Fix uv_init_torch

Author: junwha

uv/Dockerfile

@@ -69,10 +69,13 @@ BASE_DIR=/ddiff-base/py\${PY_VERSION}-torch\${TORCH_VER}
 mkdir -p "\$BASE_DIR" && cd "\$BASE_DIR"
 uv venv --python "\${PY_VERSION}" --system-site-packages
 uv pip install --no-cache torch=="\${TORCH_VER}" torchvision=="\${VISION_VER}" torchaudio=="\${TORCH_VER}" --index-url "https://download.pytorch.org/whl/\${CUDA_VER}"
-echo "alias uv_init_torch\${TORCH_VER}='ln -sfn \${BASE_DIR}/.venv ./.venv && echo Created uv virtual environment with torch==\${TORCH_VER}'" >> ~/.bashrc
+echo "ln -sfn \${BASE_DIR}/.venv ./.venv && echo Created uv virtual environment with torch==\${TORCH_VER}" > /usr/local/bin/uv_init_torch\${TORCH_VER}
+chmod +x /usr/local/bin/uv_init_torch\${TORCH_VER}
 EOF
 RUN chmod +x /usr/local/bin/install_torch_env
 
+ENV UV_NO_CACHE=1
+
 # pytorch 2.4.1 (251209 Update)
 RUN install_torch_env 2.4.1 0.19.1 cu124
 

uv/README.md

@@ -16,17 +16,17 @@ We provide a monolithic base image with various PyTorch versions and essential l
 ### 1. Docker file example
 
 ```dockerfile
-FROM junwha/ddiff-base:cu12.4.1-py3.10-torch-251205
+FROM junwha/ddiff-base:cu12.4.1-py3.10-torch-251210
 
 WORKDIR /<your workspace>
 
 # 1. Initialize venv linked to a specific Torch version
 # (Note: This creates a symlink to the base image's venv, thus only one project per torch version is recommended)
-RUN uv_init_torch2.5.1
+RUN bash -c uv_init_torch2.5.1
 
 # 2. Install additional packages (without torch)
 COPY pyproject.yaml /<your workspace>
-RUN uv pip install . --no-cache --no-install torch --no-install torchvision --no-install torchaudio
+RUN uv pip install --no-install torch --no-install torchvision --no-install torchaudio . 
 
 ```
 
@@ -48,7 +48,7 @@ you can use standard uv commands. The environment is automatically detected via
     * **Build Tools:** gcc-12 g++-12 ninja-build cmake build-essential 
     * **Profiler:** Nsight System 2024
 ## Releases
-**Image Tag:** `junwha/ddiff-base:cu12.4.1-py3.10-torch-251205`
+**Image Tag:** `junwha/ddiff-base:cu12.4.1-py3.10-torch-251210`
   * **Python:** 3.10
   * **Pre-installed PyTorch Versions:** 2.4.1, 2.5.1, 2.6.0, 2.7.1, 2.9.0
   * **Compute Capabilities**: 7.0 7.5 8.0 8.6 8.9 9.0+PTX