[Bug]: gpt-oss on Ampere

[ehartford]

Your current environment

The output of python collect_env.py

==============================
        System Info
==============================
OS                           : Ubuntu 24.04.2 LTS (x86_64)
GCC version                  : (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0
Clang version                : Could not collect
CMake version                : version 3.28.3
Libc version                 : glibc-2.39

==============================
       PyTorch Info
==============================
PyTorch version              : 2.9.0.dev20250804+cu128
Is debug build               : False
CUDA used to build PyTorch   : 12.8
ROCM used to build PyTorch   : N/A

==============================
      Python Environment
==============================
Python version               : 3.12.3 (main, Jun 18 2025, 17:59:45) [GCC 13.3.0] (64-bit runtime)
Python platform              : Linux-6.8.0-71-generic-x86_64-with-glibc2.39

==============================
       CUDA / GPU Info
==============================
Is CUDA available            : True
CUDA runtime version         : Could not collect
CUDA_MODULE_LOADING set to   : 
GPU models and configuration : 
GPU 0: NVIDIA GeForce RTX 3090
GPU 1: NVIDIA GeForce RTX 3090
GPU 2: NVIDIA GeForce RTX 3090
GPU 3: NVIDIA GeForce RTX 3090
GPU 4: NVIDIA GeForce RTX 3090
GPU 5: NVIDIA GeForce RTX 3090
GPU 6: NVIDIA GeForce RTX 3090
GPU 7: NVIDIA GeForce RTX 3090

Nvidia driver version        : 580.65.06
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:                        43 bits physical, 48 bits virtual
Byte Order:                           Little Endian
CPU(s):                               64
On-line CPU(s) list:                  0-63
Vendor ID:                            AuthenticAMD
Model name:                           AMD EPYC 7702 64-Core Processor
CPU family:                           23
Model:                                49
Thread(s) per core:                   1
Core(s) per socket:                   64
Socket(s):                            1
Stepping:                             0
Frequency boost:                      enabled
CPU(s) scaling MHz:                   72%
CPU max MHz:                          2183.5930
CPU min MHz:                          1500.0000
BogoMIPS:                             4000.23
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 nopl nonstop_tsc cpuid extd_apicid aperfmperf rapl pni pclmulqdq monitor ssse3 fma cx16 sse4_1 sse4_2 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 ibrs ibpb stibp vmmcall fsgsbase bmi1 avx2 smep bmi2 cqm rdt_a rdseed adx smap clflushopt clwb sha_ni xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local clzero irperf xsaveerptr rdpru wbnoinvd amd_ppin arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold avic v_vmsave_vmload vgif v_spec_ctrl umip rdpid overflow_recov succor smca sev sev_es
Virtualization:                       AMD-V
L1d cache:                            2 MiB (64 instances)
L1i cache:                            2 MiB (64 instances)
L2 cache:                             32 MiB (64 instances)
L3 cache:                             256 MiB (16 instances)
NUMA node(s):                         1
NUMA node0 CPU(s):                    0-63
Vulnerability Gather data sampling:   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:               Mitigation; untrained return thunk; SMT disabled
Vulnerability Spec rstack overflow:   Mitigation; SMT disabled
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; Retpolines; IBPB conditional; STIBP disabled; RSB filling; PBRSB-eIBRS Not affected; BHI Not affected
Vulnerability Srbds:                  Not affected
Vulnerability Tsx async abort:        Not affected

==============================
Versions of relevant libraries
==============================
[pip3] numpy==2.2.6
[pip3] nvidia-cublas-cu12==12.8.4.1
[pip3] nvidia-cuda-cupti-cu12==12.8.90
[pip3] nvidia-cuda-nvrtc-cu12==12.8.93
[pip3] nvidia-cuda-runtime-cu12==12.8.90
[pip3] nvidia-cudnn-cu12==9.10.2.21
[pip3] nvidia-cudnn-frontend==1.13.0
[pip3] nvidia-cufft-cu12==11.3.3.83
[pip3] nvidia-cufile-cu12==1.13.1.3
[pip3] nvidia-curand-cu12==10.3.9.90
[pip3] nvidia-cusolver-cu12==11.7.3.90
[pip3] nvidia-cusparse-cu12==12.5.8.93
[pip3] nvidia-cusparselt-cu12==0.7.1
[pip3] nvidia-ml-py==12.575.51
[pip3] nvidia-nccl-cu12==2.27.5
[pip3] nvidia-nvjitlink-cu12==12.8.93
[pip3] nvidia-nvshmem-cu12==3.3.9
[pip3] nvidia-nvtx-cu12==12.8.90
[pip3] pynvml==12.0.0
[pip3] pytorch-triton==3.4.0+git11ec6354
[pip3] pyzmq==27.0.1
[pip3] torch==2.9.0.dev20250804+cu128
[pip3] torchaudio==2.8.0.dev20250804+cu128
[pip3] torchvision==0.24.0.dev20250804+cu128
[pip3] transformers==4.55.0
[pip3] triton==3.4.0+git663e04e8
[pip3] triton_kernels==1.0.0
[conda] Could not collect

