镜像与模型均来自developer.metax-tech.com/ai-models/v0.5.12/DeepSeek/%E5%90%AF%E5%8A%A8%E5%87%86%E5%A4%87.md#GPU
执行指令：bash sglang/prefill.sh *** 0.0.0.0 1 0
[19:15:16.901][MCC][E]ErrorNoBinaryForGpu: Couldn't find xcore1000 binary for current device! Please compile using option: --offload-arch=xcore1000
[19:15:16.902][MCR][E]mc_runtime_api.cpp :382 : 3207 : [7ff902c0f740] mcLaunchKernelExC: Returned mcErrorNoKernelImageForDevice
Profiling prefill latency for dynamic chunking: 0%| | 0/128 [00:00<?, ?it/s]
[2026-08-14 19:15:16 PP0] [PP Dynamic Chunk] Failed to profile prefill latency: Runtime check failed at /opt/conda/lib/python3.10/site-packages/sglang/jit_kernel/csrc/deepseek_v4/c_plan.cuh:617: CUDA error: no kernel image is available for execution on the device. Dynamic chunking will be disabled.
