Describe the bug
The MultiQueryRetriever which rewords the user prompts 2 times by writing 2 additonal prompts on each line in the output using GPT3.5 has observed an increase in error rate.
The root cause is as follows:
Normally when the prompt is reworded, GPT 3.5 generates the 2 prompts in 2 lines. Like below
However, randomly, as we have witness increasingly, somtimes there is a random empty line in between the 2 prompts, causing it to error out!
Expected behaviorMutliQueryRetriever should be able to regenerate reworded prompts without erroring.
Describe the bug The
MultiQueryRetrieverwhich rewords the user prompts 2 times by writing 2 additonal prompts on each line in the output using GPT3.5 has observed an increase in error rate.The root cause is as follows:
Normally when the prompt is reworded, GPT 3.5 generates the 2 prompts in 2 lines. Like below
However, randomly, as we have witness increasingly, somtimes there is a random empty line in between the 2 prompts, causing it to error out!
Expected behavior
MutliQueryRetrievershould be able to regenerate reworded prompts without erroring.