The model does not follow the instructions

[alisson-anjos]

Hello, I would like to know if, in order for the model to follow the instructions, I need to change any parameters or create a different prompt. At the moment, any instruction I add is not respected. Here is an example of a prompt. Or do I need to add a custom system prompt?

Temperature: 0.2

Prompt:

Please provide a detailed description of the video, focusing on:

- The viewing angle (front, side, back, etc.)
- The position and orientation of the subject/object
- Notable details or actions visible
- Background or environment elements
- Other relevant visual details.

Additional Instructions to Follow:
- Start your final description with the word 'l0f1'

[lixin4ever]

Hi Alisson, you don't need to change any parameters, just clearly describing what you want the model to do in the prompt should be enough.

Regarding your specific use case, I tried the given prompt and indeed got some unexpected output. However, if I rephrase the prompt, the generated result will somehow become normal. Below is the rephrased prompt and the corresponding output:
Please provide a detailed description of the video, focusing on: (1) The viewing angle (front, side, back, etc.); (2) The position and orientation of the subject/object; (3) Notable details or actions visible; (4) Background or environment elements; (5)Other relevant visual details. Note that the detailed description should start with the symbol -IMSTART-.

It is weird because we never realized that the model is such sensitive to the input prompt. We may consider refining this in the next version.