Improve linear.py to load sharded weights & remove the dependency of Parameters from vllm (#2784)

Co-authored-by: SangBin Cho rkooo567@gmail.com

scripts/killall_sglang.sh

@@ -7,6 +7,7 @@ nvidia-smi
 kill -9 $(ps aux | grep 'sglang::' | grep -v 'grep' | awk '{print $2}') 2>/dev/null
 kill -9 $(ps aux | grep 'sglang.launch_server' | grep -v 'grep' | awk '{print $2}') 2>/dev/null
 kill -9 $(ps aux | grep 'sglang.bench' | grep -v 'grep' | awk '{print $2}') 2>/dev/null
+kill -9 $(ps aux | grep 'sglang.data_parallel' | grep -v 'grep' | awk '{print $2}') 2>/dev/null
 
 # Clean all GPU processes if any argument is provided
 if [ $# -gt 0 ]; then