Skipping import of cpp extensions due to incompatible torch version 2.10.0a0+rocm7.11.0a20251210 for torchao version 0.14.1 Please see https://github.com/pytorch/ao/issues/2919 for more info WARNING 12-19 14:41:24 [attention.py:82] Using VLLM_V1_USE_PREFILL_DECODE_ATTENTION environment variable is deprecated and will be removed in v0.14.0 or v1.0.0, whichever is soonest. Please use --attention-config.use_prefill_decode_attention command line argument or AttentionConfig(use_prefill_decode_attention=...) config field instead. (APIServer pid=57644) INFO 12-19 14:41:24 [api_server.py:1351] vLLM API server version 0.13.0rc2.dev112+g763963aa7.d20251213 (APIServer pid=57644) INFO 12-19 14:41:24 [utils.py:253] non-default args: {'model_tag': 'openai/gpt-oss-20b', 'host': '127.0.0.1', 'model': 'openai/gpt-oss-20b', 'trust_remote_code': True, 'max_model_len': 32768, 'gpu_memory_utilization': 0.98, 'max_num_seqs': 64} (APIServer pid=57644) The argument `trust_remote_code` is to be used with Auto classes. It has no effect here and is ignored. (APIServer pid=57644) INFO 12-19 14:41:28 [model.py:514] Resolved architecture: GptOssForCausalLM (APIServer pid=57644) Parse safetensors files: 0%| | 0/3 [00:00, 'debug_dump_path': None, 'cache_dir': '', 'compile_cache_save_format': 'binary', 'backend': 'inductor', 'custom_ops': ['none'], 'splitting_ops': ['vllm::unified_attention', 'vllm::unified_attention_with_output', 'vllm::unified_mla_attention', 'vllm::unified_mla_attention_with_output', 'vllm::mamba_mixer2', 'vllm::mamba_mixer', 'vllm::short_conv', 'vllm::linear_attention', 'vllm::plamo2_mamba_mixer', 'vllm::gdn_attention_core', 'vllm::kda_attention', 'vllm::sparse_attn_indexer'], 'compile_mm_encoder': False, 'compile_sizes': [], 'compile_ranges_split_points': [2048], 'inductor_compile_config': {'enable_auto_functionalized_v2': False, 'combo_kernels': True, 'benchmark_combo_kernel': True}, 'inductor_passes': {}, 'cudagraph_mode': , 'cudagraph_num_of_warmups': 1, 'cudagraph_capture_sizes': [1, 2, 4, 8, 16, 24, 32, 40, 48, 56, 64, 72, 80, 88, 96, 104, 112, 120, 128, 136, 144, 152, 160, 168, 176, 184, 192, 200, 208, 216, 224, 232, 240, 248, 256, 272, 288, 304, 320, 336, 352, 368, 384, 400, 416, 432, 448, 464, 480, 496, 512, 528, 544, 560, 576, 592, 608, 624, 640, 656, 672, 688, 704, 720, 736, 752, 768, 784, 800, 816, 832, 848, 864, 880, 896, 912, 928, 944, 960, 976, 992, 1008, 1024], 'cudagraph_copy_inputs': False, 'cudagraph_specialize_lora': True, 'use_inductor_graph_partition': False, 'pass_config': {'fuse_norm_quant': False, 'fuse_act_quant': False, 'fuse_attn_quant': False, 'eliminate_noops': True, 'enable_sp': False, 'fuse_gemm_comms': False, 'fuse_allreduce_rms': False}, 'max_cudagraph_capture_size': 1024, 'dynamic_shapes_config': {'type': , 'evaluate_guards': False}, 'local_cache_dir': None} (EngineCore_DP0 pid=57811) INFO 12-19 14:41:33 [parallel_state.py:1203] world_size=1 rank=0 local_rank=0 distributed_init_method=tcp://192.168.1.122:59097 backend=nccl (EngineCore_DP0 pid=57811) INFO 12-19 14:41:33 [parallel_state.py:1411] rank 0 in world size 1 is assigned as DP rank 0, PP rank 0, PCP rank 0, TP rank 0, EP rank 0 (EngineCore_DP0 pid=57811) INFO 12-19 14:41:34 [gpu_model_runner.py:3562] Starting to load model openai/gpt-oss-20b... (EngineCore_DP0 pid=57811) INFO 12-19 14:41:34 [rocm.py:306] Using Rocm Attention backend on V1 engine. (EngineCore_DP0 pid=57811) INFO 12-19 14:41:34 [layer.py:372] Enabled separate cuda stream for MoE shared_experts (EngineCore_DP0 pid=57811) INFO 12-19 14:41:34 [mxfp4.py:170] Using Triton backend (EngineCore_DP0 pid=57811) Loading safetensors checkpoint shards: 0% Completed | 0/3 [00:00