==============================
         vLLM Info
==============================
ROCM Version                 : Could not collect
Neuron SDK Version           : N/A
vLLM Version                 : 0.10.2.dev2+gf5635d62e.d20250805 (git sha: f5635d62e, date: 20250805)
vLLM Build Flags:
  CUDA Archs: Not Set; ROCm: Disabled; Neuron: Disabled
GPU Topology:
        GPU0    GPU1    GPU2    GPU3    GPU4    GPU5    GPU6    GPU7    NIC0    NIC1    CPU Affinity    NUMA Affinity   GPU NUMA ID
GPU0     X      PIX     NODE    NODE    NODE    NODE    NODE    NODE    NODE    NODE    0-63    0               N/A
GPU1    PIX      X      NODE    NODE    NODE    NODE    NODE    NODE    NODE    NODE    0-63    0               N/A
GPU2    NODE    NODE     X      PIX     NODE    NODE    NODE    NODE    NODE    NODE    0-63    0               N/A
GPU3    NODE    NODE    PIX      X      NODE    NODE    NODE    NODE    NODE    NODE    0-63    0               N/A
GPU4    NODE    NODE    NODE    NODE     X      PIX     NODE    NODE    PHB     PHB     0-63    0               N/A
GPU5    NODE    NODE    NODE    NODE    PIX      X      NODE    NODE    PHB     PHB     0-63    0               N/A
GPU6    NODE    NODE    NODE    NODE    NODE    NODE     X      PIX     NODE    NODE    0-63    0               N/A
GPU7    NODE    NODE    NODE    NODE    NODE    NODE    PIX      X      NODE    NODE    0-63    0               N/A
NIC0    NODE    NODE    NODE    NODE    PHB     PHB     NODE    NODE     X      PIX
NIC1    NODE    NODE    NODE    NODE    PHB     PHB     NODE    NODE    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_0
  NIC1: mlx5_1

==============================
     Environment Variables
==============================
NCCL_CUMEM_ENABLE=0
PYTORCH_NVML_BASED_CUDA_CHECK=1
TORCHINDUCTOR_COMPILE_THREADS=1

🐛 Describe the bug

uv venv --python 3.12 --seed
source venv/bin/activate
uv pip install \
  --pre vllm==0.10.1+gptoss \    
  --extra-index-url https://wheels.vllm.ai/gpt-oss/ \     
  --extra-index-url https://download.pytorch.org/whl/nightly/cu128 \     
  --index-strategy unsafe-best-match
vllm serve openai/gpt-oss-120b -tp 8

AssertionError: Sinks are only supported in FlashAttention 3

I also tried:

VLLM_FLASH_ATTN_VERSION=2 vllm serve openai/gpt-oss-120b -tp 8 --hf-overrides '{"sink_token_len": 0, "use_sliding_window": false}'

AssertionError: Sinks are only supported in FlashAttention 3

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.

[XinBow99]

same issue on L40s (Ada Lovelace)

[LucasWilkinson]

I was able to get it to run on 2xA100 by building from source (the 0.10.1+gptoss wheels are currently only built for sm90 and sm100) using instructions in: #22259

then doing:

uv pip uninstall triton
uv pip install triton==3.4.0

and adding VLLM_ATTENTION_BACKEND=TRITON_ATTN_VLLM_V1

i.e.:

VLLM_ATTENTION_BACKEND=TRITON_ATTN_VLLM_V1 vllm serve openai/gpt-oss-120b --tensor-parallel-size 2 --async-scheduling

Note: after doing

uv pip install -U -e . --verbose --no-build-isolation

You may have to reinstall the torch nightly if you get a missing symbol error (followed by uv pip install triton==3.4.0)

[Koki-Itai]

@LucasWilkinson
Thank you. I successfully launched vllm serve on an A100 environment, but encountered an error when sending a completion request to the served endpoint.

[Nietism]

@LucasWilkinson Thank you. I successfully launched vllm serve on an A100 environment, but encountered an error when sending a completion request to the served endpoint.

I'm experiencing the same issue. vLLM could start service successfully after the modification, but there would be an error when sending a completion request.

