DiscoverArxiv PapersCritical Tokens Matter: Token-Level Contrastive Estimation Enhence LLM’s Reasoning Capability
Critical Tokens Matter: Token-Level Contrastive Estimation Enhence LLM’s Reasoning Capability

Critical Tokens Matter: Token-Level Contrastive Estimation Enhence LLM’s Reasoning Capability

Update: 2024-12-02
Share

Description





This paper introduces cDPO, a method for identifying critical tokens in LLMs that lead to incorrect reasoning, enhancing model alignment through token-level rewards and improving performance on reasoning tasks.




https://arxiv.org/abs//2411.19943




YouTube: https://www.youtube.com/@ArxivPapers




TikTok: https://www.tiktok.com/@arxiv_papers




Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv-papers/id1692476016




Spotify: https://podcasters.spotify.com/pod/show/arxiv-papers





---

Support this podcast: https://podcasters.spotify.com/pod/show/arxiv-papers/support
Comments 
Top Podcasts
The Best New Comedy Podcast Right Now – June 2024The Best News Podcast Right Now – June 2024The Best New Business Podcast Right Now – June 2024The Best New Sports Podcast Right Now – June 2024The Best New True Crime Podcast Right Now – June 2024The Best New Joe Rogan Experience Podcast Right Now – June 20The Best New Dan Bongino Show Podcast Right Now – June 20The Best New Mark Levin Podcast – June 2024
In Channel
[QA] Transformers Struggle to Learn to Search

[QA] Transformers Struggle to Learn to Search

2024-12-0907:41

Transformers Struggle to Learn to Search

Transformers Struggle to Learn to Search

2024-12-0919:58

[QA] Navigation World Models

[QA] Navigation World Models

2024-12-0707:36

Navigation World Models

Navigation World Models

2024-12-0719:47

[QA] Motion Prompting: Controlling Video Generation with Motion Trajectories

[QA] Motion Prompting: Controlling Video Generation with Motion Trajectories

2024-12-0707:21

Motion Prompting: Controlling Video Generation with Motion Trajectories

Motion Prompting: Controlling Video Generation with Motion Trajectories

2024-12-0716:36

[QA] Infinity: Scaling Bitwise AutoRegressive Modeling for High-Resolution Image Synthesis

[QA] Infinity: Scaling Bitwise AutoRegressive Modeling for High-Resolution Image Synthesis

2024-12-0608:01

Infinity: Scaling Bitwise AutoRegressive Modeling for High-Resolution Image Synthesis

Infinity: Scaling Bitwise AutoRegressive Modeling for High-Resolution Image Synthesis

2024-12-0621:32

[QA] NVILA: Efficient Frontier Visual Language Models

[QA] NVILA: Efficient Frontier Visual Language Models

2024-12-0607:38

NVILA: Efficient Frontier Visual Language Models

NVILA: Efficient Frontier Visual Language Models

2024-12-0623:11

[QA] o1-Coder: an o1 Replication for Coding

[QA] o1-Coder: an o1 Replication for Coding

2024-12-0309:05

o1-Coder: an o1 Replication for Coding

o1-Coder: an o1 Replication for Coding

2024-12-0319:21

[QA] Efficient Track Anything

[QA] Efficient Track Anything

2024-12-0307:55

Efficient Track Anything

Efficient Track Anything

2024-12-0319:04

[QA] Reverse Thinking Makes LLMs Stronger Reasoners

[QA] Reverse Thinking Makes LLMs Stronger Reasoners

2024-12-0208:08

Reverse Thinking Makes LLMs Stronger Reasoners

Reverse Thinking Makes LLMs Stronger Reasoners

2024-12-0216:36

[QA] Critical Tokens Matter: Token-Level Contrastive Estimation Enhence LLM’s Reasoning Capability

[QA] Critical Tokens Matter: Token-Level Contrastive Estimation Enhence LLM’s Reasoning Capability

2024-12-0207:29

Critical Tokens Matter: Token-Level Contrastive Estimation Enhence LLM’s Reasoning Capability

Critical Tokens Matter: Token-Level Contrastive Estimation Enhence LLM’s Reasoning Capability

2024-12-0212:00

[QA] JetFormer: an autoregressive generative model of raw images and text

[QA] JetFormer: an autoregressive generative model of raw images and text

2024-12-0207:29

JetFormer: an autoregressive generative model of raw images and text

JetFormer: an autoregressive generative model of raw images and text

2024-12-0223:02

loading
Google Play
Download from Google Play
Castbox
Download from App Store
usUnited States
00:00
00:00
x

0.5x

0.8x

1.0x

1.25x

1.5x

2.0x

3.0x

Sleep Timer

Off

End of Episode

5 Minutes

10 Minutes

15 Minutes

30 Minutes

45 Minutes

60 Minutes

120 Minutes

Critical Tokens Matter: Token-Level Contrastive Estimation Enhence LLM’s Reasoning Capability

Critical Tokens Matter: Token-Level Contrastive Estimation Enhence LLM’s Reasoning Capability

Igor Melnyk