Training mistral instruct - dataset and prompt format

[dinonovak]

Hi,
I am trying to set up training for mistral instruct, can you please advise what dataset config to use.
I have dataset configured in instruct format https://huggingface.co/datasets/dnovak232/sql_create_context-v4-mssql-instruct

but how do I need to configure .yaml for it. I see this recommendation for instruct format below, but not sure as mistral uses [INST] {instruction} [/INST] {answer} ?

datasets:

• path: repo
  type:
  system_prompt: ""
  field_system: system
  format: "[INST] {instruction} [/INST]"
  no_input_format: "[INST] {instruction} [/INST]"

[gardner]

It looks like it matches the alpaca format described on the README.md

Have you tried something like this?

datasets:
  - path: dnovak232/sql_create_context-v4-mssql-instruct
    type: alpaca

[NanoCode012]

Yes! The one you said works. Alternatively, you can use the following

type: sharegpt
conversation: mistral