Saved in:
Bibliographic Details
Main Authors: Chen, Nuo, Wu, Ning, Liang, Shining, Gong, Ming, Shou, Linjun, Zhang, Dongmei, Li, Jia
Format: Preprint
Published: 2023
Subjects:
Online Access:https://arxiv.org/abs/2312.04333
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914634976460800
author Chen, Nuo
Wu, Ning
Liang, Shining
Gong, Ming
Shou, Linjun
Zhang, Dongmei
Li, Jia
author_facet Chen, Nuo
Wu, Ning
Liang, Shining
Gong, Ming
Shou, Linjun
Zhang, Dongmei
Li, Jia
contents This paper presents an in-depth analysis of Large Language Models (LLMs), focusing on LLaMA, a prominent open-source foundational model in natural language processing. Instead of assessing LLaMA through its generative output, we design multiple-choice tasks to probe its intrinsic understanding in high-order tasks such as reasoning and computation. We examine the model horizontally, comparing different sizes, and vertically, assessing different layers. We unveil several key and uncommon findings based on the designed probing tasks: (1) Horizontally, enlarging model sizes almost could not automatically impart additional knowledge or computational prowess. Instead, it can enhance reasoning abilities, especially in math problem solving, and helps reduce hallucinations, but only beyond certain size thresholds; (2) In vertical analysis, the lower layers of LLaMA lack substantial arithmetic and factual knowledge, showcasing logical thinking, multilingual and recognitive abilities, with top layers housing most computational power and real-world knowledge.
format Preprint
id arxiv_https___arxiv_org_abs_2312_04333
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Is Bigger and Deeper Always Better? Probing LLaMA Across Scales and Layers
Chen, Nuo
Wu, Ning
Liang, Shining
Gong, Ming
Shou, Linjun
Zhang, Dongmei
Li, Jia
Computation and Language
This paper presents an in-depth analysis of Large Language Models (LLMs), focusing on LLaMA, a prominent open-source foundational model in natural language processing. Instead of assessing LLaMA through its generative output, we design multiple-choice tasks to probe its intrinsic understanding in high-order tasks such as reasoning and computation. We examine the model horizontally, comparing different sizes, and vertically, assessing different layers. We unveil several key and uncommon findings based on the designed probing tasks: (1) Horizontally, enlarging model sizes almost could not automatically impart additional knowledge or computational prowess. Instead, it can enhance reasoning abilities, especially in math problem solving, and helps reduce hallucinations, but only beyond certain size thresholds; (2) In vertical analysis, the lower layers of LLaMA lack substantial arithmetic and factual knowledge, showcasing logical thinking, multilingual and recognitive abilities, with top layers housing most computational power and real-world knowledge.
title Is Bigger and Deeper Always Better? Probing LLaMA Across Scales and Layers
topic Computation and Language
url https://arxiv.org/abs/2312.04333