�[1;36m(EngineCore_0 pid=1650169)�[0;0m ERROR 08-06 16:12:44 [core.py:720] torch.AcceleratorError: CUDA error: no kernel image is available for execution on the device
�[1;36m(EngineCore_0 pid=1650169)�[0;0m ERROR 08-06 16:12:44 [core.py:720] Search for `cudaErrorNoKernelImageForDevice' in https://docs.nvidia.com/cuda/cuda-runtime-api/group__CUDART__TYPES.html for more information.
�[1;36m(EngineCore_0 pid=1650169)�[0;0m ERROR 08-06 16:12:44 [core.py:720] CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
�[1;36m(EngineCore_0 pid=1650169)�[0;0m ERROR 08-06 16:12:44 [core.py:720] For debugging consider passing CUDA_LAUNCH_BLOCKING=1
�[1;36m(EngineCore_0 pid=1650169)�[0;0m ERROR 08-06 16:12:44 [core.py:720] Compile with `TORCH_USE_CUDA_DSA` to enable device-side assertions.

[Ammo-5201314]

Thank you, it runs successfully on vllm, but there is an error in the inference.
�[1;36m(APIServer pid=1789342)�[0;0m /data/WorkSpace/model/GPT-OSS-120B/.venv/lib/python3.12/site-packages/pydantic/_internal/_generate_schema.py:2225: UnsupportedFieldAttributeWarning: The 'deprecated' attribute with value 'max_tokens is deprecated in favor of the max_completion_tokens field' was provided to the Field() function, which has no effect in the context it was used. 'deprecated' is field-specific metadata, and can only be attached to a model field using Annotated metadata or by assignment. This may have happened because an Annotated type alias using the type statement was used, or if the Field() function was attached to a single member of a union type.
�[1;36m(APIServer pid=1789342)�[0;0m warnings.warn(
�[1;36m(VllmWorker pid=1790746)�[0;0m ERROR 08-06 16:23:53 [multiproc_executor.py:596] WorkerProc hit an exception.
�[1;36m(VllmWorker pid=1790746)�[0;0m ERROR 08-06 16:23:53 [multiproc_executor.py:596] Traceback (most recent call last):
�[1;36m(VllmWorker pid=1790746)�[0;0m ERROR 08-06 16:23:53 [multiproc_executor.py:596] File "/data/WorkSpace/model/GPT-OSS-120B/.venv/lib/python3.12/site-packages/vllm/v1/executor/multiproc_executor.py", line 591, in worker_busy_loop
�[1;36m(VllmWorker pid=1790746)�[0;0m ERROR 08-06 16:23:53 [multiproc_executor.py:596] output = func(*args, **kwargs)
�[1;36m(VllmWorker pid=1790746)�[0;0m ERROR 08-06 16:23:53 [multiproc_executor.py:596] ^^^^^^^^^^^^^^^^^^^^^
�[1;36m(VllmWorker pid=1790746)�[0;0m ERROR 08-06 16:23:53 [multiproc_executor.py:596] File "/data/WorkSpace/model/GPT-OSS-120B/.venv/lib/python3.12/site-packages/torch/utils/_contextlib.py", line 120, in decorate_context
�[1;36m(VllmWorker pid=1790746)�[0;0m ERROR 08-06 16:23:53 [multiproc_executor.py:596] return func(*args, **kwargs)
�[1;36m(VllmWorker pid=1790746)�[0;0m ERROR 08-06 16:23:53 [multiproc_executor.py:596] ^^^^^^^^^^^^^^^^^^^^^
�[1;36m(VllmWorker pid=1790746)�[0;0m ERROR 08-06 16:23:53 [multiproc_executor.py:596] File "/data/WorkSpace/model/GPT-OSS-120B/.venv/lib/python3.12/site-packages/vllm/v1/worker/gpu_worker.py", line 362, in execute_model
�[1;36m(VllmWorker pid=1790746)�[0;0m ERROR 08-06 16:23:53 [multiproc_executor.py:596] output = self.model_runner.execute_model(scheduler_output,
�[1;36m(VllmWorker pid=1790746)�[0;0m ERROR 08-06 16:23:53 [multiproc_executor.py:596] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
�[1;36m(VllmWorker pid=1790746)�[0;0m ERROR 08-06 16:23:53 [multiproc_executor.py:596] File "/data/WorkSpace/model/GPT-OSS-120B/.venv/lib/python3.12/site-packages/torch/utils/_contextlib.py", line 120, in decorate_context
�[1;36m(VllmWorker pid=1790746)�[0;0m ERROR 08-06 16:23:53 [multiproc_executor.py:596] return func(*args, **kwargs)
�[1;36m(VllmWorker pid=1790746)�[0;0m ERROR 08-06 16:23:53 [multiproc_executor.py:596] ^^^^^^^^^^^^^^^^^^^^^
