Mirrors/modelscope
1
0
Fork 0
mirror of https://github.com/modelscope/modelscope.git synced 2026-08-29 10:08:40 +02:00
Code Issues Packages Projects Releases Wiki Activity

update docker image

Browse Source
This commit is contained in:
hjh0119
2026-08-20 00:42:40 +08:00
parent 06e123e06e
commit 88e36d7ec2
2 changed files with 27 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
docker/Dockerfile.ubuntu
View File

@@ -78,14 +78,16 @@ RUN export PIP_EXTRA_INDEX_URL="${PIP_EXTRA_INDEX_URL}" && \
if [ "$IMAGE_TYPE" = "gpu" ]; then \
pip install --no-cache-dir math_verify "gradio<5.33" "deepspeed<0.19" ray -U && \
pip install --no-cache-dir mcore-bridge -i https://pypi.org/simple/ -U && \
pip install --no-cache-dir liger_kernel wandb swanlab nvitop pre-commit "transformers<5.9" "trl<1.0" "peft<0.20" huggingface-hub -U && \
pip install --no-cache-dir --no-build-isolation "transformer_engine[pytorch]==2.16.0"; \
pip install --no-cache-dir liger_kernel wandb swanlab nvitop pre-commit transformers "trl<1.0" peft huggingface-hub -U && \
NVTE_WITH_NCCL_EP=0 NVTE_FRAMEWORK=pytorch MAX_JOBS=16 \
pip install --no-cache-dir --no-build-isolation "transformer_engine[pytorch]==2.18.0"; \
cd /tmp && GIT_LFS_SKIP_SMUDGE=1 git clone https://github.com/NVIDIA/apex && \
cd apex && pip install -v --disable-pip-version-check --no-cache-dir --no-build-isolation --config-settings "--build-option=--cpp_ext" --config-settings "--build-option=--cuda_ext" ./ && \
cd apex && pip install -v --disable-pip-version-check --no-cache-dir --no-build-isolation --config-settings "--build-option=--cpp_ext" --config-settings "--build-option=--cuda_ext" ./ || \
echo "WARNING: apex build failed; megatron-core falls back to Torch Norm."; \
cd / && rm -fr /tmp/apex && pip cache purge; \
pip install --no-cache-dir "megatron-core==0.16.*" -U; \
pip install --no-cache-dir "megatron-core==0.18.*" -U; \
elif [ "$IMAGE_TYPE" = "cpu" ]; then \
pip install --no-cache-dir huggingface-hub "transformers<5.9" peft diffusers -U; \
pip install --no-cache-dir huggingface-hub transformers peft diffusers -U; \
else \
pip install "transformers<5.0" "tokenizers<0.22" "trl<0.23" "diffusers<0.35" --no-dependencies; \
fi

27
docker/build_image.py
View File

@@ -357,12 +357,18 @@ class StableGPUImageBuilder(Builder):
"""Dependencies will be stable versions"""
def init_args(self, args: Any) -> Any:
if not args.base_image:
# torch>=2.13 is built against CUDA 13 and pulls its own nvidia-cudnn-cu13,
# so the plain -devel tag is enough (matches vLLM's own Dockerfile).
args.base_image = 'nvidia/cuda:13.0.3-devel-ubuntu22.04'
if not args.cuda_version:
args.cuda_version = '13.0.3'
if not args.torch_version:
args.torch_version = '2.10.0'
args.torchaudio_version = '2.10.0'
args.torchvision_version = '0.25.0'
args.torch_version = '2.13.0'
args.torchaudio_version = '2.11.0'
args.torchvision_version = '0.28.0'
if not args.vllm_version:
args.vllm_version = '0.19.1'
args.vllm_version = '0.27.1'
return super().init_args(args)
def generate_dockerfile(self) -> str:
@@ -373,7 +379,14 @@ class StableGPUImageBuilder(Builder):
self.args.python_version)
extra_content += """
RUN export PIP_EXTRA_INDEX_URL=https://pypi.org/simple && \
pip install --no-cache-dir -U icecream soundfile pybind11 py-spy
pip install --no-cache-dir -U icecream soundfile pybind11 py-spy onnx onnxscript
"""
# vLLM >= 0.27 requires flashinfer >= 0.6.14, whose wheels are no longer
# published to PyPI. Pre-install it from flashinfer's own index so the later
# `pip install vllm` resolves without touching install.sh.
extra_content += """
RUN pip install --no-cache-dir flashinfer-python \
--extra-index-url https://flashinfer.ai/whl/ && pip cache purge
"""
version_args = (
f'{self.args.torch_version} {self.args.torchvision_version} {self.args.torchaudio_version} '
@@ -603,7 +616,7 @@ class AmdImageBuilder(Builder):
parts = version.strip().split('.')
if len(parts) >= 2 and parts[0].isdigit() and parts[1].isdigit():
return f'py{parts[0]}{parts[1]}'
return f'py{re.sub(r"[^0-9]", "", version)}'
return f'py{re.sub(r'[^0-9]', '', version)}'
@classmethod
def _run_capture(cls, *cmd: str) -> subprocess.CompletedProcess:
@@ -758,7 +771,7 @@ class AmdImageBuilder(Builder):
}
print('Probed AMD base image versions:')
for key, value in versions.items():
print(f' {key}: {value or "unknown"}')
print(f' {key}: {value or 'unknown'}')
return versions
def generate_dockerfile(self) -> str: