Bump vLLM version + more options when loading models in vLLM (#1782)

* Bump vLLM version to 0.3.2

* Add vLLM model loading options

* Remove transformers-exllama

* Fix install exllama

backend/python/exllama/Makefile

@@ -1,7 +1,8 @@
+export CONDA_ENV_PATH = "exllama.yml"
+
 .PHONY: exllama
 exllama:
-	$(MAKE) -C ../common-env/transformers
-	bash install.sh
+	bash install.sh ${CONDA_ENV_PATH}
 
 .PHONY: run
 run:

backend/python/exllama/install.sh

@@ -1,14 +1,22 @@
 #!/bin/bash
+set -ex
 
-##
-## A bash script installs the required dependencies of VALL-E-X and prepares the environment
 export PATH=$PATH:/opt/conda/bin
 
-# Activate conda environment
-source activate transformers
+# Check if environment exist
+conda_env_exists(){
+    ! conda list --name "${@}" >/dev/null 2>/dev/null
+}
 
-echo $CONDA_PREFIX
+if conda_env_exists "exllama" ; then
+    echo "Creating virtual environment..."
+    conda env create --name exllama --file $1
+    echo "Virtual environment created."
+else
+    echo "Virtual environment already exists."
+fi
 
+source activate exllama
 
 git clone https://github.com/turboderp/exllama $CONDA_PREFIX/exllama && pushd $CONDA_PREFIX/exllama && pip install -r requirements.txt && popd
 

backend/python/exllama/run.sh

@@ -2,11 +2,10 @@
 
 ##
 ## A bash script wrapper that runs the exllama server with conda
-
 export PATH=$PATH:/opt/conda/bin
 
 # Activate conda environment
-source activate transformers
+source activate exllama
 
 # get the directory where the bash script is located
 DIR="$( cd "$( dirname "${BASH_SOURCE[0]}" )" >/dev/null 2>&1 && pwd )"