�[1;36m(VllmWorker pid=1790746)�[0;0m ERROR 08-06 16:23:53 [multiproc_executor.py:596] File "/data/WorkSpace/model/GPT-OSS-120B/.venv/lib/python3.12/site-packages/vllm/v1/worker/gpu_model_runner.py", line 1606, in execute_model
�[1;36m(VllmWorker pid=1790746)�[0;0m ERROR 08-06 16:23:53 [multiproc_executor.py:596] sample_hidden_states = hidden_states[logits_indices]
�[1;36m(VllmWorker pid=1790746)�[0;0m ERROR 08-06 16:23:53 [multiproc_executor.py:596] ~~~~~~~~~~~~~^^^^^^^^^^^^^^^^
�[1;36m(VllmWorker pid=1790746)�[0;0m ERROR 08-06 16:23:53 [multiproc_executor.py:596] torch.AcceleratorError: CUDA error: no kernel image is available for execution on the device
�[1;36m(VllmWorker pid=1790746)�[0;0m ERROR 08-06 16:23:53 [multiproc_executor.py:596] Search for cudaErrorNoKernelImageForDevice' in https://docs.nvidia.com/cuda/cuda-runtime-api/group__CUDART__TYPES.html for more information. �[1;36m(VllmWorker pid=1790746)�[0;0m ERROR 08-06 16:23:53 [multiproc_executor.py:596] CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect. �[1;36m(VllmWorker pid=1790746)�[0;0m ERROR 08-06 16:23:53 [multiproc_executor.py:596] For debugging consider passing CUDA_LAUNCH_BLOCKING=1 �[1;36m(VllmWorker pid=1790746)�[0;0m ERROR 08-06 16:23:53 [multiproc_executor.py:596] Compile with TORCH_USE_CUDA_DSAto enable device-side assertions. �[1;36m(VllmWorker pid=1790746)�[0;0m ERROR 08-06 16:23:53 [multiproc_executor.py:596] �[1;36m(VllmWorker pid=1790746)�[0;0m ERROR 08-06 16:23:53 [multiproc_executor.py:596] Traceback (most recent call last): �[1;36m(VllmWorker pid=1790746)�[0;0m ERROR 08-06 16:23:53 [multiproc_executor.py:596] File "/data/WorkSpace/model/GPT-OSS-120B/.venv/lib/python3.12/site-packages/vllm/v1/executor/multiproc_executor.py", line 591, in worker_busy_loop �[1;36m(VllmWorker pid=1790746)�[0;0m ERROR 08-06 16:23:53 [multiproc_executor.py:596] output = func(*args, **kwargs) �[1;36m(VllmWorker pid=1790746)�[0;0m ERROR 08-06 16:23:53 [multiproc_executor.py:596] ^^^^^^^^^^^^^^^^^^^^^ �[1;36m(VllmWorker pid=1790746)�[0;0m ERROR 08-06 16:23:53 [multiproc_executor.py:596] File "/data/WorkSpace/model/GPT-OSS-120B/.venv/lib/python3.12/site-packages/torch/utils/_contextlib.py", line 120, in decorate_context �[1;36m(VllmWorker pid=1790746)�[0;0m ERROR 08-06 16:23:53 [multiproc_executor.py:596] return func(*args, **kwargs) �[1;36m(VllmWorker pid=1790746)�[0;0m ERROR 08-06 16:23:53 [multiproc_executor.py:596] ^^^^^^^^^^^^^^^^^^^^^ �[1;36m(VllmWorker pid=1790746)�[0;0m ERROR 08-06 16:23:53 [multiproc_executor.py:596] File "/data/WorkSpace/model/GPT-OSS-120B/.venv/lib/python3.12/site-packages/vllm/v1/worker/gpu_worker.py", line 362, in execute_model �[1;36m(VllmWorker pid=1790746)�[0;0m ERROR 08-06 16:23:53 [multiproc_executor.py:596] output = self.model_runner.execute_model(scheduler_output, �[1;36m(VllmWorker pid=1790746)�[0;0m ERROR 08-06 16:23:53 [multiproc_executor.py:596] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ �[1;36m(VllmWorker pid=1790746)�[0;0m ERROR 08-06 16:23:53 [multiproc_executor.py:596] File "/data/WorkSpace/model/GPT-OSS-120B/.venv/lib/python3.12/site-packages/torch/utils/_contextlib.py", line 120, in decorate_context �[1;36m(VllmWorker pid=1790746)�[0;0m ERROR 08-06 16:23:53 [multiproc_executor.py:596] return func(*args, **kwargs) �[1;36m(VllmWorker pid=1790746)�[0;0m ERROR 08-06 16:23:53 [multiproc_executor.py:596] ^^^^^^^^^^^^^^^^^^^^^ �[1;36m(VllmWorker pid=1790746)�[0;0m ERROR 08-06 16:23:53 [multiproc_executor.py:596] File "/data/WorkSpace/model/GPT-OSS-120B/.venv/lib/python3.12/site-packages/vllm/v1/worker/gpu_model_runner.py", line 1606, in execute_model �[1;36m(VllmWorker pid=1790746)�[0;0m ERROR 08-06 16:23:53 [multiproc_executor.py:596] sample_hidden_states = hidden_states[logits_indices] �[1;36m(VllmWorker pid=1790746)�[0;0m ERROR 08-06 16:23:53 [multiproc_executor.py:596] ~~~~~~~~~~~~~^^^^^^^^^^^^^^^^ �[1;36m(VllmWorker pid=1790746)�[0;0m ERROR 08-06 16:23:53 [multiproc_executor.py:596] torch.AcceleratorError: CUDA error: no kernel image is available for execution on the device �[1;36m(VllmWorker pid=1790746)�[0;0m ERROR 08-06 16:23:53 [multiproc_executor.py:596] Search forcudaErrorNoKernelImageForDevice' in https://docs.nvidia.com/cuda/cuda-runtime-api/group__CUDART__TYPES.html for more information.
