Earn 25,200 ($252.00)

DESCRIPTION
A program to intake multiple writing samples and fine-tune a GPT-3 model (either DaVinci or Curie) to match the voice/linguistic style of the writing samples. The program will use the fine-tuned model to generate an output in the same writing style when given a prompt. The output should be at least five paragraphs in length, equivalent to a typical blog post. There are no interface or hosting requirements, a command line/ locally hosted solution is fine.

REQUIREMENTS

• The program must intake writing samples in the form of .txt files and use them to fine-tune a GPT-3 model.
• The user should be able to select the base model for fine-tuning, either DaVinci or Curie.
• The fine-tuned model should not alter the content of the output, only the writing style.
• The program should generate two writing samples: one using the base model and one using the fine-tuned model. Both samples should have a minimum length of five paragraphs.
• The program should provide metrics for evaluating the performance of the fine-tuned model, such as weights and the gradient of loss.
• The program should be written in Python and include thorough comments in the code.
• The program should be easy for someone with basic programming knowledge to use, even if it is a command-line application.
• The program should be able to process any amount of input text, regardless of the number of .txt files used.
• The program should create a single jsonl file or an alternative method for fine-tuning the model.

EXPECTED FLOW

1. The user will input multiple .txt files containing writing samples.
2. The program will create a file to fine-tune the model using the input writing samples.
3. The program will fine-tune the selected GPT-3 model (either DaVinci or Curie).
4. The user will enter a prompt for the program to generate an output.
5. The program will generate two writing samples: one using the base model and one using the fine-tuned model.
6. The program will output performance metrics for the fine-tuned model.

ACCEPTANCE CRITERIA

• The program should intake writing samples in the form of .txt files and use them to fine-tune a GPT-3 model.
• The base model for fine-tuning should be selectable by the user between DaVinci and Curie.
• The fine-tuned model should not alter the content of the output, only the writing style.
• The program should generate two writing samples: one using the base model and one using the fine-tuned model. Both samples should have a minimum length of five paragraphs.
• The program should provide metrics for evaluating the performance of the fine-tuned model, such as weights and the gradient of loss.
• The program should be written in Python and include thorough comments in the code.
• The program should be easy for someone with basic programming knowledge to use, even if it is a command-line application.
• The program should be able to process any amount of input text, regardless of the number of .txt files used.
• The program should create a single jsonl file or an alternative method for fine-tuning the model.
• The program should correctly follow the expected flow as described in the requirements.