fix bug

Dockerfile

@@ -7,6 +7,11 @@ RUN apt-get update && \
     git lfs install && \
     rm -rf /var/lib/apt/lists/*
 
+COPY --chown=user ./vllm_hacked/model_executor/models/llama.py /opt/conda/lib/python3.11/site-packages/vllm/model_executor/models/llama.py
+COPY --chown=user ./vllm_hacked/v1/sample/sampler.py /opt/conda/lib/python3.11/site-packages/vllm/v1/sample/sampler.py
+COPY --chown=user ./vllm_hacked/v1/sample/metadata.py /opt/conda/lib/python3.11/site-packages/vllm/v1/sample/metadata.py
+COPY --chown=user ./vllm_hacked/sampling_params.py /opt/conda/lib/python3.11/site-packages/vllm/sampling_params.py
+
 RUN useradd -m -u 1000 user
 USER user
 ENV PATH="/home/user/.local/bin:$PATH"
@@ -17,4 +22,4 @@ COPY --chown=user ./requirements.txt requirements.txt
 RUN pip install --no-cache-dir --upgrade -r requirements.txt
 
 COPY --chown=user . /app
-CMD ["sh", "app.sh"]
+CMD ["python3", "app.py"]

app.sh

@@ -1,15 +0,0 @@
-export USER=root
-export PYTHONDONTWRITEBYTECODE=1
-export TRANSFORMERS_CACHE="$(pwd)/third_party/hub"
-export NCCL_HOME=/usr/local/tccl
-export PYTHONPATH="$(pwd)/codeclm/tokenizer/":"$(pwd)":"$(pwd)/codeclm/tokenizer/Flow1dVAE/":"$(pwd)/codeclm/tokenizer/":$PYTHONPATH
-export OMP_NUM_THREADS=1
-export MKL_NUM_THREADS=1
-export CUDA_LAUNCH_BLOCKING=0
-
-
-sudo cp ./vllm_hacked/model_executor/models/llama.py /opt/conda/lib/python3.11/site-packages/vllm/model_executor/models/llama.py
-sudo cp ./vllm_hacked/v1/sample/sampler.py /opt/conda/lib/python3.11/site-packages/vllm/v1/sample/sampler.py
-sudo cp ./vllm_hacked/v1/sample/metadata.py /opt/conda/lib/python3.11/site-packages/vllm/v1/sample/metadata.py
-sudo cp ./vllm_hacked/sampling_params.py /opt/conda/lib/python3.11/site-packages/vllm/sampling_params.py
-python3 app.py