Convert and quantize script

Author: adriacabeza

convert_and_quantize.sh

@@ -0,0 +1,28 @@
+echo "💪🏼 Download and compile Llama.cpp"
+if [ ! -f "./llama.cpp/main" ]
+then
+  if [ ! -d "./llama.cpp/" ]
+  then
+  git clone https://github.com/StrikingLoo/llama.cpp
+  fi
+  cd llama.cpp && \
+    make && \
+    mv main ../api/llama
+    cd ..
+else
+  echo "✅ Repository already cloned and compiled"
+fi
+
+echo "💬 Convert weights"
+if ls models/*/*.bin 1> /dev/null 2>&1; then
+    echo "✅ Weights already converted"
+else
+  if ls models/*/*.pth 1> /dev/null 2>&1; then
+    # Convert model to ggml FP16 format
+    python3 llama.cpp/convert-pth-to-ggml.py weights/7B/ 1
+    # Quantize the model to 4-bits
+    python3 llama.cpp/quantize.py weights/7B
+  else
+    echo "⚠️ Weight files do not exist. Please download them and place them inside the models folder"
+  fi
+fi