Datasets:
What is prompt for test commandR
Could you release the prompt for test commandR, I get a very low score to test commandr, thanks for your help
Hi,
We observe that Command-R-Plus usually generates very long answer, while the output length of ChatRAG-Bench are usually not very long or in about one sentence.
We follow the similar instruction strategy to ChatQA. For doc2dial, quac, qrecc, inscit, and hybridial, we give an additional instruction, "Please give an answer in just one sentence"; For coqa which has very short answer, we give an additional instruction, "Answer the following question with a short span. The answer needs to be just in a few words"; For topiocqa with fluctuating output lengths, we give an additional instruction, "Answer the following questions in JUST a few words or one sentence."; We give, "Answer the following questions with one or a list of entities.", for sqa, and "Answer the following questions with just a number from context or just the math arithmetic using +, -, *, or /." for convfinqa.