You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Thanks for the great work. Unfortunately, we are unable to reproduce your results for ReAct / Reflexion on HotpotQA.
E.g. You say that ReAct+gpt-3.5-turbo has a baseline accuracy of 0.26 in Table 5 of your article. However, we tried to reproduce this result using gpt-3.5-turbo on your dataset hotpot-qa-distractor-sample.joblib, but only get a baseline(the 1st trial) accuracy of 0.09. You can see the detailed trajs here: https://github.com/haoyb22/Reflexion_hotpotqa/blob/main/100_questions_5_trials.txt
And we find that the trajs we got show diffferent features from yours. For example, the ReAct agent using gpt-3.5-turbo always tries to answer more than one step a time and doesn't follow the framework. The last bug causes error in Action steps and always stops the program, so we change your agent.py a bit just make the program continue to work. You can see the changed agent.py here: https://github.com/haoyb22/Reflexion_hotpotqa/blob/main/agents.py
Hi,
Thanks for the great work. Unfortunately, we are unable to reproduce your results for ReAct / Reflexion on HotpotQA.
E.g. You say that ReAct+gpt-3.5-turbo has a baseline accuracy of 0.26 in Table 5 of your article. However, we tried to reproduce this result using gpt-3.5-turbo on your dataset hotpot-qa-distractor-sample.joblib, but only get a baseline(the 1st trial) accuracy of 0.09. You can see the detailed trajs here: https://github.com/haoyb22/Reflexion_hotpotqa/blob/main/100_questions_5_trials.txt
And we find that the trajs we got show diffferent features from yours. For example, the ReAct agent using gpt-3.5-turbo always tries to answer more than one step a time and doesn't follow the framework. The last bug causes error in Action steps and always stops the program, so we change your agent.py a bit just make the program continue to work. You can see the changed agent.py here: https://github.com/haoyb22/Reflexion_hotpotqa/blob/main/agents.py
Please clarify how to get the results you get?
@noahshinn
The text was updated successfully, but these errors were encountered: