Update installation instructions to prevent/resolve nccl error, update poetry lock to vllm 0.4.2, update flash attention instructions for vllm 0.4.2

Author: XkunW

README.md

@@ -23,6 +23,7 @@ pip install wheel
 export CUDA_HOME=/pkgs/cuda-12.3
 
 pip install flash-attn --no-build-isolation
+pip install vllm-flash-attn
 ```
 
 ## Launch an inference server