I want to fine-tune a LLM with an instructions dataset, which consists of pairs of prompts and completions. I have seen a lot of tutorials on how to fine-tune LLMs with supervised datasets. Almost all of them use Trainer or SFTTrainer from Hugging Face. The strange thing that shocked me is that the…

Instruction tuning llm

nielsr January 1, 2024, 6:15pm 2

Hi,

That’s supported in the TRL library using the DataCollatorForCompletionOnlyLM class: Supervised Fine-tuning Trainer

1 Like

Topic		Replies	Views
Fine tune with SFTTrainer Intermediate	5	2906	March 15, 2024
Fine tune GPT2/LLaMA in seq2seq manner 🤗Transformers	2	908	January 14, 2024
Slower train with collator for completion only 🤗Transformers	1	670	April 7, 2024
Supervised Fine-tuning Trainer - where is the 'supervised' part? Beginners	0	350	July 3, 2023
How to train LLM only on response 🤗Transformers	2	1076	July 24, 2023