Issue Station
Sign In
Tools hereToolbox

vllm - ✅(Solved) Fix [Bug]: RuntimeError triton error during profile_run with Qwen3.5-MoE vision encoder on ROCm [1 pull requests, 2 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#37992•Fetched 2026-04-08 01:22:07
View on GitHub
Comments
2
Participants
2
Timeline
13
Reactions
0
Author
xuebwang-amdxuebwang-amd
Participants
github-actions[bot]github-actions[bot]xuebwang-amdxuebwang-amd
Timeline (top)
mentioned ×3subscribed ×3commented ×2labeled ×2

Error Message

Traceback (most recent call last): File "/workspace/vllm/vllm/v1/executor/multiproc_executor.py", line 944, in worker_busy_loop output = func(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.12/dist-packages/torch/utils/_contextlib.py", line 120, in decorate_context return func(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^ File "/workspace/vllm/vllm/v1/worker/gpu_worker.py", line 370, in determine_available_memory self.model_runner.profile_run() File "/workspace/vllm/vllm/v1/worker/gpu_model_runner.py", line 5536, in profile_run dummy_encoder_outputs = self.model.embed_multimodal( ^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/workspace/vllm/vllm/model_executor/models/qwen3_vl.py", line 2398, in embed_multimodal image_embeddings = self._process_image_input(multimodal_input) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/workspace/vllm/vllm/model_executor/models/qwen3_vl.py", line 1818, in _process_image_input image_embeds = self.visual(pixel_values, grid_thw=grid_thw) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1775, in _wrapped_call_impl return self._call_impl(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1786, in _call_impl return forward_call(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/workspace/vllm/vllm/model_executor/models/qwen3_vl.py", line 639, in forward hidden_states = blk( ^^^^ File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1775, in _wrapped_call_impl return self._call_impl(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1786, in _call_impl return forward_call(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/workspace/vllm/vllm/model_executor/models/qwen3_vl.py", line 256, in forward x = x + self.attn( ^^^^^^^^^^ File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1775, in _wrapped_call_impl return self._call_impl(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1786, in _call_impl return forward_call(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/workspace/vllm/vllm/model_executor/models/qwen2_5_vl.py", line 389, in forward qk_rotated = self.apply_rotary_emb( ^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1775, in _wrapped_call_impl return self._call_impl(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1786, in _call_impl return forward_call(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/workspace/vllm/vllm/model_executor/custom_op.py", line 136, in forward return self._forward_method(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/workspace/vllm/vllm/model_executor/layers/rotary_embedding/common.py", line 266, in forward_hip output = self.apply_rotary_emb_flash_attn( ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.12/dist-packages/flash_attn/ops/triton/rotary.py", line 159, in apply_rotary torch.library.wrap_triton(rotary_kernel)[grid]( File "/usr/local/lib/python3.12/dist-packages/torch/_higher_order_ops/triton_kernel_wrap.py", line 2049, in call return tracing_triton_hopifier_singleton.call_triton_kernel( ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.12/dist-packages/torch/_higher_order_ops/triton_kernel_wrap.py", line 1893, in call_triton_kernel return self.call_HOP(variable, grids, combined_args_raw, tx) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.12/dist-packages/torch/_higher_order_ops/triton_kernel_wrap.py", line 2003, in call_HOP return triton_kernel_wrapper_mutation( ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.12/dist-packages/torch/_higher_order_ops/triton_kernel_wrap.py", line 980, in call return super().call( ^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.12/dist-packages/torch/_ops.py", line 536, in call return wrapper() ^^^^^^^^^ File "/usr/local/lib/python3.12/dist-packages/torch/_ops.py", line 532, in wrapper return self.dispatch( ^^^^^^^^^^^^^^ File "/usr/local/lib/python3.12/dist-packages/torch/_ops.py", line 520, in dispatch return kernel(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.12/dist-packages/torch/_higher_order_ops/triton_kernel_wrap.py", line 1094, in triton_kernel_wrapper_mutation_dense kernel[grid_fn](*args, **kwargs, **constant_args) File "/usr/local/lib/python3.12/dist-packages/triton/runtime/jit.py", line 393, in <lambda> return lambda *args, **kwargs: self.run(grid=grid, warmup=False, *args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.12/dist-packages/triton/runtime/jit.py", line 623, in run kernel.run(grid_0, grid_1, grid_2, stream, kernel.function, kernel.packed_metadata, launch_metadata, File "/usr/local/lib/python3.12/dist-packages/triton/backends/amd/driver.py", line 647, in call self.launch(self.launch_cooperative_grid, gridX, gridY, gridZ, stream, function, profile_scratch, *args) RuntimeError: Triton Error [HIP]: Code: 1, Messsage: invalid argument

Root Cause

Root Cause The Qwen3.5 vision encoder has hidden_size=1152 and num_heads=16, giving head_dim=72 (not a power of 2) which is is invalid on MI325X.