�[1;36m(VllmWorker pid=1790746)�[0;0m ERROR 08-06 16:23:53 [multiproc_executor.py:596] CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
�[1;36m(VllmWorker pid=1790746)�[0;0m ERROR 08-06 16:23:53 [multiproc_executor.py:596] For debugging consider passing CUDA_LAUNCH_BLOCKING=1
�[1;36m(VllmWorker pid=1790746)�[0;0m ERROR 08-06 16:23:53 [multiproc_executor.py:596] Compile with TORCH_USE_CUDA_DSA to enable device-side assertions.
�[1;36m(VllmWorker pid=1790746)�[0;0m ERROR 08-06 16:23:53 [multiproc_executor.py:596]
�[1;36m(VllmWorker pid=1790746)�[0;0m ERROR 08-06 16:23:53 [multiproc_executor.py:596]
�[1;36m(EngineCore_0 pid=1790344)�[0;0m ERROR 08-06 16:23:53 [dump_input.py:69] Dumping input data for V1 LLM engine (v0.10.2.dev2+gf5635d62e.d20250806) with config: model='/data/hf_cache/openai-mirror/gpt-oss-120b', speculative_config=None, tokenizer='/data/hf_cache/openai-mirror/gpt-oss-120b', skip_tokenizer_init=False, tokenizer_mode=auto, revision=None, override_neuron_config={}, tokenizer_revision=None, trust_remote_code=False, dtype=torch.bfloat16, max_seq_len=131072, download_dir=None, load_format=auto, tensor_parallel_size=1, pipeline_parallel_size=1, disable_custom_all_reduce=False, quantization=mxfp4, enforce_eager=False, kv_cache_dtype=auto, device_config=cuda, decoding_config=DecodingConfig(backend='auto', disable_fallback=False, disable_any_whitespace=False, disable_additional_properties=False, reasoning_backend='openai'), observability_config=ObservabilityConfig(show_hidden_metrics_for_version=None, otlp_traces_endpoint=None, collect_detailed_traces=None), seed=0, served_model_name=GPT-OSS-120B, num_scheduler_steps=1, multi_step_stream_outputs=True, enable_prefix_caching=True, chunked_prefill_enabled=True, use_async_output_proc=True, pooler_config=None, compilation_config={"level":3,"debug_dump_path":"","cache_dir":"","backend":"","custom_ops":[],"splitting_ops":["vllm.unified_attention","vllm.unified_attention_with_output","vllm.mamba_mixer2"],"use_inductor":true,"compile_sizes":[],"inductor_compile_config":{"enable_auto_functionalized_v2":false},"inductor_passes":{},"use_cudagraph":true,"cudagraph_num_of_warmups":1,"cudagraph_capture_sizes":[1024,1008,992,976,960,944,928,912,896,880,864,848,832,816,800,784,768,752,736,720,704,688,672,656,640,624,608,592,576,560,544,528,512,496,480,464,448,432,416,400,384,368,352,336,320,304,288,272,256,248,240,232,224,216,208,200,192,184,176,168,160,152,144,136,128,120,112,104,96,88,80,72,64,56,48,40,32,24,16,8,4,2,1],"cudagraph_copy_inputs":false,"full_cuda_graph":false,"max_capture_size":1024,"local_cache_dir":null},
�[1;36m(EngineCore_0 pid=1790344)�[0;0m ERROR 08-06 16:23:53 [dump_input.py:76] Dumping scheduler output for model execution: SchedulerOutput(scheduled_new_reqs=[NewRequestData(req_id=chatcmpl-1dc6e4beb4ac40f6a908fc8a363f26aa,prompt_token_ids_len=90,mm_inputs=[],mm_hashes=[],mm_positions=[],sampling_params=SamplingParams(n=1, presence_penalty=0.0, frequency_penalty=0.0, repetition_penalty=1.0, temperature=1.0, top_p=1.0, top_k=0, min_p=0.0, seed=None, stop=[], stop_token_ids=[199999], bad_words=[], include_stop_str_in_output=False, ignore_eos=False, max_tokens=130982, min_tokens=0, logprobs=None, prompt_logprobs=None, skip_special_tokens=True, spaces_between_special_tokens=True, truncate_prompt_tokens=None, guided_decoding=None, extra_args=None),block_ids=([1, 