metadata

license: other
tags:
  - trl
  - sft
  - generated_from_trainer
base_model: meta-llama/Meta-Llama-3-8B-Instruct
model-index:
  - name: Llama3-stanford-encyclopedia-philosophy-QA
    results: []
language:
  - en
pipeline_tag: text-generation

Llama3-stanford-encyclopedia-philosophy-QA

This model is a Qlora finetune of meta-llama/Meta-Llama-3-8B on the Stanford Encyclopedia of Philosophy-instruct dataset. It is meant for answering philosophical questions in a more formal tone.

Model description

The model was trained with the following system prompt:

"You are an expert and informative yet accessible Philosophy university professor. Students will pose you philosophical questions, answer them in a correct and rigorous but not to obscure way."

Furthermore, the chat dataset was formatted using the Llama3-instruct chat format:

<|begin_of_text|><|start_header_id|>system<|end_header_id|>

{{ system_prompt }}<|eot_id|><|start_header_id|>user<|end_header_id|>

{{ user_message }}<|eot_id|><|start_header_id|>assistant<|end_header_id|>

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 0.0002
train_batch_size: 8
eval_batch_size: 8
seed: 42
gradient_accumulation_steps: 4
total_train_batch_size: 32
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: constant
lr_scheduler_warmup_ratio: 0.03
num_epochs: 3

Training results

Framework versions

PEFT 0.10.0
Transformers 4.40.1
Pytorch 2.2.2+cu121
Datasets 2.19.0
Tokenizers 0.19.1