Stanford Alpaca Regenerate: An Instruction-following LLaMA Model

Updated to the latest OpenAI Python SDK. For the data generation part, we mostly keep it and instead replace the model training part with axolotl config.

create dataset

You need a FRIENDLI_TOKEN. You can get it from the document.

## Generate 52 new data -> regen.json
python -m generate_instruction generate_instruction_following_data \
  --output_dir ./ \
  --num_instructions_to_generate 52

# convert regen.json to jsonl -> output.jsonl
python convert_to_jsonl.py

Example of completed data

$ head -n 1 output.jsonl
{"instruction": "What can you infer from the following conversation?", "input": "John: How was your weekend?\nJane: It was great. I went to the beach with friends and had a lot of fun.", "output": "Jane had a great weekend and enjoyed her time at the beach with friends."}