Run adapter verification after GPU allocation

Dockerfile

@@ -43,8 +43,7 @@ RUN pip install --no-cache-dir \
 
 ENV PYTHONPATH=/opt/sglang-omni
 
-RUN /opt/audio8-tts/sglang_omni/scripts/install_adapter.sh /opt/sglang-omni && \
-    python3 /opt/audio8-tts/sglang_omni/scripts/verify_install.py
+RUN /opt/audio8-tts/sglang_omni/scripts/install_adapter.sh /opt/sglang-omni
 
 COPY --chown=1000:1000 app.py ./app.py
 COPY --chown=1000:1000 static ./static
@@ -58,8 +57,7 @@ USER 1000
 
 ARG MODEL_REPO=Audio8/Audio8-TTS-Preview-0.6b
 ARG MODEL_REVISION=1b17c91db5f4dccb6914aa4aa5cb0e56661a6c17
-RUN python3 -c "from huggingface_hub import snapshot_download; snapshot_download(repo_id='${MODEL_REPO}', revision='${MODEL_REVISION}', local_dir='/models/audio8-tts-0.6b')" && \
-    python3 /opt/audio8-tts/sglang_omni/scripts/verify_install.py --model-path /models/audio8-tts-0.6b
+RUN python3 -c "from huggingface_hub import snapshot_download; snapshot_download(repo_id='${MODEL_REPO}', revision='${MODEL_REVISION}', local_dir='/models/audio8-tts-0.6b')"
 
 ENV HOME=/home/user \
     HF_HOME=/home/user/.cache/huggingface \

scripts/run_backend.sh

@@ -8,6 +8,9 @@ export SGLANG_OMNI_ROOT="${SGLANG_OMNI_ROOT:-/opt/sglang-omni}"
 export PYTHONPATH="${SGLANG_OMNI_ROOT}${PYTHONPATH:+:${PYTHONPATH}}"
 export FLASHINFER_WORKSPACE_BASE="${FLASHINFER_WORKSPACE_BASE:-/tmp/audio8-flashinfer}"
 
+python3 "${ADAPTER_ROOT}/sglang_omni/scripts/verify_install.py" \
+  --model-path "${MODEL_PATH}"
+
 if [[ "${AUDIO8_TTS_VERIFY_FAST_KV:-0}" == "1" ]]; then
   python3 "${ADAPTER_ROOT}/sglang_omni/scripts/test_fast_kv_cache.py" \
     --model-path "${MODEL_PATH}" \