#!/bin/bash export VLLM_ALLOW_RUNTIME_LORA_UPDATING=True vllm serve $MODEL \ --enable-lora \ --dtype bfloat16 \ --max-lora-rank 64 \ --enable-prefix-caching \ > $(readlink -ef $(dirname $0))/vllm.log \ 2> $(readlink -ef $(dirname $0))/vllm.err # --lora-modules "{\"name\": \"myadapter\", \"path\": \"$ADAPTER\"}" \ # --lora-modules "{\"name\": \"myadapter\", \"path\": \"$ADAPTER\", \"base_model_name\": \"$MODEL\"}" \