nvs:1000

NIM_NAME=moonshotai/Kimi-K2.6 NIM_MODEL_NAME=moonshotai/Kimi-K2.6 MANIFEST_FILE=model_manifest.yaml python3 -m pip uninstall -y imageio-ffmpeg moviepy av opencv-python-headless torchcodec ||
  true &&
  rm -rf /usr/local/lib/python3.12/dist-packages/imageio_ffmpeg* /usr/local/lib/python3.12/dist-packages/moviepy* /usr/local/lib/python3.12/dist-packages/av* /usr/local/lib/python3.12/dist-packages/opencv_python_headless* /usr/local/lib/python3.12/dist-packages/cv2 /usr/local/lib/python3.12/dist-packages/torchcodec* /usr/local/lib/python3.12/dist-packages/imageio/plugins/ffmpeg.py /usr/local/lib/python3.12/dist-packages/imageio/plugins/__pycache__/ffmpeg* &&
  rm -rf /root/.cache/pip

NIM_NAME=moonshotai/Kimi-K2.6 NIM_MODEL_NAME=moonshotai/Kimi-K2.6 MANIFEST_FILE=model_manifest.yaml if [ -d /usr/local/lib/python3.12/dist-packages/flashinfer_cubin ]; then mkdir -p /usr/local/lib/python3.12/dist-packages/flashinfer_cubin/cubins &&
  chown -R nvs:1000 /usr/local/lib/python3.12/dist-packages/flashinfer_cubin/cubins; fi

NIM_NAME=moonshotai/Kimi-K2.6 NIM_MODEL_NAME=moonshotai/Kimi-K2.6 MANIFEST_FILE=model_manifest.yaml patch -d / -p1 < /workspace/patch/flashinfer_ensure_symlink_filelock.patch &&
  patch -d /usr/local/lib/python3.12/dist-packages/ -p1 < /workspace/patch/nim_inference_sdk_change.patch &&
  patch -d /sgl-workspace/sglang -p1 < /workspace/patch/sglang_21902.patch &&
  patch -d /sgl-workspace/sglang -p1 < /workspace/patch/anthropic-stop-sequence-reporting.patch &&
  patch -d /sgl-workspace/sglang -p1 < /workspace/patch/anthropic-thinking-type-widen.patch &&
  patch -d /sgl-workspace/sglang -p1 < /workspace/patch/anthropic-thinking-default-disabled.patch &&
  patch -d /sgl-workspace/sglang -p1 < /workspace/patch/sglang_response.patch &&
  patch -d /sgl-workspace/sglang -p1 < /workspace/patch/sglang_tokenizer_parallel_sample_shm_fix.patch

patch/sglang_api.py /usr/local/lib/python3.12/dist-packages/nimlib/nim_inference_api_builder/sglang_api.py

patch/inference.py /opt/nim/inference.py

NIM_MODEL_LOADER_EXTRA_CONFIG={"enable_multithread_load": "true","num_threads": 32}

NIM_REASONING_PARSER_NAME=kimi_k2

NIM_TOOL_CALL_PARSER_NAME=kimi_k2

model_manifest.yaml /opt/nim/etc/default/model_manifest.yaml