2, 3, 4, 5, 6], [7, 8, 9, 10, 11, 12]),num_computed_tokens=0,lora_request=None)], scheduled_cached_reqs=CachedRequestData(req_ids=[], resumed_from_preemption=[], new_token_ids=[], new_block_ids=[], num_computed_tokens=[]), num_scheduled_tokens={chatcmpl-1dc6e4beb4ac40f6a908fc8a363f26aa: 90}, total_num_scheduled_tokens=90, scheduled_spec_decode_tokens={}, scheduled_encoder_inputs={}, num_common_prefix_blocks=[6, 0], finished_req_ids=[], free_encoder_input_ids=[], structured_output_request_ids={}, grammar_bitmask=null, kv_connector_metadata=null)
�[1;36m(EngineCore_0 pid=1790344)�[0;0m ERROR 08-06 16:23:53 [dump_input.py:79] Dumping scheduler stats: SchedulerStats(num_running_reqs=1, num_waiting_reqs=0, step_counter=0, current_wave=0, kv_cache_usage=0.0010161012974831962, prefix_cache_stats=PrefixCacheStats(reset=False, requests=1, queries=90, hits=0), spec_decoding_stats=None, num_corrupted_reqs=0)
�[1;36m(EngineCore_0 pid=1790344)�[0;0m ERROR 08-06 16:23:53 [core.py:720] EngineCore encountered a fatal error.
�[1;36m(EngineCore_0 pid=1790344)�[0;0m ERROR 08-06 16:23:53 [core.py:720] Traceback (most recent call last):
�[1;36m(EngineCore_0 pid=1790344)�[0;0m ERROR 08-06 16:23:53 [core.py:720] File "/data/WorkSpace/model/GPT-OSS-120B/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 711, in run_engine_core
�[1;36m(EngineCore_0 pid=1790344)�[0;0m ERROR 08-06 16:23:53 [core.py:720] engine_core.run_busy_loop()
�[1;36m(EngineCore_0 pid=1790344)�[0;0m ERROR 08-06 16:23:53 [core.py:720] File "/data/WorkSpace/model/GPT-OSS-120B/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 738, in run_busy_loop
�[1;36m(EngineCore_0 pid=1790344)�[0;0m ERROR 08-06 16:23:53 [core.py:720] self._process_engine_step()
�[1;36m(EngineCore_0 pid=1790344)�[0;0m ERROR 08-06 16:23:53 [core.py:720] File "/data/WorkSpace/model/GPT-OSS-120B/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 763, in _process_engine_step
�[1;36m(EngineCore_0 pid=1790344)�[0;0m ERROR 08-06 16:23:53 [core.py:720] outputs, model_executed = self.step_fn()
�[1;36m(EngineCore_0 pid=1790344)�[0;0m ERROR 08-06 16:23:53 [core.py:720] ^^^^^^^^^^^^^^
�[1;36m(EngineCore_0 pid=1790344)�[0;0m ERROR 08-06 16:23:53 [core.py:720] File "/data/WorkSpace/model/GPT-OSS-120B/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 353, in step_with_batch_queue
�[1;36m(EngineCore_0 pid=1790344)�[0;0m ERROR 08-06 16:23:53 [core.py:720] model_output = self.execute_model_with_error_logging(
�[1;36m(EngineCore_0 pid=1790344)�[0;0m ERROR 08-06 16:23:53 [core.py:720] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
�[1;36m(EngineCore_0 pid=1790344)�[0;0m ERROR 08-06 16:23:53 [core.py:720] File "/data/WorkSpace/model/GPT-OSS-120B/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 276, in execute_model_with_error_logging
�[1;36m(EngineCore_0 pid=1790344)�[0;0m ERROR 08-06 16:23:53 [core.py:720] raise err
�[1;36m(EngineCore_0 pid=1790344)�[0;0m ERROR 08-06 16:23:53 [core.py:720] File "/data/WorkSpace/model/GPT-OSS-120B/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 267, in execute_model_with_error_logging
�[1;36m(EngineCore_0 pid=1790344)�[0;0m ERROR 08-06 16:23:53 [core.py:720] return model_fn(scheduler_output)
�[1;36m(EngineCore_0 pid=1790344)�[0;0m ERROR 08-06 16:23:53 [core.py:720] ^^^^^^^^^^^^^^^^^^^^^^^^^^
�[1;36m(EngineCore_0 pid=1790344)�[0;0m ERROR 08-06 16:23:53 [core.py:720] File "/data/WorkSpace/model/GPT-OSS-120B/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 354, in
�[1;36m(EngineCore_0 pid=1790344)�[0;0m ERROR 08-06 16:23:53 [core.py:720] lambda _: future.result(), scheduler_output)
