ARG BASE_IMAGE=nvcr.io/nvidia/nemo:26.06 FROM ${BASE_IMAGE} ARG INSTALL_FLASH_ATTN4=0 ARG PIP_INDEX_URL=https://mirrors.aliyun.com/pypi/simple/ ARG PIP_EXTRA_INDEX_URL=https://pypi.org/simple ENV PIP_INDEX_URL=${PIP_INDEX_URL} ENV PIP_EXTRA_INDEX_URL=${PIP_EXTRA_INDEX_URL} ENV PIP_DISABLE_PIP_VERSION_CHECK=1 RUN if [ "${INSTALL_FLASH_ATTN4}" = "1" ]; then \ python3 -m pip uninstall -y flash-attn flash-attn-3 flash-attn-4 || true; \ MAX_JOBS="${MAX_JOBS:-16}" python3 -m pip install --no-cache-dir --no-build-isolation \ "flash-attn-4" "nvidia-cutlass-dsl"; \ fi RUN set -eux; \ python3 - <<'PY' from pathlib import Path import shutil site = Path("/usr/local/lib/python3.12/dist-packages") src = site / "vllm/third_party/triton_kernels" dst = site / "triton_kernels" if not (src / "matmul_ogs.py").exists() or not (src / "tensor.py").exists(): raise SystemExit(f"missing vLLM triton_kernels source: {src}") if not dst.exists(): raise SystemExit(f"missing top-level triton_kernels package: {dst}") backup = site / "triton_kernels_nv26_backup" if backup.exists(): shutil.rmtree(backup) shutil.copytree(dst, backup, ignore=shutil.ignore_patterns("__pycache__")) shutil.rmtree(dst) shutil.copytree(src, dst, ignore=shutil.ignore_patterns("__pycache__")) PY RUN python3 - <<'PY' import importlib.metadata as metadata import importlib.util required = ["torch", "transformer-engine", "megatron-core", "megatron-bridge"] for package in required: print(package, metadata.version(package)) try: print("flash-attn-4", metadata.version("flash-attn-4")) except metadata.PackageNotFoundError: print("flash-attn-4 MISSING") try: print("nvidia-cutlass-dsl", metadata.version("nvidia-cutlass-dsl")) except metadata.PackageNotFoundError: print("nvidia-cutlass-dsl MISSING") spec = importlib.util.find_spec("triton_kernels.matmul_ogs") if spec is None: raise SystemExit("triton_kernels.matmul_ogs MISSING") print("triton_kernels.matmul_ogs", spec.origin) PY