[AMD] CI: pin antlr4-python3-runtime back after lmms-eval (unblocks ROCm 7.2 stage-b evals) (#34768)

scripts/ci/amd/amd_ci_install_dependency.sh

@@ -150,6 +150,12 @@ else
   docker exec ci_sglang git config --global --add safe.directory /lmms-eval
   install_with_retry docker exec -w /lmms-eval ci_sglang pip install --cache-dir=/sgl-data/pip-cache -e .
 
+  # lmms-eval v0.4.1 pulls latex2sympy2, which pins antlr4-python3-runtime==4.7.2
+  # and uninstalls the 4.9.3 that sgl-eval's latex2sympy2_extended requires, so
+  # every `sgl-eval run mmlu` dies with "Unsupported ANTLR version 4.7.2". Pin it
+  # back, the same way the CUDA installer does after its own lmms-eval install.
+  install_with_retry docker exec ci_sglang pip install --cache-dir=/sgl-data/pip-cache "antlr4-python3-runtime==4.9.3" --force-reinstall --no-deps
+
   git_clone_with_retry https://github.com/akao-amd/human-eval.git human-eval
   docker cp human-eval ci_sglang:/
   install_with_retry docker exec -w /human-eval ci_sglang pip install --cache-dir=/sgl-data/pip-cache -e .