Rename serverless test file, set default model to Phi 2 for test, removed jq install, and env vars that are set the same in utils.py already, ignore .cache in git

.gitignore

@@ -1 +1,2 @@
 .aider*
+.cache

runpod.dockerfile

@@ -19,16 +19,13 @@ RUN python3.11 -m pip install pytest cmake \
 # Install llama-cpp-python (build with cuda)
 ENV CMAKE_ARGS="-DLLAMA_CUBLAS=on"
 RUN python3.11 -m pip install git+https://github.com/lukestanley/llama-cpp-python.git@expose_json_grammar_convert_function --upgrade --no-cache-dir --force-reinstall
-RUN apt-get update; apt-get install jq -y
-ADD runpod_handler.py .
 
+ADD runpod_handler.py .
 ADD chill.py .
 ADD utils.py .
 ADD promptObjects.py .
 
-#ENV REPO_ID="TheBloke/phi-2-GGUF"
-#ENV MODEL_FILE="phi-2.Q2_K.gguf"
-ENV N_GPU_LAYERS=-1
-ENV CONTEXT_SIZE=2048
+ENV REPO_ID="TheBloke/phi-2-GGUF"
+ENV MODEL_FILE="phi-2.Q2_K.gguf"
 CMD python3.11 -u /runpod_handler.py