Fix Action

Fix / Workaround

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

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): 256 On-line CPU(s) list: 0-255 Vendor ID: AuthenticAMD Model name: AMD EPYC 9555 64-Core Processor CPU family: 26 Model: 2 Thread(s) per core: 2 Core(s) per socket: 64 Socket(s): 2 Stepping: 1 Frequency boost: enabled CPU max MHz: 4409.3750 CPU min MHz: 1500.0000 BogoMIPS: 6399.98 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 amd_lbr_v2 nopl nonstop_tsc cpuid extd_apicid aperfmperf rapl pni pclmulqdq monitor ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand lahf_lm cmp_legacy svm extapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ibs skinit wdt tce topoext perfctr_core perfctr_nb bpext perfctr_llc mwaitx cpb cat_l3 cdp_l3 hw_pstate ssbd mba perfmon_v2 ibrs ibpb stibp ibrs_enhanced vmmcall fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid cqm rdt_a avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local user_shstk avx_vnni avx512_bf16 clzero irperf xsaveerptr rdpru wbnoinvd amd_ppin cppc amd_ibpb_ret arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold avic v_vmsave_vmload vgif x2avic v_spec_ctrl vnmi avx512vbmi umip pku ospke avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg avx512_vpopcntdq la57 rdpid bus_lock_detect movdiri movdir64b overflow_recov succor smca fsrm avx512_vp2intersect flush_l1d debug_swap Virtualization: AMD-V L1d cache: 6 MiB (128 instances) L1i cache: 4 MiB (128 instances) L2 cache: 128 MiB (128 instances) L3 cache: 512 MiB (16 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 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 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; STIBP always-on; PBRSB-eIBRS Not affected; BHI Not affected Vulnerability Srbds: Not affected Vulnerability Tsa: Not affected Vulnerability Tsx async abort: Not affected Vulnerability Vmscape: Not affected

Error message The server crashes during engine initialization at the memory-profiling stage (profile_run), before any user request is served. The profiling run creates dummy multimodal inputs and forwards them through the Qwen3.5 vision encoder (ViT). All 8 TP workers fail simultaneously with the same error:

 Traceback (most recent call last):
   File "/workspace/vllm/vllm/v1/executor/multiproc_executor.py", line 944, in worker_busy_loop
     output = func(*args, **kwargs)
              ^^^^^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/utils/_contextlib.py", line 120, in decorate_context
     return func(*args, **kwargs)
            ^^^^^^^^^^^^^^^^^^^^^
   File "/workspace/vllm/vllm/v1/worker/gpu_worker.py", line 370, in determine_available_memory
     self.model_runner.profile_run()
   File "/workspace/vllm/vllm/v1/worker/gpu_model_runner.py", line 5536, in profile_run
     dummy_encoder_outputs = self.model.embed_multimodal(
                             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/workspace/vllm/vllm/model_executor/models/qwen3_vl.py", line 2398, in embed_multimodal
     image_embeddings = self._process_image_input(multimodal_input)
                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/workspace/vllm/vllm/model_executor/models/qwen3_vl.py", line 1818, in _process_image_input
     image_embeds = self.visual(pixel_values, grid_thw=grid_thw)
                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1775, in _wrapped_call_impl
     return self._call_impl(*args, **kwargs)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1786, in _call_impl
     return forward_call(*args, **kwargs)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/workspace/vllm/vllm/model_executor/models/qwen3_vl.py", line 639, in forward
     hidden_states = blk(
                     ^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1775, in _wrapped_call_impl
     return self._call_impl(*args, **kwargs)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1786, in _call_impl
     return forward_call(*args, **kwargs)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/workspace/vllm/vllm/model_executor/models/qwen3_vl.py", line 256, in forward
     x = x + self.attn(
             ^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1775, in _wrapped_call_impl
     return self._call_impl(*args, **kwargs)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1786, in _call_impl
     return forward_call(*args, **kwargs)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/workspace/vllm/vllm/model_executor/models/qwen2_5_vl.py", line 389, in forward
     qk_rotated = self.apply_rotary_emb(
                  ^^^^^^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1775, in _wrapped_call_impl
     return self._call_impl(*args, **kwargs)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1786, in _call_impl
     return forward_call(*args, **kwargs)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/workspace/vllm/vllm/model_executor/custom_op.py", line 136, in forward
     return self._forward_method(*args, **kwargs)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/workspace/vllm/vllm/model_executor/layers/rotary_embedding/common.py", line 266, in forward_hip
     output = self.apply_rotary_emb_flash_attn(
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/flash_attn/ops/triton/rotary.py", line 159, in apply_rotary
     torch.library.wrap_triton(rotary_kernel)[grid](
   File "/usr/local/lib/python3.12/dist-packages/torch/_higher_order_ops/triton_kernel_wrap.py", line 2049, in __call__
     return tracing_triton_hopifier_singleton.call_triton_kernel(
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/_higher_order_ops/triton_kernel_wrap.py", line 1893, in call_triton_kernel
     return self.call_HOP(variable, grids, combined_args_raw, tx)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/_higher_order_ops/triton_kernel_wrap.py", line 2003, in call_HOP
     return triton_kernel_wrapper_mutation(
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/_higher_order_ops/triton_kernel_wrap.py", line 980, in __call__
     return super().__call__(
            ^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/_ops.py", line 536, in __call__
     return wrapper()
            ^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/_ops.py", line 532, in wrapper
     return self.dispatch(
            ^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/_ops.py", line 520, in dispatch
     return kernel(*args, **kwargs)
            ^^^^^^^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/_higher_order_ops/triton_kernel_wrap.py", line 1094, in triton_kernel_wrapper_mutation_dense
     kernel[grid_fn](*args, **kwargs, **constant_args)
   File "/usr/local/lib/python3.12/dist-packages/triton/runtime/jit.py", line 393, in <lambda>
     return lambda *args, **kwargs: self.run(grid=grid, warmup=False, *args, **kwargs)
                                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/triton/runtime/jit.py", line 623, in run
     kernel.run(grid_0, grid_1, grid_2, stream, kernel.function, kernel.packed_metadata, launch_metadata,
   File "/usr/local/lib/python3.12/dist-packages/triton/backends/amd/driver.py", line 647, in __call__
     self.launch(self.launch_cooperative_grid, gridX, gridY, gridZ, stream, function, profile_scratch, *args)
 RuntimeError: Triton Error [HIP]:  Code: 1, Messsage: invalid argument