�[1;36m(EngineCore_0 pid=1790344)�[0;0m ERROR 08-06 16:23:53 [core.py:720] ^^^^^^^^^^^^^^^
�[1;36m(EngineCore_0 pid=1790344)�[0;0m ERROR 08-06 16:23:53 [core.py:720] File "/home/ubuntu/.local/share/uv/python/cpython-3.12.11-linux-x86_64-gnu/lib/python3.12/concurrent/futures/_base.py", line 456, in result
�[1;36m(EngineCore_0 pid=1790344)�[0;0m ERROR 08-06 16:23:53 [core.py:720] return self.__get_result()
�[1;36m(EngineCore_0 pid=1790344)�[0;0m ERROR 08-06 16:23:53 [core.py:720] ^^^^^^^^^^^^^^^^^^^
�[1;36m(EngineCore_0 pid=1790344)�[0;0m ERROR 08-06 16:23:53 [core.py:720] File "/home/ubuntu/.local/share/uv/python/cpython-3.12.11-linux-x86_64-gnu/lib/python3.12/concurrent/futures/_base.py", line 401, in __get_result
�[1;36m(EngineCore_0 pid=1790344)�[0;0m ERROR 08-06 16:23:53 [core.py:720] raise self._exception
�[1;36m(EngineCore_0 pid=1790344)�[0;0m ERROR 08-06 16:23:53 [core.py:720] File "/home/ubuntu/.local/share/uv/python/cpython-3.12.11-linux-x86_64-gnu/lib/python3.12/concurrent/futures/thread.py", line 59, in run
�[1;36m(EngineCore_0 pid=1790344)�[0;0m ERROR 08-06 16:23:53 [core.py:720] result = self.fn(*self.args, **self.kwargs)
�[1;36m(EngineCore_0 pid=1790344)�[0;0m ERROR 08-06 16:23:53 [core.py:720] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
�[1;36m(EngineCore_0 pid=1790344)�[0;0m ERROR 08-06 16:23:53 [core.py:720] File "/data/WorkSpace/model/GPT-OSS-120B/.venv/lib/python3.12/site-packages/vllm/v1/executor/multiproc_executor.py", line 230, in get_response
�[1;36m(EngineCore_0 pid=1790344)�[0;0m ERROR 08-06 16:23:53 [core.py:720] raise RuntimeError(
�[1;36m(EngineCore_0 pid=1790344)�[0;0m ERROR 08-06 16:23:53 [core.py:720] RuntimeError: Worker failed with error 'CUDA error: no kernel image is available for execution on the device
�[1;36m(EngineCore_0 pid=1790344)�[0;0m ERROR 08-06 16:23:53 [core.py:720] Search for cudaErrorNoKernelImageForDevice' in https://docs.nvidia.com/cuda/cuda-runtime-api/group__CUDART__TYPES.html for more information. �[1;36m(EngineCore_0 pid=1790344)�[0;0m ERROR 08-06 16:23:53 [core.py:720] CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect. �[1;36m(EngineCore_0 pid=1790344)�[0;0m ERROR 08-06 16:23:53 [core.py:720] For debugging consider passing CUDA_LAUNCH_BLOCKING=1 �[1;36m(EngineCore_0 pid=1790344)�[0;0m ERROR 08-06 16:23:53 [core.py:720] Compile with TORCH_USE_CUDA_DSA` to enable device-side assertions.
�[1;36m(EngineCore_0 pid=1790344)�[0;0m ERROR 08-06 16:23:53 [core.py:720] ', please check the stack trace above for the root cause
�[1;36m(VllmWorker pid=1790746)�[0;0m INFO 08-06 16:23:53 [multiproc_executor.py:520] Parent process exited, terminating worker
�[1;36m(APIServer pid=1789342)�[0;0m ERROR 08-06 16:23:53 [async_llm.py:430] AsyncLLM output_handler failed.
�[1;36m(APIServer pid=1789342)�[0;0m ERROR 08-06 16:23:53 [async_llm.py:430] Traceback (most recent call last):
�[1;36m(APIServer pid=1789342)�[0;0m ERROR 08-06 16:23:53 [async_llm.py:430] File "/data/WorkSpace/model/GPT-OSS-120B/.venv/lib/python3.12/site-packages/vllm/v1/engine/async_llm.py", line 389, in output_handler
�[1;36m(APIServer pid=1789342)�[0;0m ERROR 08-06 16:23:53 [async_llm.py:430] outputs = await engine_core.get_output_async()
�[1;36m(APIServer pid=1789342)�[0;0m ERROR 08-06 16:23:53 [async_llm.py:430] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
�[1;36m(APIServer pid=1789342)�[0;0m ERROR 08-06 16:23:53 [async_llm.py:430] File "/data/WorkSpace/model/GPT-OSS-120B/.venv/lib/python3.12/site-packages/vllm/v1/engine/core_client.py", line 809, in get_output_async
�[1;36m(APIServer pid=1789342)�[0;0m ERROR 08-06 16:23:53 [async_llm.py:430] raise self._format_exception(outputs) from None

