rework this prompt: make large dataset for training text model to create 'prompt' based…
- User
rework this prompt:
make large dataset for training text model to create 'prompt' based on 'answer', most of the time there be image caption in 'answer', so make more fields with like: 'Make caption', 'Describe this image in fer words' and etc., make large variaty of different prompts.
Dataset should be in json format, with fields: 'good', 'bad', 'question' , 'bad' - answer, 'good' - prompt, 'question' - something like 'make prompt'
prompt should in some way reflect what`s inside 'good', for example, if answer is detailed, prompt should contain detailed, if its short, than it should contain short, etc.
- Assistant
Create a comprehensive dataset for training a text model to generate 'prompts' based on given 'answers'. The dataset should primarily focus on scenarios where the 'answer' includes image captions, leading to prompt variations such as 'Create a caption', 'Summarize this image in a few words', and more. Ensure there is a wide variety of different prompt types.
The dataset should be structured in JSON format, containing the fields: 'good', 'bad', 'question'. Here is the structure you'll need:
- 'good': Contains the prompt that effectively corresponds to the answer.
- 'bad': Represents a response that doesn't accurately reflect the provided good prompt.
- 'question': A guiding instruction like 'generate prompt'.
When crafting prompts, ensure they mirror the nature of the 'good' answer. For instance, if the answer is detailed, the prompt should ask for a detailed response. Conversely, if the answer is succinct, the prompt should request brevity.