4 ms·
We got very similar findings: we published a paper that show that smaller LLMs (3-7b) when finetuned with LoRA can match or outperform GPT-4 on a variety of tas
by w4nderlust 2y ago
We got very similar findings: we published a paper that show that smaller LLMs (3-7b) when finetuned with LoRA can match or outperform GPT-4 on a variety of tasks (29 out of 31) including classification, summarization, info extraction, "reasoning". https://arxiv.org/abs/2405.00732 https://arxiv.org/abs/2405.00732 (Predibase cofounder and coauthor of the paper)