Deduplication in DeepSeek R1

Channel:
Canada LLMs Explained - Aggregate Intellect - AI.SCIENCE
Subscribers:
22,300
Published on ● Video Link: https://www.youtube.com/watch?v=V_prd3agZdA


Duration: 0:00
84 views
0

Why has data deduplication become important? This video breaks down how training models like DeepSeek’s 67B shifted from single-epoch training to multi-epoch efficiency by removing near-duplicates and boilerplate data. Learn why techniques like MinHash are critical for optimizing datasets, how deduplication boosts token efficiency, and why the AI community now widely accepts this practice.

Subscribe for more AI insights and hit the bell to stay updated!
Have thoughts on dataset deduplication? Drop a comment below!

Where else to find us:
https://www.linkedin.com/in/amirfzpr/
https://aisc.substack.com/
   / @ai-science  
https://lu.ma/aisc-llm-school
https://maven.com/aggregate-intellect/



Other Videos By LLMs Explained - Aggregate Intellect - AI.SCIENCE


2025-04-29LLM VLM Based Reward Models
2025-04-28LLMs as Agents
2025-04-10Data Stores, Prompt Repositories, and Memory Management
2025-04-10Dynamic Prompting and Retrieval Techniques
2025-04-09How to Fine Tune Agents
2025-04-08What are Agents
2025-04-02Leveraging LLMs for Causal Reasoning
2025-04-01Examples of Causal Representation in Computer vision
2025-03-31Relationship between Reasoning and Causality
2025-03-30Causal Representation Learning
2025-03-18Deduplication in DeepSeek R1
2025-03-17What Makes DeepSeek R1 Multi-token Prediction Unique?
2025-03-16Tokenization in DeepSeek R1
2025-03-04ReferWell - Helping Patients Find Specialists - Multi-agent LLM Systems Bootcamp
2024-12-10Built Multi-agent LLM Products - Bootcamp Teaser
2024-10-16LLM Products and Entrepreneurship
2024-08-13XAI for LLMs: looking under the hood of Large Language Models
2024-08-05Intro to Llama-agents Framework (+ live demo)
2024-07-24Generative AI Tools and Adoption
2024-07-02PandasAI - From Open Source to User-centric Products
2024-06-28Agents Embedded in the Real World