/dev/shm/.tmp_yi/workspace/omni-wan/vllm_omni/version.py:55: RuntimeWarning: vLLM and vLLM-Omni appear to have mismatched major/minor versions: --> vLLM-Omni version 0.27.0rc2.dev81+g199ef7779 --> vLLM version 0.26.0 This will likely cause compatibility issues. warn_if_misaligned_vllm_version() INFO 08-28 09:22:31 [patch.py:252] NVFP4 W4A4 weight_scale NaN-clamp: installed. INFO 08-28 09:22:37 [quack_fp8.py:76] Quack FP8 fused-bias GEMM enabled (CuteDSL). INFO 08-28 09:22:37 [quack_fp8.py:155] Patched FlashInfer FP8 ScaledMM to use quack fused-bias GEMM. INFO 08-28 09:22:37 [patch.py:503] inductor factorable-divisibility patch: installed. INFO 08-28 09:22:37 [patch.py:568] [cumem-cuda] CuMemAllocator._python_free_callback patched: asleep guard extended to all platforms. INFO 08-28 09:22:40 [serve.py:194] Detected diffusion model: /dev/shm/.tmp_yi/models/Wan-AI/Wan2.2-T2V-A14B-Diffusers INFO 08-28 09:22:40 [logo.py:52] █ █ █▄ ▄█ ▄▀▀▀▀▄ █▄ ▄█ █▄ █ ▀█▀ INFO 08-28 09:22:40 [logo.py:52] ▄▄ ▄█ █ █ █ ▀▄▀ █ ▄▄▄ █ █ █ ▀▄▀ █ █ ▀▄ █ █ INFO 08-28 09:22:40 [logo.py:52] █▄█▀ █ █ █ █ █ █ █ █ █ ▀▄█ █ INFO 08-28 09:22:40 [logo.py:52] ▀▀ ▀▀▀▀▀ ▀▀▀▀▀ ▀ ▀ ▀▀▀▀ ▀ ▀ ▀ ▀ ▀▀▀ INFO 08-28 09:22:40 [logo.py:52] (APIServer pid=859182) INFO 08-28 09:22:40 [api_utils.py:345] vLLM server version 0.26.0, serving model /dev/shm/.tmp_yi/models/Wan-AI/Wan2.2-T2V-A14B-Diffusers (APIServer pid=859182) INFO 08-28 09:22:40 [api_utils.py:273] non-default args: {'model_tag': '/dev/shm/.tmp_yi/models/Wan-AI/Wan2.2-T2V-A14B-Diffusers', 'host': '127.0.0.1', 'port': 8143, 'model': '/dev/shm/.tmp_yi/models/Wan-AI/Wan2.2-T2V-A14B-Diffusers'} (APIServer pid=859182) INFO 08-28 09:22:40 [omni_base.py:195] [AsyncOmni] Initializing with model /dev/shm/.tmp_yi/models/Wan-AI/Wan2.2-T2V-A14B-Diffusers (APIServer pid=859182) INFO 08-28 09:22:40 [async_omni_engine.py:139] [AsyncOmniEngine] Initializing with model /dev/shm/.tmp_yi/models/Wan-AI/Wan2.2-T2V-A14B-Diffusers (APIServer pid=859182) INFO 08-28 09:22:40 [config_factory.py:230] Detected pipeline 'wan2_2_ti2v' from model_index.json (_class_name='WanPipeline') (APIServer pid=859182) INFO 08-28 09:22:40 [config_factory.py:230] Detected pipeline 'wan2_2_ti2v' from model_index.json (_class_name='WanPipeline') (APIServer pid=859182) INFO 08-28 09:22:40 [async_omni_engine.py:236] [AsyncOmniEngine] Launching Orchestrator thread with 1 stages (APIServer pid=859182) INFO 08-28 09:22:40 [multiproc_executor.py:296] Starting server... (APIServer pid=859182) WARNING 08-28 09:22:40 [multiproc_executor.py:1070] Reducing Torch parallelism from 128 threads to 1 to avoid unnecessary CPU contention. Set OMP_NUM_THREADS in the external environment to tune this value as needed. /dev/shm/.tmp_yi/workspace/omni-wan/vllm_omni/version.py:55: RuntimeWarning: vLLM and vLLM-Omni appear to have mismatched major/minor versions: --> vLLM-Omni version 0.27.0rc2.dev81+g199ef7779 --> vLLM version 0.26.0 This will likely cause compatibility issues. warn_if_misaligned_vllm_version() INFO 08-28 09:22:49 [patch.py:252] NVFP4 W4A4 weight_scale NaN-clamp: installed. INFO 08-28 09:22:49 [quack_fp8.py:76] Quack FP8 fused-bias GEMM enabled (CuteDSL). INFO 08-28 09:22:49 [quack_fp8.py:155] Patched FlashInfer FP8 ScaledMM to use quack fused-bias GEMM. INFO 08-28 09:22:50 [patch.py:503] inductor factorable-divisibility patch: installed. INFO 08-28 09:22:50 [patch.py:568] [cumem-cuda] CuMemAllocator._python_free_callback patched: asleep guard extended to all platforms. INFO 08-28 09:22:52 [diffusion_worker.py:871] Worker 0 created result MessageQueue INFO 08-28 09:22:52 [scheduler.py:252] Chunked prefill is enabled with max_num_batched_tokens=2048. INFO 08-28 09:22:52 [vllm.py:1109] Asynchronous scheduling is enabled. INFO 08-28 09:22:52 [kernel.py:295] Final IR op priority after setting platform defaults: IrOpPriorityConfig(rms_norm=['native'], fused_add_rms_norm=['native']) INFO 08-28 09:22:52 [diffusion_worker.py:315] Final IR op priority after setting vLLM-Omni overrides: IrOpPriorityConfig(rms_norm=['vllm_c', 'native'], fused_add_rms_norm=['vllm_c', 'native']) INFO 08-28 09:22:52 [diffusion_worker.py:327] Worker 0: Initialized device and distributed environment. INFO 08-28 09:22:52 [parallel_state.py:646] Building SP subgroups from explicit sp_group_ranks (sp_size=1, ulysses=1, ring=1, use_ulysses_low=True). INFO 08-28 09:22:52 [parallel_state.py:688] SP group details for rank 0: sp_group=[0], ulysses_group=[0], ring_group=[0] Loading weights: 0%| | 0/242 [00:00