PR fix notes

PR #37993: [ROCm] Fall back to native rotary embedding when flash_attn triton kernel fails

Description (problem / solution / changelog)

Purpose

To fix the reported issue https://github.com/vllm-project/vllm/issues/37992

Test Plan

Example model: amd/Qwen3.5-397B-A17B-MTP-PTPC-FP8

Test Result

Changed files

  • vllm/model_executor/layers/rotary_embedding/common.py (modified, +12/-3)

Code Example

Collecting environment information...
==============================
        System Info
==============================
OS                           : Ubuntu 22.04.5 LTS (x86_64)
GCC version                  : (Ubuntu 11.4.0-1ubuntu1~22.04.2) 11.4.0
Clang version                : 20.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-7.0.0 25314 f4087f6b428f0e6f575ebac8a8a724dab123d06e)
CMake version                : version 3.31.10
Libc version                 : glibc-2.35

==============================
       PyTorch Info
==============================
PyTorch version              : 2.9.1+git8907517
Is debug build               : False
CUDA used to build PyTorch   : N/A
ROCM used to build PyTorch   : 7.0.51831-a3e329ad8

==============================
      Python Environment
==============================
Python version               : 3.12.12 (main, Oct 10 2025, 08:52:57) [GCC 11.4.0] (64-bit runtime)
Python platform              : Linux-6.8.0-101-generic-x86_64-with-glibc2.35

==============================
       CUDA / GPU Info
==============================
Is CUDA available            : True
CUDA runtime version         : Could not collect
CUDA_MODULE_LOADING set to   : 
GPU models and configuration :  (gfx942:sramecc+:xnack-)
Nvidia driver version        : Could not collect
cuDNN version                : Could not collect
HIP runtime version          : 7.0.51831
MIOpen runtime version       : 3.5.0
Is XNNPACK available         : True

==============================
          CPU Info
==============================
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):                                  256
On-line CPU(s) list:                     0-255
Vendor ID:                               AuthenticAMD
Model name:                              AMD EPYC 9555 64-Core Processor
CPU family:                              26
Model:                                   2
Thread(s) per core:                      2
Core(s) per socket:                      64
Socket(s):                               2
Stepping:                                1
Frequency boost:                         enabled
CPU max MHz:                             4409.3750
CPU min MHz:                             1500.0000
BogoMIPS:                                6399.98
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 amd_lbr_v2 nopl nonstop_tsc cpuid extd_apicid aperfmperf rapl pni pclmulqdq monitor ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand lahf_lm cmp_legacy svm extapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ibs skinit wdt tce topoext perfctr_core perfctr_nb bpext perfctr_llc mwaitx cpb cat_l3 cdp_l3 hw_pstate ssbd mba perfmon_v2 ibrs ibpb stibp ibrs_enhanced vmmcall fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid cqm rdt_a avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local user_shstk avx_vnni avx512_bf16 clzero irperf xsaveerptr rdpru wbnoinvd amd_ppin cppc amd_ibpb_ret arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold avic v_vmsave_vmload vgif x2avic v_spec_ctrl vnmi avx512vbmi umip pku ospke avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg avx512_vpopcntdq la57 rdpid bus_lock_detect movdiri movdir64b overflow_recov succor smca fsrm avx512_vp2intersect flush_l1d debug_swap
Virtualization:                          AMD-V
L1d cache:                               6 MiB (128 instances)
L1i cache:                               4 MiB (128 instances)
L2 cache:                                128 MiB (128 instances)
L3 cache:                                512 MiB (16 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 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 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; STIBP always-on; PBRSB-eIBRS Not affected; BHI Not affected
Vulnerability Srbds:                     Not affected
Vulnerability Tsa:                       Not affected
Vulnerability Tsx async abort:           Not affected
Vulnerability Vmscape:                   Not affected

