[AMD CI] Local cache fallback. (#13452)

Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>

scripts/ci/amd_ci_install_dependency.sh

@@ -12,7 +12,8 @@ else
 fi
 
 # Install the required dependencies in CI.
-docker exec ci_sglang pip install --upgrade pip
+docker exec ci_sglang chown -R root:root /sgl-data/pip-cache
+docker exec ci_sglang pip install --cache-dir=/sgl-data/pip-cache --upgrade pip
 docker exec ci_sglang pip uninstall sgl-kernel -y || true
 docker exec -w /sglang-checkout/sgl-kernel ci_sglang bash -c "rm -f pyproject.toml && mv pyproject_rocm.toml pyproject.toml && python3 setup_rocm.py install"
 
@@ -20,18 +21,18 @@ case "${GPU_ARCH}" in
   mi35x)
     echo "Runner uses ${GPU_ARCH}; will fetch mi35x image."
     docker exec ci_sglang rm -rf python/pyproject.toml && mv python/pyproject_other.toml python/pyproject.toml
-    docker exec ci_sglang pip install -e "python[dev_hip]" --no-deps # TODO: only for mi35x
+    docker exec ci_sglang pip install --cache-dir=/sgl-data/pip-cache -e "python[dev_hip]" --no-deps # TODO: only for mi35x
     # For lmms_evals evaluating MMMU
     docker exec -w / ci_sglang git clone --branch v0.4.1 --depth 1 https://github.com/EvolvingLMMs-Lab/lmms-eval.git
-    docker exec -w /lmms-eval ci_sglang pip install -e . --no-deps # TODO: only for mi35x
+    docker exec -w /lmms-eval ci_sglang pip install --cache-dir=/sgl-data/pip-cache -e . --no-deps # TODO: only for mi35x
     ;;
   mi30x|mi300|mi325)
     echo "Runner uses ${GPU_ARCH}; will fetch mi30x image."
     docker exec ci_sglang rm -rf python/pyproject.toml && mv python/pyproject_other.toml python/pyproject.toml
-    docker exec ci_sglang pip install -e "python[dev_hip]"
+    docker exec ci_sglang pip install --cache-dir=/sgl-data/pip-cache -e "python[dev_hip]"
     # For lmms_evals evaluating MMMU
     docker exec -w / ci_sglang git clone --branch v0.4.1 --depth 1 https://github.com/EvolvingLMMs-Lab/lmms-eval.git
-    docker exec -w /lmms-eval ci_sglang pip install -e .
+    docker exec -w /lmms-eval ci_sglang pip install --cache-dir=/sgl-data/pip-cache -e .
     ;;
   *)
     echo "Runner architecture '${GPU_ARCH}' unrecognised;" >&2
@@ -39,11 +40,11 @@ case "${GPU_ARCH}" in
 esac
 
 docker exec -w / ci_sglang git clone https://github.com/merrymercy/human-eval.git
-docker exec -w /human-eval ci_sglang pip install -e .
+docker exec -w /human-eval ci_sglang pip install --cache-dir=/sgl-data/pip-cache -e .
 
 docker exec -w / ci_sglang mkdir -p /dummy-grok
 mkdir -p dummy-grok && wget https://sharkpublic.blob.core.windows.net/sharkpublic/sglang/dummy_grok.json -O dummy-grok/config.json
 docker cp ./dummy-grok ci_sglang:/
 
-docker exec ci_sglang pip install huggingface_hub[hf_xet]
-docker exec ci_sglang pip install pytest
+docker exec ci_sglang pip install --cache-dir=/sgl-data/pip-cache huggingface_hub[hf_xet]
+docker exec ci_sglang pip install --cache-dir=/sgl-data/pip-cache pytest