Build Stanford's ALPACA on a free Flan-T5-Large LLM. If you have the compute resources (GPU), why not try the Flan-T5-XXL (11B) model to fine-tune on the Alpaca 52K specific data set? Or create your own ALPACA style data set?
Enjoy wild and free Alpacas!
All files of my video are available via git clone for you (I was asked if one has no continuous online access to Jupyter or COLAB? So today .py!):
https://github.com/declare-lab/flan-alpaca
Notice: All rights are with Univ of Singapore for this code /git! Thank you to the team in Singapore.
New videos will include Parameter Efficient Fine-Tuning (PEFT) like Low-Rank Adaptation (LoRA) for memory optimization (incl INT.8 model weights and parameters), but enough with teasing future videos, let's enjoy this one.
ALPACA Data Set:
-----------------------------
@tatsu-lab/alpaca (https://crfm.stanford.edu/2023/03/13/alpaca.html)
Alpaca is a data set of 52,000 instructions:
The data fields are as follows:
A. instruction: describes the task the model should perform. Each of the 52K instructions is unique.
B. input: optional context or input for the task. For example, when the instruction is "Summarize the following article", the input is the article. Around 40% of the examples have an input.
C. output: the answer to the instruction as generated by OpenAI's text-davinci-003 (not for free).
D. text: the instruction, input and output formatted with the prompt template used by the authors for fine-tuning their models.
That's it! Could it be that a huge data set is enough to fine-tune even a tiny LLM (smaller 10GB) for a great performance on a downstream task? Supervised fine-tuning of LLMs on a huge data set really is THE best solution for performance? What a surprise! (Please note the sarcastic tone of my last remark...)
@misc{alpaca,
author = {Rohan Taori and Ishaan Gulrajani and Tianyi Zhang and Yann Dubois and Xuechen Li and Carlos Guestrin and Percy Liang and Tatsunori B. Hashimoto },
title = {Stanford Alpaca: An Instruction-following LLaMA model},
year = {2023},
publisher = {GitHub},
journal = {GitHub repository},
howpublished = {\url{https://github.com/tatsu-lab/stanford_alpaca}},
}
#ai
#finetuning
#finetuning
#languagemodel
#alpaca_LLM
23 Comments