Alpaca format mapping?

[EllenOrange]

Hi folks,

I have a dataset that is structured like this:

    {
        "instruction": "<my instruction>",
        "input": "<my input>",
        "output": "<my response>"
    }

I'm doing SFT training with --template alpaca

I ran an evaluation of my lora traning and it reported that the format it's using is:

<s> Below is an instruction that describes a task. Write a response that appropriately completes the request.

### Instruction:
<my instruction>
<my input>

### Response:

The model I'm doing the SFT training against expects

<s> Below is an instruction that describes a task. Write a response that appropriately completes the request.

### Instruction:
<my instruction>
### Input:
<my input>
### Response:

Looking in template.py it seems the result I'm getting is what the code is aiming to do, but I don't see a template that will result in the format the model expects.

Is there any way of accomplishing this?

[hiyouga]

We adopted the alpaca template without an input field since most samples in alpaca 52k have no input.
Using the alpaca template in this framework, you can choose to pre-process the training set by prepending the tokens to the inputs:

{
  "instruction": "<your instruction>",
  "input": "### Input:\n<your input>",
  "output": "<your response>"
}

[EllenOrange]

Makes sense, thank you!