at concurrency 4, and can be mitigated by adjusting the speculative configuration or concurrency level.

Guidance

  • Verify that the DeepSeek-V4 MTP path supports concurrency greater than 1 by checking the documentation or source code.
  • Adjust the --max-concurrency parameter to a lower value (e.g., 2) to see if the benchmark completes successfully.
  • Consider modifying the speculative configuration, such as changing the num_speculative_tokens value, to improve concurrency support.
  • Investigate the shared-memory broadcast stalls reported by the engine, which may indicate a resource contention or synchronization issue.

Example

No code snippet is provided as the issue is related to configuration and concurrency rather than code.

Notes

The exact cause of the issue is unclear, and further investigation is needed to determine the root cause. The provided information suggests a potential concurrency limitation in the DeepSeek-V4 MTP path.

Recommendation

Apply a workaround by adjusting the --max-concurrency parameter to a lower value (e.g., 2) to see if the benchmark completes successfully, as the current configuration may not be supported by the DeepSeek-V4 MTP path.

Vote matrix · Quick signals

Works
Did the solution work? Tap to confirm.
Easy Fix
Was it a quick fix?
Time Saver
Did it save you time?
Blocking
Was it severely blocking?
Common Issue
Are others likely hitting this too?
Flaky / Intermittent
Is it intermittent?
Verified / Reproducible
Can you reproduce it reliably?
Loading…

FAQ

Expected behavior

The benchmark should make forward progress at concurrency > 1, or fail with a clear error if the current DeepSeek-V4 MTP path does not support this concurrency.

#api#installation#tensor shape#autograd error#environment variable

Still need to ship something?

×6

Another batch ranked right after the header list — different links, same matching logic.

Back to top recommendations

TRENDING

vllm - ✅(Solved) Fix [Bug]: DeepSeek-V4-Flash MTP hangs with `vllm bench serve` when concurrency > 1 on vLLM v0.20.0 [1 pull requests, 1 comments, 2 participants]