Seq-VCR: Preventing Collapse in Intermediate Transformer Representations for Enhanced Reasoning

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Arefin, Md Rifat, Subbaraj, Gopeshh, Gontier, Nicolas, LeCun, Yann, Rish, Irina, Shwartz-Ziv, Ravid, Pal, Christopher
Format: Preprint
Veröffentlicht: 2024
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866917962611425280
author Arefin, Md Rifat
Subbaraj, Gopeshh
Gontier, Nicolas
LeCun, Yann
Rish, Irina
Shwartz-Ziv, Ravid
Pal, Christopher
author_facet Arefin, Md Rifat
Subbaraj, Gopeshh
Gontier, Nicolas
LeCun, Yann
Rish, Irina
Shwartz-Ziv, Ravid
Pal, Christopher
contents Decoder-only Transformers often struggle with complex reasoning tasks, particularly arithmetic reasoning requiring multiple sequential operations. In this work, we identify representation collapse in the model's intermediate layers as a key factor limiting their reasoning capabilities. To address this, we propose Sequential Variance-Covariance Regularization (Seq-VCR), which enhances the entropy of intermediate representations and prevents collapse. Combined with dummy pause tokens as substitutes for chain-of-thought (CoT) tokens, our method significantly improves performance in arithmetic reasoning problems. In the challenging $5 \times 5$ integer multiplication task, our approach achieves $99.5\%$ exact match accuracy, outperforming models of the same size (which yield $0\%$ accuracy) and GPT-4 with five-shot CoT prompting ($44\%$). We also demonstrate superior results on arithmetic expression and longest increasing subsequence (LIS) datasets. Our findings highlight the importance of preventing intermediate layer representation collapse to enhance the reasoning capabilities of Transformers and show that Seq-VCR offers an effective solution without requiring explicit CoT supervision.
format Preprint
id arxiv_https___arxiv_org_abs_2411_02344
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Seq-VCR: Preventing Collapse in Intermediate Transformer Representations for Enhanced Reasoning
Arefin, Md Rifat
Subbaraj, Gopeshh
Gontier, Nicolas
LeCun, Yann
Rish, Irina
Shwartz-Ziv, Ravid
Pal, Christopher
Machine Learning
Computation and Language
Decoder-only Transformers often struggle with complex reasoning tasks, particularly arithmetic reasoning requiring multiple sequential operations. In this work, we identify representation collapse in the model's intermediate layers as a key factor limiting their reasoning capabilities. To address this, we propose Sequential Variance-Covariance Regularization (Seq-VCR), which enhances the entropy of intermediate representations and prevents collapse. Combined with dummy pause tokens as substitutes for chain-of-thought (CoT) tokens, our method significantly improves performance in arithmetic reasoning problems. In the challenging $5 \times 5$ integer multiplication task, our approach achieves $99.5\%$ exact match accuracy, outperforming models of the same size (which yield $0\%$ accuracy) and GPT-4 with five-shot CoT prompting ($44\%$). We also demonstrate superior results on arithmetic expression and longest increasing subsequence (LIS) datasets. Our findings highlight the importance of preventing intermediate layer representation collapse to enhance the reasoning capabilities of Transformers and show that Seq-VCR offers an effective solution without requiring explicit CoT supervision.
title Seq-VCR: Preventing Collapse in Intermediate Transformer Representations for Enhanced Reasoning
topic Machine Learning
Computation and Language
url https://arxiv.org/abs/2411.02344