README.md

metadata

base_model: daryl149/llama-2-7b-chat-hf
datasets:
  - Hypersniper/philosophy_dialogue

Overview

Fine-tuned Llama-2 7B with a philosopher conversation dataset (originally from Hypersniper/philosophy_dialogue). Used QLoRA for fine-tuning. Trained for one epoch on a 40GB GPU (NVIDIA A100) instance.

The version here is the fp16 HuggingFace model.

Prompt style

The model was trained with the following prompt style:

<s>[INST] <<SYS>>
{{ system_prompt }}
<</SYS>>

{{ user_message }} [/INST]
{{ response }} </s>