Adding --in-suffix option

[44670]

Hi!

I would like to submit a pull request to your project for adding the "--in-suffix" option. This option will allow users to specify the input file suffix when running the LLM model on CPU.

As a team working on OpenBuddy, a multilingual open model with the ability to understand user's questions and generate creative contents, we find llama.cpp's project incredibly useful for running LLM models on personal hardware. We appreciate your hard work and dedication in creating this project.

Attached is a screenshot showing successful testing on our own model. We follow a prompt format of "User: [question]\nAssistant:", which requires us to add "Assistant:" after the user's input in interactive mode for the model to understand the role switch and output the answer correctly.

Thank you for considering this pull request. We look forward to contributing to the llama.cpp project.

[44670]

By the way, here is the command-line that we have been testing:

bin/main -m 7b-q4_0.bin --interactive-first --reverse-prompt "User:" --in-prefix " " --in-suffix "Assistant:" -f prompt.txt

Please let me know if you need further explanation.

[44670]

Hi, I have added a new commit that will print the input-suffix on the screen before generating.

This will not affect the generated output, but it will enhance the user experience by making it feel more like a conversation.

[ggerganov]

Feel free to add a link to OpenBuddy to the README!

[44670]

Thanks for merging! And we are more than happy to add our model into the readme file!

We will make another pull request soon.