This repository contains the code and data for fine-tuning a Llama-7B model on a dataset containing user inputs and mid-journey prompts as responses.

The dataset used for fine-tuning the model is stored in a variable named fine_tuning_data. It contains three columns: User, Prompt, and Text.

The User column contains the user inputs, while the Prompt column contains the corresponding mid-journey prompts. The Text column is generated using the following template: ### Instruction: <USER PROMPT> ### Response: <RESPONSE / MID-JOURNEY PROMPTS

This template adds special tokens ### Instruction: and ### Response: to the user input and mid-journey prompt, respectively. These special tokens are used by the model to distinguish between the instruction and response parts of the text.

The Llama-7B model is fine-tuned on the fine_tuning_data dataset using standard techniques for fine-tuning language models. The fine-tuned model can then be used to generate mid-journey prompts given a user input.

The following bitsandbytes quantization config was used during training:

• load_in_8bit: False
• load_in_4bit: True
• llm_int8_threshold: 6.0
• llm_int8_skip_modules: None
• llm_int8_enable_fp32_cpu_offload: False
• llm_int8_has_fp16_weight: False
• bnb_4bit_quant_type: nf4
• bnb_4bit_use_double_quant: True
• bnb_4bit_compute_dtype: bfloat16

• PEFT 0.4.0