==============================
Versions of relevant libraries
==============================
[pip3] conch-triton-kernels==1.2.1
[pip3] numpy==2.1.3
[pip3] onnx==1.19.0
[pip3] onnx-ir==0.2.0
[pip3] onnxscript==0.6.2
[pip3] onnxslim==0.1.86
[pip3] pyzmq==27.1.0
[pip3] torch==2.9.1+git8907517
[pip3] torchaudio==2.9.0+eaa9e4e
[pip3] torchvision==0.24.1+d801a34
[pip3] transformers==4.57.6
[pip3] triton==3.4.0
[pip3] triton_kernels==1.0.0
[conda] Could not collect

==============================
         vLLM Info
==============================
ROCM Version                 : 7.0.51831-a3e329ad8
vLLM Version                 : 0.17.1rc1.dev34+g0b4aea4d4.d20260314 (git sha: 0b4aea4d4, date: 20260314)
vLLM Build Flags:
  CUDA Archs: Not Set; ROCm: Disabled
GPU Topology:
  ============================ ROCm System Management Interface ============================
================================ Weight between two GPUs =================================
       GPU0         GPU1         GPU2         GPU3         GPU4         GPU5         GPU6         GPU7         
GPU0   0            15           15           15           15           15           15           15           
GPU1   15           0            15           15           15           15           15           15           
GPU2   15           15           0            15           15           15           15           15           
GPU3   15           15           15           0            15           15           15           15           
GPU4   15           15           15           15           0            15           15           15           
GPU5   15           15           15           15           15           0            15           15           
GPU6   15           15           15           15           15           15           0            15           
GPU7   15           15           15           15           15           15           15           0            

================================= Hops between two GPUs ==================================
       GPU0         GPU1         GPU2         GPU3         GPU4         GPU5         GPU6         GPU7         
GPU0   0            1            1            1            1            1            1            1            
GPU1   1            0            1            1            1            1            1            1            
GPU2   1            1            0            1            1            1            1            1            
GPU3   1            1            1            0            1            1            1            1            
GPU4   1            1            1            1            0            1            1            1            
GPU5   1            1            1            1            1            0            1            1            
GPU6   1            1            1            1            1            1            0            1            
GPU7   1            1            1            1            1            1            1            0            

=============================== Link Type between two GPUs ===============================
       GPU0         GPU1         GPU2         GPU3         GPU4         GPU5         GPU6         GPU7         
GPU0   0            XGMI         XGMI         XGMI         XGMI         XGMI         XGMI         XGMI         
GPU1   XGMI         0            XGMI         XGMI         XGMI         XGMI         XGMI         XGMI         
GPU2   XGMI         XGMI         0            XGMI         XGMI         XGMI         XGMI         XGMI         
GPU3   XGMI         XGMI         XGMI         0            XGMI         XGMI         XGMI         XGMI         
GPU4   XGMI         XGMI         XGMI         XGMI         0            XGMI         XGMI         XGMI         
GPU5   XGMI         XGMI         XGMI         XGMI         XGMI         0            XGMI         XGMI         
GPU6   XGMI         XGMI         XGMI         XGMI         XGMI         XGMI         0            XGMI         
GPU7   XGMI         XGMI         XGMI         XGMI         XGMI         XGMI         XGMI         0            

======================================= Numa Nodes =======================================
GPU[0]		: (Topology) Numa Node: 0
GPU[0]		: (Topology) Numa Affinity: 0
GPU[1]		: (Topology) Numa Node: 0
GPU[1]		: (Topology) Numa Affinity: 0
GPU[2]		: (Topology) Numa Node: 0
GPU[2]		: (Topology) Numa Affinity: 0
GPU[3]		: (Topology) Numa Node: 0
GPU[3]		: (Topology) Numa Affinity: 0
GPU[4]		: (Topology) Numa Node: 1
GPU[4]		: (Topology) Numa Affinity: 1
GPU[5]		: (Topology) Numa Node: 1
GPU[5]		: (Topology) Numa Affinity: 1
GPU[6]		: (Topology) Numa Node: 1
GPU[6]		: (Topology) Numa Affinity: 1
GPU[7]		: (Topology) Numa Node: 1
GPU[7]		: (Topology) Numa Affinity: 1
================================== End of ROCm SMI Log ===================================

==============================
     Environment Variables
==============================
PYTORCH_ROCM_ARCH=gfx90a;gfx942;gfx950;gfx1100;gfx1101;gfx1200;gfx1201;gfx1150;gfx1151
LD_LIBRARY_PATH=/opt/rocm/lib:/usr/local/lib:
PYTORCH_NVML_BASED_CUDA_CHECK=1
TORCHINDUCTOR_COMPILE_THREADS=1