[ayrnb]

@LucasWilkinson Thank you. I successfully launched vllm serve on an A100 environment, but encountered an error when sending a completion request to the served endpoint.

I'm experiencing the same issue. vLLM could start service successfully after the modification, but there would be an error when sending a completion request.

�[1;36m(EngineCore_0 pid=1650169)�[0;0m ERROR 08-06 16:12:44 [core.py:720] torch.AcceleratorError: CUDA error: no kernel image is available for execution on the device
�[1;36m(EngineCore_0 pid=1650169)�[0;0m ERROR 08-06 16:12:44 [core.py:720] Search for `cudaErrorNoKernelImageForDevice' in https://docs.nvidia.com/cuda/cuda-runtime-api/group__CUDART__TYPES.html for more information.
�[1;36m(EngineCore_0 pid=1650169)�[0;0m ERROR 08-06 16:12:44 [core.py:720] CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
�[1;36m(EngineCore_0 pid=1650169)�[0;0m ERROR 08-06 16:12:44 [core.py:720] For debugging consider passing CUDA_LAUNCH_BLOCKING=1
�[1;36m(EngineCore_0 pid=1650169)�[0;0m ERROR 08-06 16:12:44 [core.py:720] Compile with `TORCH_USE_CUDA_DSA` to enable device-side assertions.

+1

[rdlh]

@LucasWilkinson 's fix working too for L40S clusters 👌

[ayrnb]

@LucasWilkinson 's fix working too for L40S clusters 👌

@rdlh Is there any error in inference? when i set VLLM_ATTENTION_BACKEND=TRITON_ATTN_VLLM_V1, the server can work, but an error CUDA error: no kernel image is available for execution on the device occurs in inference.

[CNYoki]

I was able to get it to run on 2xA100 by building from source (the 0.10.1+gptoss wheels are currently only built for sm90 and sm100) using instructions in: #22259

then doing:

uv pip uninstall triton
uv pip install triton==3.4.0

and adding VLLM_ATTENTION_BACKEND=TRITON_ATTN_VLLM_V1

i.e.:

VLLM_ATTENTION_BACKEND=TRITON_ATTN_VLLM_V1 vllm serve openai/gpt-oss-120b --tensor-parallel-size 2 --async-scheduling

Note: after doing

uv pip install -U -e . --verbose --no-build-isolation

You may have to reinstall the torch nightly if you get a missing symbol error (followed by uv pip install triton==3.4.0)

Does not work on A800. Model can run but will be error when generate.

[moaddib666]

Same for my side, looks like Ampere not supported, I'd re-check the https://docs.vllm.ai/projects/recipes/en/latest/OpenAI/GPT-OSS.html and not Ampere cards was mentioned....

(EngineCore_0 pid=70) ERROR 08-06 03:44:21 [core.py:720] torch.AcceleratorError: CUDA error: no kernel image is available for execution on the device
(EngineCore_0 pid=70) ERROR 08-06 03:44:21 [core.py:720] Search for `cudaErrorNoKernelImageForDevice' in https://docs.nvidia.com/cuda/cuda-runtime-api/group__CUDART__TYPES.html for more information.
(EngineCore_0 pid=70) ERROR 08-06 03:44:21 [core.py:720] CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.