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/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 )"