Currently, the prepare.sh script installs dependencies for both llm_judge and model_worker. The latter is needed because common.py depends on model_adapter.py, which depends on large dependencies like torch. However, this seems unnecessary. common.py only needs a few components from model_adapter.py. If we can separate those components, we may not need to install dependencies for model_worker.
Currently, the
prepare.shscript installs dependencies for bothllm_judgeandmodel_worker. The latter is needed becausecommon.pydepends onmodel_adapter.py, which depends on large dependencies liketorch. However, this seems unnecessary.common.pyonly needs a few components frommodel_adapter.py. If we can separate those components, we may not need to install dependencies formodel_worker.