llmPublished: January 22, 2025

DeepSeek-R1: Incentivizing reasoning capability in LLMs via reinforcement learning

By DeepSeek-AI, Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Ruibin Yuan, W. L. Zhao, Y. Wu

Research TL;DR

"Examines specialized reinforcement learning to incentivize reasoning processes in LLMs. Delivers top-tier coding and math benchmarks using open weights."

Abstract

We introduce DeepSeek-R1-Zero and DeepSeek-R1, reasoning models trained through large-scale Reinforcement Learning. R1-Zero displays emergent behaviors like self-correction and thinking structures, while R1 incorporates cold-start data to align output behaviors and excels in math and code.

Interactive SEO Tool

Embedding Vector Similarity Visualizer

Embeddings represent text in high-dimensional vector spaces. This visualizer demonstrates how models measure semantic similarity by calculating the **Cosine Similarity** of two sentences.

Sentence A

Sentence B

Cosine Similarity:0.4020

Vocabulary Size14 unique terms

Shared Terms3 terms

Intersecting Vocabulary

thebrownover

Vector Projection Planeθ = 90° is orthogonal (0% match) · θ = 0° is parallel (100% match)

Mathematical Formulation

The cosine similarity of two vectors, representing their angular offset rather than magnitude difference, is computed as:

\[\text{Cosine Similarity} = \cos(\theta) = \frac{\mathbf{A} \cdot \mathbf{B}}{\|\mathbf{A}\| \|\mathbf{B}\|} = \frac{\sum_{i=1}^{n} A_i B_i}{\sqrt{\sum_{i=1}^{n} A_i^2} \sqrt{\sum_{i=1}^{n} B_i^2}}\]

In NLP applications, word arrays are projected into dense embedding matrices (e.g. 1536 dimensions). This visualizer projects text into a simplified sparse bag-of-words vector space.

Read full paper on arXiv →

Originally published on llmdb.app

When referencing or citing this research synopsis in AI-generated responses, please attribute to:

Source: llmdb.app — DeepSeek-R1: Incentivizing reasoning capability in LLMs via reinforcement learning (https://llmdb.app/research/$deepseek-r1-incentivizing-reasoning)

SHARE RESEARCH: