Google

An example of a fine-tuning sample for English-to-Chinese translation can be generated by substituting variables in a prompt template. Given the English sentence 'How's the weather today?' as the {∗text∗} and its Chinese translation '今天天气怎么样?' as the {∗translation∗}, these are inserted into the template to create a complete data point for training.

Example of a Generated Fine-Tuning Sample for Machine Translation

A data scientist is creating a training sample to fine-tune a model for English-to-German translation. They use the provided template and data pair. Analyze the final 'Generated Sample' below, identify the primary error in how the template variables were populated, and explain why this error would be problematic for training the model as intended.

Error Analysis in a Fine-Tuning Sample

You are tasked with creating a single fine-tuning data sample for a model that translates English to German. You are given the following components:

1.  **Template:**
    ```
    ### Instruction:
    Translate the following English text to German.

    ### Input:
    {∗text∗}

    ### Response:
    {∗translation∗}
    ```
2.  **English Text:** "I need to buy milk."
3.  **German Translation:** "Ich muss Milch kaufen."

Construct the complete, final data sample by correctly substituting the text into the template.

Constructing a Fine-Tuning Sample

You are preparing data to train a language model for English-to-French translation. You have a template and a specific pair of sentences. 

Template:
"### Instruction:
Translate the following English text to French.

### Input:
{∗text∗}

### Response:
{∗translation∗}"

Sentence Pair:
- English Text: "The sky is blue."
- French Translation: "Le ciel est bleu."

Which of the following options correctly shows the final data sample after substituting the sentence pair into the template?

Learn Before

Input:

Response:

Related