new vllm support

Dockerfile

@@ -2,8 +2,6 @@
 #FROM vllm/vllm-openai:deepseekv4-cu130
 FROM vllm/vllm-openai:v0.20.0-cu130
 
-RUN curl -fsSL https://raw.githubusercontent.com/vllm-project/vllm/v0.20.0/tools/install_deepgemm.sh | bash
-
 # Fix the broken ass nightly build that forgot to include pandas
 #RUN pip install --no-cache-dir pandas
 
@@ -24,6 +22,9 @@ RUN apt-get update && apt-get install -y git \
     pip install --no-cache-dir --no-build-isolation . && \
     rm -rf /tmp/lmcache && export CACHE_BUSTER=2
 
+# Needs git to complete
+RUN curl -fsSL https://raw.githubusercontent.com/vllm-project/vllm/v0.20.0/tools/install_deepgemm.sh | bash
+
 # Copy over nemotron reasonong parser
 COPY ./super_v3_reasoning_parser.py /opt/super_v3_reasoning_parser.py