Add transformers patch layer for GLM-5/Qwen3.5 model type support

SGLang v0.5.10 pins an older transformers that doesn't have
glm_moe_dsa or qwen3_5_moe in its AutoConfig registry. Add a
thin Dockerfile layer that upgrades transformers to the commit
with these model types and verifies registration.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: JordanNanos

docker/build-sglang-bnxt-mi325x.sh

@@ -79,6 +79,17 @@ docker build \
     -f "${BUILD_CONTEXT}/rocm.Dockerfile" \
     "${BUILD_CONTEXT}/"
 
+# ---------- Patch transformers for GLM-5/Qwen3.5 model type support ----------
+PATCH_DOCKERFILE="$(dirname "$0")/patch-transformers.Dockerfile"
+if [[ -f "${PATCH_DOCKERFILE}" ]]; then
+    echo "[build] Patching transformers for glm_moe_dsa/qwen3_5_moe support"
+    docker build \
+        --build-arg BASE_IMAGE="${IMAGE_TAG}" \
+        -t "${IMAGE_TAG}" \
+        -f "${PATCH_DOCKERFILE}" \
+        "$(dirname "$0")/"
+fi
+
 # ---------- Push ----------
 echo "[build] Pushing ${IMAGE_TAG}"
 docker push "${IMAGE_TAG}"

docker/patch-transformers.Dockerfile

@@ -0,0 +1,10 @@
+ARG BASE_IMAGE
+FROM ${BASE_IMAGE}
+
+# Upgrade transformers to get glm_moe_dsa and qwen3_5_moe model type support.
+# The SGLang v0.5.10 base image pins an older transformers that doesn't have these.
+RUN pip install --no-cache-dir -U \
+    "git+https://github.com/huggingface/transformers.git@6ed9ee36f608fd145168377345bfc4a5de12e1e2" \
+    && python3 -c "import transformers; print(f'transformers {transformers.__version__}')" \
+    && python3 -c "from transformers import AutoConfig; AutoConfig.for_model('glm_moe_dsa')" \
+    && echo "glm_moe_dsa model type verified"