Missing `jax_agent` module
Opened this issue · 0 comments
I was trying evaluate a BC trained model/policy using the script llm_rl_scripts/car_dealer/misc/eval_car_dealer.py
but I cannot find a jax_agent
module. (Actually, it doesn't find algorithms
. I assumed that it was a reference to the package LLM_RL/algorithms/
but that module is not there.)
This also affects other scripts as a quick grep search finds:
llm_rl_scripts/car_dealer/misc/cache_embs.py:from algorithms.jax_agent import JaxAgentPolicy, Inference
llm_rl_scripts/car_dealer/misc/eval_car_dealer.py:from algorithms.jax_agent import Inference, JaxAgentPolicy
llm_rl_scripts/car_dealer/mc_returns/train_mc.py:from algorithms.jax_agent import Inference, JaxAgentPolicy, ReRankerPolicy, build_agent_proposer
llm_rl_scripts/car_dealer/ilql/train_ilql.py:from algorithms.jax_agent import Inference, JaxAgentPolicy, ReRankerPolicy, build_agent_proposer
llm_rl_scripts/twenty_questions/misc/eval_guesser.py:from algorithms.jax_agent import Inference, JaxAgentPolicy
Notice that this had already been noted in #7 but left unresolved when closed.
By the way, that eval script follows a different pattern that the one presented in the README but it was the only I could find for the car_dealer
environment. Are there other ways to evaluate the policies? And, even better, is there a way to interact with a learned seller policy (instead of the LLM buyer that is)?