Chat Bison

[ShreyasMahajan1]

I am trying to generate few questions using chat bison.  Here are the topics : Classical dances, Temple architecture, World War I and II, Current affairs. But the questions which are generated are something totally different from the above mentioned topic. Here is a question which is generated for instance : 

"प्रश्न : किस मुगल शासक ने अपने शासनकाल में 'जजिया कर' को समाप्त कर दिया था? " (Translation : "Question: Which Mughal ruler abolished 'Jizya tax' during his reign?") As you can see the question generated is not relevant to the given topics. The question generated is related to Mughal era which is not what I want and not what I have specified.

Why is this happening and how can i prevent this. Any help will be appreciated. Thanks.  

[Poala_Tenorio]

Here are a few steps you can take to ensure relevance:

• Check Input Prompt: Ensure that your input prompt clearly specifies the topics you want questions on. For example, your prompt should include phrases like "Generate questions on Classical dances, Temple architecture, World War I and II, Current affairs."
• Model Training: If you're fine-tuning or training your own model, make sure you're using a dataset that primarily focuses on the topics you're interested in. If the model is fine-tuned on a diverse dataset, it might generate questions from a wide range of topics.
• Input Examples: Provide specific examples related to the topics you want questions on in your input prompt. This can help guide the model to focus on generating questions relevant to those topics.
• Review Generated Questions: Always review the questions generated to ensure they align with your desired topics. If you notice irrelevant questions, try adjusting your input prompt or providing more specific examples.
• Experiment with Prompt Formatting: Sometimes, tweaking the format of your input prompt can lead to more relevant outputs. You can try different variations to see what works best.

By following these steps, you should be able to generate questions more closely aligned with the topics you're interested in. If you continue to encounter issues, feel free to provide more details or examples for further assistance.

[nimrah-waqar]

The mismatch between the specified topics and the generated questions may be due to several factors:

1. Training Data: The model might have been trained on diverse topics, including historical events like the Mughal era.

2. Topic Specification: Ensure that the specified topics align accurately with the desired questions.

3. Fine-tuning: Consider fine-tuning the model on a dataset specific to the desired topics.

4. Quality Control: Implement manual review or filtering of generated questions to ensure relevance.

To prevent this, verify topic specification, consider fine-tuning, and implement quality control measures.