---

vllm serve amd/Qwen3.5-397B-A17B-MTP-PTPC-FP8 \
    --tensor-parallel-size 8 \
    --dtype auto \
    --enforce-eager \
    --no-enable-chunked-prefill \
    --speculative-config '{"method": "mtp", "num_speculative_tokens": 1}' \
    --gpu-memory-utilization 0.95 \
    --port 8008

---

Traceback (most recent call last):
   File "/workspace/vllm/vllm/v1/executor/multiproc_executor.py", line 944, in worker_busy_loop
     output = func(*args, **kwargs)
              ^^^^^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/utils/_contextlib.py", line 120, in decorate_context
     return func(*args, **kwargs)
            ^^^^^^^^^^^^^^^^^^^^^
   File "/workspace/vllm/vllm/v1/worker/gpu_worker.py", line 370, in determine_available_memory
     self.model_runner.profile_run()
   File "/workspace/vllm/vllm/v1/worker/gpu_model_runner.py", line 5536, in profile_run
     dummy_encoder_outputs = self.model.embed_multimodal(
                             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/workspace/vllm/vllm/model_executor/models/qwen3_vl.py", line 2398, in embed_multimodal
     image_embeddings = self._process_image_input(multimodal_input)
                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/workspace/vllm/vllm/model_executor/models/qwen3_vl.py", line 1818, in _process_image_input
     image_embeds = self.visual(pixel_values, grid_thw=grid_thw)
                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1775, in _wrapped_call_impl
     return self._call_impl(*args, **kwargs)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1786, in _call_impl
     return forward_call(*args, **kwargs)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/workspace/vllm/vllm/model_executor/models/qwen3_vl.py", line 639, in forward
     hidden_states = blk(
                     ^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1775, in _wrapped_call_impl
     return self._call_impl(*args, **kwargs)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1786, in _call_impl
     return forward_call(*args, **kwargs)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/workspace/vllm/vllm/model_executor/models/qwen3_vl.py", line 256, in forward
     x = x + self.attn(
             ^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1775, in _wrapped_call_impl
     return self._call_impl(*args, **kwargs)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1786, in _call_impl
     return forward_call(*args, **kwargs)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/workspace/vllm/vllm/model_executor/models/qwen2_5_vl.py", line 389, in forward
     qk_rotated = self.apply_rotary_emb(
                  ^^^^^^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1775, in _wrapped_call_impl
     return self._call_impl(*args, **kwargs)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1786, in _call_impl
     return forward_call(*args, **kwargs)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/workspace/vllm/vllm/model_executor/custom_op.py", line 136, in forward
     return self._forward_method(*args, **kwargs)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/workspace/vllm/vllm/model_executor/layers/rotary_embedding/common.py", line 266, in forward_hip
     output = self.apply_rotary_emb_flash_attn(
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/flash_attn/ops/triton/rotary.py", line 159, in apply_rotary
     torch.library.wrap_triton(rotary_kernel)[grid](
   File "/usr/local/lib/python3.12/dist-packages/torch/_higher_order_ops/triton_kernel_wrap.py", line 2049, in __call__
     return tracing_triton_hopifier_singleton.call_triton_kernel(
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/_higher_order_ops/triton_kernel_wrap.py", line 1893, in call_triton_kernel
     return self.call_HOP(variable, grids, combined_args_raw, tx)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/_higher_order_ops/triton_kernel_wrap.py", line 2003, in call_HOP
     return triton_kernel_wrapper_mutation(
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/_higher_order_ops/triton_kernel_wrap.py", line 980, in __call__
     return super().__call__(
            ^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/_ops.py", line 536, in __call__
     return wrapper()
            ^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/_ops.py", line 532, in wrapper
     return self.dispatch(
            ^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/_ops.py", line 520, in dispatch
     return kernel(*args, **kwargs)
            ^^^^^^^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/_higher_order_ops/triton_kernel_wrap.py", line 1094, in triton_kernel_wrapper_mutation_dense
     kernel[grid_fn](*args, **kwargs, **constant_args)
   File "/usr/local/lib/python3.12/dist-packages/triton/runtime/jit.py", line 393, in <lambda>
     return lambda *args, **kwargs: self.run(grid=grid, warmup=False, *args, **kwargs)
                                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/triton/runtime/jit.py", line 623, in run
     kernel.run(grid_0, grid_1, grid_2, stream, kernel.function, kernel.packed_metadata, launch_metadata,
   File "/usr/local/lib/python3.12/dist-packages/triton/backends/amd/driver.py", line 647, in __call__
     self.launch(self.launch_cooperative_grid, gridX, gridY, gridZ, stream, function, profile_scratch, *args)
 RuntimeError: Triton Error [HIP]:  Code: 1, Messsage: invalid argument
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 11.4.0-1ubuntu1~22.04.2) 11.4.0
Clang version                : 20.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-7.0.0 25314 f4087f6b428f0e6f575ebac8a8a724dab123d06e)
CMake version                : version 3.31.10
Libc version                 : glibc-2.35

