rmihaylov / mpttune

Tune MPTs
Apache License 2.0
84 stars 16 forks source link

Add argument for "input" along with "instruction" #6

Open sidharthiimc opened 1 year ago

sidharthiimc commented 1 year ago

image

rmihaylov commented 1 year ago

It is either prompt or instruction. You would feed in prompt when the model in hand is not finetuned with an instruction dataset. On the other hand, you will feed in an instruction when one uses an instruction-finetuned model.

sidharthiimc commented 1 year ago

I understand the prompt and instruction can't be passed together. Let's say I want to use one data point Alpaca dataset for inference after fine tuning. For example :

{ "instruction": "Identify the odd one out.", "input": "Twitter, Instagram, Telegram", "output": "Telegram" }

How can I pass this instruction + input in the generate.py?

I need an argument for "input" here -

image

or do you have any other solution for it? How can I structure it in prompt?