==============================
       PyTorch Info
==============================
PyTorch version              : 2.9.1+git8907517
Is debug build               : False
CUDA used to build PyTorch   : N/A
ROCM used to build PyTorch   : 7.0.51831-a3e329ad8

==============================
      Python Environment
==============================
Python version               : 3.12.12 (main, Oct 10 2025, 08:52:57) [GCC 11.4.0] (64-bit runtime)
Python platform              : Linux-6.8.0-101-generic-x86_64-with-glibc2.35

==============================
       CUDA / GPU Info
==============================
Is CUDA available            : True
CUDA runtime version         : Could not collect
CUDA_MODULE_LOADING set to   : 
GPU models and configuration :  (gfx942:sramecc+:xnack-)
Nvidia driver version        : Could not collect
cuDNN version                : Could not collect
HIP runtime version          : 7.0.51831
MIOpen runtime version       : 3.5.0
Is XNNPACK available         : True

==============================
          CPU Info
==============================
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):                                  256
On-line CPU(s) list:                     0-255
Vendor ID:                               AuthenticAMD
Model name:                              AMD EPYC 9555 64-Core Processor
CPU family:                              26
Model:                                   2
Thread(s) per core:                      2
Core(s) per socket:                      64
Socket(s):                               2
Stepping:                                1
Frequency boost:                         enabled
CPU max MHz:                             4409.3750
CPU min MHz:                             1500.0000
BogoMIPS:                                6399.98
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 amd_lbr_v2 nopl nonstop_tsc cpuid extd_apicid aperfmperf rapl pni pclmulqdq monitor ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand lahf_lm cmp_legacy svm extapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ibs skinit wdt tce topoext perfctr_core perfctr_nb bpext perfctr_llc mwaitx cpb cat_l3 cdp_l3 hw_pstate ssbd mba perfmon_v2 ibrs ibpb stibp ibrs_enhanced vmmcall fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid cqm rdt_a avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local user_shstk avx_vnni avx512_bf16 clzero irperf xsaveerptr rdpru wbnoinvd amd_ppin cppc amd_ibpb_ret arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold avic v_vmsave_vmload vgif x2avic v_spec_ctrl vnmi avx512vbmi umip pku ospke avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg avx512_vpopcntdq la57 rdpid bus_lock_detect movdiri movdir64b overflow_recov succor smca fsrm avx512_vp2intersect flush_l1d debug_swap
Virtualization:                          AMD-V
L1d cache:                               6 MiB (128 instances)
L1i cache:                               4 MiB (128 instances)
L2 cache:                                128 MiB (128 instances)
L3 cache:                                512 MiB (16 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 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 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; STIBP always-on; PBRSB-eIBRS Not affected; BHI Not affected
Vulnerability Srbds:                     Not affected
Vulnerability Tsa:                       Not affected
Vulnerability Tsx async abort:           Not affected
Vulnerability Vmscape:                   Not affected

==============================
Versions of relevant libraries
==============================
[pip3] conch-triton-kernels==1.2.1
[pip3] numpy==2.1.3
[pip3] onnx==1.19.0
[pip3] onnx-ir==0.2.0
[pip3] onnxscript==0.6.2
[pip3] onnxslim==0.1.86
[pip3] pyzmq==27.1.0
[pip3] torch==2.9.1+git8907517
[pip3] torchaudio==2.9.0+eaa9e4e
[pip3] torchvision==0.24.1+d801a34
[pip3] transformers==4.57.6
[pip3] triton==3.4.0
[pip3] triton_kernels==1.0.0
[conda] Could not collect

==============================
         vLLM Info
==============================
ROCM Version                 : 7.0.51831-a3e329ad8
vLLM Version                 : 0.17.1rc1.dev34+g0b4aea4d4.d20260314 (git sha: 0b4aea4d4, date: 20260314)
vLLM Build Flags:
  CUDA Archs: Not Set; ROCm: Disabled
GPU Topology:
  ============================ ROCm System Management Interface ============================
================================ Weight between two GPUs =================================
       GPU0         GPU1         GPU2         GPU3         GPU4         GPU5         GPU6         GPU7         
GPU0   0            15           15           15           15           15           15           15           
GPU1   15           0            15           15           15           15           15           15           
GPU2   15           15           0            15           15           15           15           15           
GPU3   15           15           15           0            15           15           15           15           
GPU4   15           15           15           15           0            15           15           15           
GPU5   15           15           15           15           15           0            15           15           
GPU6   15           15           15           15           15           15           0            15           
GPU7   15           15           15           15           15           15           15           0            

================================= Hops between two GPUs ==================================
       GPU0         GPU1         GPU2         GPU3         GPU4         GPU5         GPU6         GPU7         
GPU0   0            1            1            1            1            1            1            1            
GPU1   1            0            1            1            1            1            1            1            
GPU2   1            1            0            1            1            1            1            1            
GPU3   1            1            1            0            1            1            1            1            
GPU4   1            1            1            1            0            1            1            1            
GPU5   1            1            1            1            1            0            1            1            
GPU6   1            1            1            1            1            1            0            1            
GPU7   1            1            1            1            1            1            1            0            

=============================== Link Type between two GPUs ===============================
       GPU0         GPU1         GPU2         GPU3         GPU4         GPU5         GPU6         GPU7         
GPU0   0            XGMI         XGMI         XGMI         XGMI         XGMI         XGMI         XGMI         
GPU1   XGMI         0            XGMI         XGMI         XGMI         XGMI         XGMI         XGMI         
GPU2   XGMI         XGMI         0            XGMI         XGMI         XGMI         XGMI         XGMI         
GPU3   XGMI         XGMI         XGMI         0            XGMI         XGMI         XGMI         XGMI         
GPU4   XGMI         XGMI         XGMI         XGMI         0            XGMI         XGMI         XGMI         
GPU5   XGMI         XGMI         XGMI         XGMI         XGMI         0            XGMI         XGMI         
GPU6   XGMI         XGMI         XGMI         XGMI         XGMI         XGMI         0            XGMI         
GPU7   XGMI         XGMI         XGMI         XGMI         XGMI         XGMI         XGMI         0            

======================================= Numa Nodes =======================================
GPU[0]		: (Topology) Numa Node: 0
GPU[0]		: (Topology) Numa Affinity: 0
GPU[1]		: (Topology) Numa Node: 0
GPU[1]		: (Topology) Numa Affinity: 0
GPU[2]		: (Topology) Numa Node: 0
GPU[2]		: (Topology) Numa Affinity: 0
GPU[3]		: (Topology) Numa Node: 0
GPU[3]		: (Topology) Numa Affinity: 0
GPU[4]		: (Topology) Numa Node: 1
GPU[4]		: (Topology) Numa Affinity: 1
GPU[5]		: (Topology) Numa Node: 1
GPU[5]		: (Topology) Numa Affinity: 1
GPU[6]		: (Topology) Numa Node: 1
GPU[6]		: (Topology) Numa Affinity: 1
GPU[7]		: (Topology) Numa Node: 1
GPU[7]		: (Topology) Numa Affinity: 1
================================== End of ROCm SMI Log ===================================

==============================
     Environment Variables
==============================
PYTORCH_ROCM_ARCH=gfx90a;gfx942;gfx950;gfx1100;gfx1101;gfx1200;gfx1201;gfx1150;gfx1151
LD_LIBRARY_PATH=/opt/rocm/lib:/usr/local/lib:
PYTORCH_NVML_BASED_CUDA_CHECK=1
TORCHINDUCTOR_COMPILE_THREADS=1
</details>

🐛 Describe the bug

Environment

  • Model: amd/Qwen3.5-397B-A17B-MTP-PTPC-FP8
  • ROCm: MI325X (gfx942)

Command

vllm serve amd/Qwen3.5-397B-A17B-MTP-PTPC-FP8 \
    --tensor-parallel-size 8 \
    --dtype auto \
    --enforce-eager \
    --no-enable-chunked-prefill \
    --speculative-config '{"method": "mtp", "num_speculative_tokens": 1}' \
    --gpu-memory-utilization 0.95 \
    --port 8008

Error message The server crashes during engine initialization at the memory-profiling stage (profile_run), before any user request is served. The profiling run creates dummy multimodal inputs and forwards them through the Qwen3.5 vision encoder (ViT). All 8 TP workers fail simultaneously with the same error:

 Traceback (most recent call last):
   File "/workspace/vllm/vllm/v1/executor/multiproc_executor.py", line 944, in worker_busy_loop
     output = func(*args, **kwargs)
              ^^^^^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/utils/_contextlib.py", line 120, in decorate_context
     return func(*args, **kwargs)
            ^^^^^^^^^^^^^^^^^^^^^
   File "/workspace/vllm/vllm/v1/worker/gpu_worker.py", line 370, in determine_available_memory
     self.model_runner.profile_run()
   File "/workspace/vllm/vllm/v1/worker/gpu_model_runner.py", line 5536, in profile_run
     dummy_encoder_outputs = self.model.embed_multimodal(
                             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/workspace/vllm/vllm/model_executor/models/qwen3_vl.py", line 2398, in embed_multimodal
     image_embeddings = self._process_image_input(multimodal_input)
                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/workspace/vllm/vllm/model_executor/models/qwen3_vl.py", line 1818, in _process_image_input
     image_embeds = self.visual(pixel_values, grid_thw=grid_thw)
                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1775, in _wrapped_call_impl
     return self._call_impl(*args, **kwargs)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1786, in _call_impl
     return forward_call(*args, **kwargs)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/workspace/vllm/vllm/model_executor/models/qwen3_vl.py", line 639, in forward
     hidden_states = blk(
                     ^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1775, in _wrapped_call_impl
     return self._call_impl(*args, **kwargs)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1786, in _call_impl
     return forward_call(*args, **kwargs)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/workspace/vllm/vllm/model_executor/models/qwen3_vl.py", line 256, in forward
     x = x + self.attn(
             ^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1775, in _wrapped_call_impl
     return self._call_impl(*args, **kwargs)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1786, in _call_impl
     return forward_call(*args, **kwargs)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/workspace/vllm/vllm/model_executor/models/qwen2_5_vl.py", line 389, in forward
     qk_rotated = self.apply_rotary_emb(
                  ^^^^^^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1775, in _wrapped_call_impl
     return self._call_impl(*args, **kwargs)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py", line 1786, in _call_impl
     return forward_call(*args, **kwargs)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/workspace/vllm/vllm/model_executor/custom_op.py", line 136, in forward
     return self._forward_method(*args, **kwargs)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/workspace/vllm/vllm/model_executor/layers/rotary_embedding/common.py", line 266, in forward_hip
     output = self.apply_rotary_emb_flash_attn(
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/flash_attn/ops/triton/rotary.py", line 159, in apply_rotary
     torch.library.wrap_triton(rotary_kernel)[grid](
   File "/usr/local/lib/python3.12/dist-packages/torch/_higher_order_ops/triton_kernel_wrap.py", line 2049, in __call__
     return tracing_triton_hopifier_singleton.call_triton_kernel(
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/_higher_order_ops/triton_kernel_wrap.py", line 1893, in call_triton_kernel
     return self.call_HOP(variable, grids, combined_args_raw, tx)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/_higher_order_ops/triton_kernel_wrap.py", line 2003, in call_HOP
     return triton_kernel_wrapper_mutation(
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/_higher_order_ops/triton_kernel_wrap.py", line 980, in __call__
     return super().__call__(
            ^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/_ops.py", line 536, in __call__
     return wrapper()
            ^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/_ops.py", line 532, in wrapper
     return self.dispatch(
            ^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/_ops.py", line 520, in dispatch
     return kernel(*args, **kwargs)
            ^^^^^^^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/_higher_order_ops/triton_kernel_wrap.py", line 1094, in triton_kernel_wrapper_mutation_dense
     kernel[grid_fn](*args, **kwargs, **constant_args)
   File "/usr/local/lib/python3.12/dist-packages/triton/runtime/jit.py", line 393, in <lambda>
     return lambda *args, **kwargs: self.run(grid=grid, warmup=False, *args, **kwargs)
                                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/triton/runtime/jit.py", line 623, in run
     kernel.run(grid_0, grid_1, grid_2, stream, kernel.function, kernel.packed_metadata, launch_metadata,
   File "/usr/local/lib/python3.12/dist-packages/triton/backends/amd/driver.py", line 647, in __call__
     self.launch(self.launch_cooperative_grid, gridX, gridY, gridZ, stream, function, profile_scratch, *args)
 RuntimeError: Triton Error [HIP]:  Code: 1, Messsage: invalid argument

Root Cause The Qwen3.5 vision encoder has hidden_size=1152 and num_heads=16, giving head_dim=72 (not a power of 2) which is is invalid on MI325X.

This affects any model whose vision encoder has a non-power-of-2 head dimension.

Solution Fall back to native rotary embedding

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

Fix Plan

To fix the issue, we need to fall back to native rotary embedding for models with non-power-of-2 head dimensions. Here are the steps:

  • Identify the affected models: Check the hidden_size and num_heads of each model's vision encoder to determine if the head_dim is a power of 2.
  • Modify the model code: For affected models, modify the forward method of the vision encoder to use native rotary embedding instead of Triton's optimized implementation.
  • Update the apply_rotary_emb method: Replace the call to self.apply_rotary_emb_flash_attn with a call to the native rotary embedding implementation.

Example code:

# Replace this line:
output = self.apply_rotary_emb_flash_attn(
# With this line:
output = self.apply_rotary_emb_native(

Native rotary embedding implementation:

def apply_rotary_emb_native(self, x, grid_thw):
    # Implement native rotary embedding logic here
    pass

Verification

To verify that the fix worked, run the following command:

vllm serve amd/Qwen3.5-397B-A17B-MTP-PTPC-FP8 \
    --tensor-parallel-size 8 \
    --dtype auto \
    --enforce-eager \
    --no-enable-chunked-prefill \
    --speculative-config '{"method": "mtp", "num_speculative_tokens": 1}' \
    --gpu-memory-utilization 0.95 \
    --port 8008

Check that the server initializes successfully and serves requests without crashing.

Extra Tips

  • Make sure to test the modified model code thoroughly to ensure that it produces the correct results.
  • Consider adding a check to automatically detect models with non-power-of-2 head dimensions and fall back to native rotary embedding accordingly.
  • If you encounter any issues during the verification process, refer to the documentation page for troubleshooting guides and FAQs.

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…
#api#output truncation#response parsing#generation 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]: RuntimeError triton error during profile_run with Qwen3.5-MoE vision encoder on ROCm [1 pull requests, 2 comments, 2 participants]